ha_tokudb.cc 245 KB
Newer Older
1
#ifdef USE_PRAGMA_IMPLEMENTATION
2
#pragma implementation          // gcc: Class implementation
3 4 5
#endif

#define MYSQL_SERVER 1
6
#include "hatoku_defines.h"
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7
#include "hatoku_cmp.h"
Zardosht Kasheff's avatar
Zardosht Kasheff committed
8 9 10 11 12 13
extern "C" {
#include "stdint.h"
#if defined(_WIN32)
#include "misc.h"
#endif
}
14

15 16 17 18 19 20 21 22
static inline void *thd_data_get(THD *thd, int slot) {
    return thd->ha_data[slot].ha_ptr;
}

static inline void thd_data_set(THD *thd, int slot, void *data) {
    thd->ha_data[slot].ha_ptr = data;
}

23 24 25 26 27 28 29 30 31 32 33 34
#undef PACKAGE
#undef VERSION
#undef HAVE_DTRACE
#undef _DTRACE_VERSION

/* We define DTRACE after mysql_priv.h in case it disabled dtrace in the main server */
#ifdef HAVE_DTRACE
#define _DTRACE_VERSION 1
#else
#endif

#include "ha_tokudb.h"
Zardosht Kasheff's avatar
Zardosht Kasheff committed
35
#include "hatoku_hton.h"
36 37
#include <mysql/plugin.h>

Zardosht Kasheff's avatar
Zardosht Kasheff committed
38 39 40 41
static const char *ha_tokudb_exts[] = {
    ha_tokudb_ext,
    NullS
};
42

Zardosht Kasheff's avatar
Zardosht Kasheff committed
43 44 45
//
// This offset is calculated starting from AFTER the NULL bytes
//
Yoni Fogel's avatar
Yoni Fogel committed
46
static inline uint32_t get_fixed_field_size(KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
47 48
    uint offset = 0;
    for (uint i = 0; i < table_share->fields; i++) {
49 50
        if (kc_info->field_lengths[i] && !bitmap_is_set(&kc_info->key_filters[keynr],i)) {
            offset += kc_info->field_lengths[i];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
51 52 53 54 55 56
        }
    }
    return offset;
}


Yoni Fogel's avatar
Yoni Fogel committed
57
static inline uint32_t get_len_of_offsets(KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
58 59
    uint len = 0;
    for (uint i = 0; i < table_share->fields; i++) {
60 61
        if (kc_info->length_bytes[i] && !bitmap_is_set(&kc_info->key_filters[keynr],i)) {
            len += kc_info->num_offset_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
62 63 64 65 66 67
        }
    }
    return len;
}


68
static int allocate_key_and_col_info ( TABLE_SHARE* table_share, KEY_AND_COL_INFO* kc_info) {
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87
    int error;
    //
    // initialize all of the bitmaps
    //
    for (uint i = 0; i < MAX_KEY + 1; i++) {
        error = bitmap_init(
            &kc_info->key_filters[i],
            NULL,
            table_share->fields,
            false
            );
        if (error) {
            goto exit;
        }
    }
    
    //
    // create the field lengths
    //
Yoni Fogel's avatar
Yoni Fogel committed
88
    kc_info->field_lengths = (uint16_t *)my_malloc(table_share->fields*sizeof(uint16_t), MYF(MY_WME | MY_ZEROFILL));
89
    kc_info->length_bytes= (uchar *)my_malloc(table_share->fields, MYF(MY_WME | MY_ZEROFILL));
Yoni Fogel's avatar
Yoni Fogel committed
90
    kc_info->blob_fields= (uint32_t *)my_malloc(table_share->fields*sizeof(uint32_t), MYF(MY_WME | MY_ZEROFILL));
91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109
    
    if (kc_info->field_lengths == NULL || 
        kc_info->length_bytes == NULL || 
        kc_info->blob_fields == NULL ) {
        error = ENOMEM;
        goto exit;
    }
exit:
    if (error) {
        for (uint i = 0; MAX_KEY + 1; i++) {
            bitmap_free(&kc_info->key_filters[i]);
        }
        my_free(kc_info->field_lengths, MYF(MY_ALLOW_ZERO_PTR));
        my_free(kc_info->length_bytes, MYF(MY_ALLOW_ZERO_PTR));
        my_free(kc_info->blob_fields, MYF(MY_ALLOW_ZERO_PTR));
    }
    return error;
}

110 111 112 113 114
/** @brief
    Simple lock controls. The "share" it creates is a structure we will
    pass to each tokudb handler. Do you have to have one of these? Well, you have
    pieces that are used for locking, and they are needed to function.
*/
115
static TOKUDB_SHARE *get_share(const char *table_name, TABLE_SHARE* table_share) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
116 117
    TOKUDB_SHARE *share = NULL;
    int error = 0;
118
    uint length;
119

120 121 122
    pthread_mutex_lock(&tokudb_mutex);
    length = (uint) strlen(table_name);

123
    if (!(share = (TOKUDB_SHARE *) my_hash_search(&tokudb_open_tables, (uchar *) table_name, length))) {
124 125
        char *tmp_name;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
126 127 128 129
        //
        // create share and fill it with all zeroes
        // hence, all pointers are initialized to NULL
        //
130 131 132 133 134
        if (!(share = (TOKUDB_SHARE *) 
            my_multi_malloc(MYF(MY_WME | MY_ZEROFILL), 
                            &share, sizeof(*share),
                            &tmp_name, length + 1, 
                            NullS))) {
135 136 137 138 139 140 141
            pthread_mutex_unlock(&tokudb_mutex);
            return NULL;
        }
        share->use_count = 0;
        share->table_name_length = length;
        share->table_name = tmp_name;
        strmov(share->table_name, table_name);
142

143 144
        error = allocate_key_and_col_info(table_share, &share->kc_info);
        if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
145 146 147
            goto exit;
        }

148
        memset((void *) share->key_file, 0, sizeof(share->key_file));
149

Zardosht Kasheff's avatar
Zardosht Kasheff committed
150 151 152 153
        error = my_hash_insert(&tokudb_open_tables, (uchar *) share);
        if (error) {
            goto exit;
        }
154 155
        thr_lock_init(&share->lock);
        pthread_mutex_init(&share->mutex, MY_MUTEX_INIT_FAST);
156
        my_rwlock_init(&share->num_DBs_lock, 0);
157 158
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
159 160 161 162 163 164
exit:
    if (error) {
        pthread_mutex_destroy(&share->mutex);
        my_free((uchar *) share, MYF(0));
        share = NULL;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
165
    pthread_mutex_unlock(&tokudb_mutex);
166
    return share;
167 168
}

169

170
static void free_key_and_col_info (KEY_AND_COL_INFO* kc_info) {
171 172 173 174 175 176
    for (uint i = 0; i < MAX_KEY+1; i++) {
        bitmap_free(&kc_info->key_filters[i]);
    }
    
    for (uint i = 0; i < MAX_KEY+1; i++) {
        my_free(kc_info->cp_info[i], MYF(MY_ALLOW_ZERO_PTR));
177
        kc_info->cp_info[i] = NULL; // 3144
178 179 180 181 182 183 184
    }
    
    my_free(kc_info->field_lengths, MYF(MY_ALLOW_ZERO_PTR));
    my_free(kc_info->length_bytes, MYF(MY_ALLOW_ZERO_PTR));
    my_free(kc_info->blob_fields, MYF(MY_ALLOW_ZERO_PTR));
}

185
static int free_share(TOKUDB_SHARE * share, bool mutex_is_locked) {
186 187 188 189 190 191 192 193 194
    int error, result = 0;

    pthread_mutex_lock(&tokudb_mutex);

    if (mutex_is_locked)
        pthread_mutex_unlock(&share->mutex);
    if (!--share->use_count) {
        DBUG_PRINT("info", ("share->use_count %u", share->use_count));

Zardosht Kasheff's avatar
Zardosht Kasheff committed
195 196 197 198 199 200 201 202 203 204 205
        //
        // number of open DB's may not be equal to number of keys we have because add_index
        // may have added some. So, we loop through entire array and close any non-NULL value
        // It is imperative that we reset a DB to NULL once we are done with it.
        //
        for (uint i = 0; i < sizeof(share->key_file)/sizeof(share->key_file[0]); i++) {
            if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
                TOKUDB_TRACE("dbclose:%p\n", share->key_file[i]);
            }
            if (share->key_file[i]) { 
                error = share->key_file[i]->close(share->key_file[i], 0);
206
                assert(error == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
207 208 209 210 211
                if (error) {
                    result = error;
                }
                share->key_file[i] = NULL;
            }
212
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
213

214
        free_key_and_col_info(&share->kc_info);
215

Zardosht Kasheff's avatar
Zardosht Kasheff committed
216
        if (share->status_block && (error = share->status_block->close(share->status_block, 0))) {
217
            assert(error == 0);
218
            result = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
219 220
        }
        
221

222
        my_hash_delete(&tokudb_open_tables, (uchar *) share);
223 224
        thr_lock_delete(&share->lock);
        pthread_mutex_destroy(&share->mutex);
225
        rwlock_destroy(&share->num_DBs_lock);
226 227 228 229 230
        my_free((uchar *) share, MYF(0));
    }
    pthread_mutex_unlock(&tokudb_mutex);

    return result;
231 232
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
233

Zardosht Kasheff's avatar
Zardosht Kasheff committed
234
#define HANDLE_INVALID_CURSOR() \
Zardosht Kasheff's avatar
Zardosht Kasheff committed
235 236 237 238 239
    if (cursor == NULL) { \
        error = last_cursor_error; \
        goto cleanup; \
    }

240 241 242 243
const char *ha_tokudb::table_type() const {
    extern const char * const tokudb_hton_name;
    return tokudb_hton_name;
} 
244

245 246 247
const char *ha_tokudb::index_type(uint inx) {
    return "BTREE";
}
248

Zardosht Kasheff's avatar
Zardosht Kasheff committed
249 250 251
/* 
 *  returns NULL terminated file extension string
 */
252
const char **ha_tokudb::bas_ext() const {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
253 254
    TOKUDB_DBUG_ENTER("ha_tokudb::bas_ext");
    DBUG_RETURN(ha_tokudb_exts);
255 256
}

257 258 259 260 261 262 263 264
static inline bool is_insert_ignore (THD* thd) {
    //
    // from http://lists.mysql.com/internals/37735
    //
    return thd->lex->ignore && thd->lex->duplicates == DUP_ERROR;
}

static inline bool is_replace_into(THD* thd) {
265
    return thd->lex->duplicates == DUP_REPLACE;
266 267
}

268
static inline bool do_ignore_flag_optimization(THD* thd, TABLE* table, bool opt_eligible) {
269 270
    uint pk_insert_mode = get_pk_insert_mode(thd);
    return ( 
271
        opt_eligible && 
272
        (is_replace_into(thd) || is_insert_ignore(thd)) && 
273 274 275 276 277
        ((!table->triggers && pk_insert_mode < 2) || pk_insert_mode == 0)
        );
}

ulonglong ha_tokudb::table_flags() const {
278
    return (table && do_ignore_flag_optimization(ha_thd(), table, share->replace_into_fast) ? 
279
        int_table_flags | HA_BINLOG_STMT_CAPABLE : 
Rich Prohaska's avatar
Rich Prohaska committed
280
#if defined(HA_GENERAL_ONLINE)
281
        int_table_flags | HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE | HA_ONLINE_ALTER);
282 283 284
#else
        int_table_flags | HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE);
#endif
285 286
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
287 288 289 290
//
// Returns a bit mask of capabilities of the key or its part specified by 
// the arguments. The capabilities are defined in sql/handler.h.
//
291
ulong ha_tokudb::index_flags(uint idx, uint part, bool all_parts) const {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
292
    TOKUDB_DBUG_ENTER("ha_tokudb::index_flags");
293
    assert(table_share);
294
    ulong flags = (HA_READ_NEXT | HA_READ_PREV | HA_READ_ORDER | HA_KEYREAD_ONLY | HA_READ_RANGE);
295 296 297
    if (table_share->key_info[idx].flags & HA_CLUSTERING) {
        flags |= HA_CLUSTERED_INDEX;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
298
    DBUG_RETURN(flags);
299 300
}

Yoni Fogel's avatar
Yoni Fogel committed
301

Zardosht Kasheff's avatar
Zardosht Kasheff committed
302 303 304 305 306 307 308 309 310 311
//
// struct that will be used as a context for smart DBT callbacks
// contains parameters needed to complete the smart DBT cursor call
//
typedef struct smart_dbt_info {
    ha_tokudb* ha; //instance to ha_tokudb needed for reading the row
    uchar* buf; // output buffer where row will be written
    uint keynr; // index into share->key_file that represents DB we are currently operating on
} *SMART_DBT_INFO;

312 313 314 315 316 317 318
typedef struct smart_dbt_bf_info {
    ha_tokudb* ha;
    bool need_val;
    int direction;
    THD* thd;
} *SMART_DBT_BF_INFO;

319 320 321 322 323 324
typedef struct index_read_info {
    struct smart_dbt_info smart_dbt_info;
    int cmp;
    DBT* orig_key;
} *INDEX_READ_INFO;

325

326
static int ai_poll_fun(void *extra, float progress) {
327 328 329
    LOADER_CONTEXT context = (LOADER_CONTEXT)extra;
    if (context->thd->killed) {
        sprintf(context->write_status_msg, "The process has been killed, aborting add index.");
330
        return ER_ABORTING_CONNECTION;
331
    }
332 333
    float percentage = progress * 100;
    sprintf(context->write_status_msg, "Adding of indexes about %.1f%% done", percentage);
334
    thd_proc_info(context->thd, context->write_status_msg);
335 336 337
#ifdef HA_TOKUDB_HAS_THD_PROGRESS
    thd_progress_report(context->thd, (unsigned long long) percentage, 100);
#endif
338 339 340
    return 0;
}

341
static int loader_poll_fun(void *extra, float progress) {
342 343 344
    LOADER_CONTEXT context = (LOADER_CONTEXT)extra;
    if (context->thd->killed) {
        sprintf(context->write_status_msg, "The process has been killed, aborting bulk load.");
345
        return ER_ABORTING_CONNECTION;
346
    }
347 348
    float percentage = progress * 100;
    sprintf(context->write_status_msg, "Loading of data about %.1f%% done", percentage);
349
    thd_proc_info(context->thd, context->write_status_msg);
350 351 352
#ifdef HA_TOKUDB_HAS_THD_PROGRESS
    thd_progress_report(context->thd, (unsigned long long) percentage, 100);
#endif
353 354 355
    return 0;
}

356
static void loader_ai_err_fun(DB *db, int i, int err, DBT *key, DBT *val, void *error_extra) {
357 358 359 360 361
    LOADER_CONTEXT context = (LOADER_CONTEXT)error_extra;
    assert(context->ha);
    context->ha->set_loader_error(err);
}

362
static void loader_dup_fun(DB *db, int i, int err, DBT *key, DBT *val, void *error_extra) {
363 364 365 366 367 368
    LOADER_CONTEXT context = (LOADER_CONTEXT)error_extra;
    assert(context->ha);
    context->ha->set_loader_error(err);
    if (err == DB_KEYEXIST) {
        context->ha->set_dup_value_for_pk(key);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
369 370
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
371 372 373 374 375 376 377
//
// smart DBT callback function for optimize
// in optimize, we want to flatten DB by doing
// a full table scan. Therefore, we don't
// want to actually do anything with the data, hence
// callback does nothing
//
378
static int smart_dbt_do_nothing (DBT const *key, DBT  const *row, void *context) {
Barry Perlman's avatar
Barry Perlman committed
379
  return 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
380 381
}

382 383 384
static int smart_dbt_metacallback (DBT const *key, DBT  const *row, void *context) {
    DBT* val = (DBT *)context;
    val->data = my_malloc(row->size, MYF(MY_WME|MY_ZEROFILL));
385 386 387 388
    if (val->data == NULL) return ENOMEM;
    memcpy(val->data, row->data, row->size);
    val->size = row->size;
    return 0;
389 390
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
391

392 393 394
static int
smart_dbt_callback_rowread_ptquery (DBT const *key, DBT  const *row, void *context) {
    SMART_DBT_INFO info = (SMART_DBT_INFO)context;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
395
    info->ha->extract_hidden_primary_key(info->keynr, key);
396
    return info->ha->read_row_callback(info->buf,info->keynr,row,key);
397 398
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
399 400 401
//
// Smart DBT callback function in case where we have a covering index
//
Barry Perlman's avatar
Barry Perlman committed
402 403
static int
smart_dbt_callback_keyread(DBT const *key, DBT  const *row, void *context) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
404
    SMART_DBT_INFO info = (SMART_DBT_INFO)context;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
405 406
    info->ha->extract_hidden_primary_key(info->keynr, key);
    info->ha->read_key_only(info->buf,info->keynr,key);
Barry Perlman's avatar
Barry Perlman committed
407
    return 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
408 409 410 411 412
}

//
// Smart DBT callback function in case where we do NOT have a covering index
//
Barry Perlman's avatar
Barry Perlman committed
413 414
static int
smart_dbt_callback_rowread(DBT const *key, DBT  const *row, void *context) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
415
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
416
    SMART_DBT_INFO info = (SMART_DBT_INFO)context;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
417
    info->ha->extract_hidden_primary_key(info->keynr, key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
418 419
    error = info->ha->read_primary_key(info->buf,info->keynr,row,key);
    return error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
420 421
}

422 423 424 425 426 427 428 429 430 431 432 433 434
//
// Smart DBT callback function in case where we have a covering index
//
static int
smart_dbt_callback_ir_keyread(DBT const *key, DBT  const *row, void *context) {
    INDEX_READ_INFO ir_info = (INDEX_READ_INFO)context;
    ir_info->cmp = ir_info->smart_dbt_info.ha->prefix_cmp_dbts(ir_info->smart_dbt_info.keynr, ir_info->orig_key, key);
    if (ir_info->cmp) {
        return 0;
    }
    return smart_dbt_callback_keyread(key, row, &ir_info->smart_dbt_info);
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
435 436 437 438 439 440 441 442
static int
smart_dbt_callback_lookup(DBT const *key, DBT  const *row, void *context) {
    INDEX_READ_INFO ir_info = (INDEX_READ_INFO)context;
    ir_info->cmp = ir_info->smart_dbt_info.ha->prefix_cmp_dbts(ir_info->smart_dbt_info.keynr, ir_info->orig_key, key);
    return 0;
}


443 444 445 446 447 448 449 450 451 452 453 454 455
//
// Smart DBT callback function in case where we do NOT have a covering index
//
static int
smart_dbt_callback_ir_rowread(DBT const *key, DBT  const *row, void *context) {
    INDEX_READ_INFO ir_info = (INDEX_READ_INFO)context;
    ir_info->cmp = ir_info->smart_dbt_info.ha->prefix_cmp_dbts(ir_info->smart_dbt_info.keynr, ir_info->orig_key, key);
    if (ir_info->cmp) {
        return 0;
    }
    return smart_dbt_callback_rowread(key, row, &ir_info->smart_dbt_info);
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
456 457 458 459 460
//
// macro for Smart DBT callback function, 
// so we do not need to put this long line of code in multiple places
//
#define SMART_DBT_CALLBACK ( this->key_read ? smart_dbt_callback_keyread : smart_dbt_callback_rowread ) 
461
#define SMART_DBT_IR_CALLBACK ( this->key_read ? smart_dbt_callback_ir_keyread : smart_dbt_callback_ir_rowread ) 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
462 463


Zardosht Kasheff's avatar
Zardosht Kasheff committed
464 465 466 467
//
// macro that modifies read flag for cursor operations depending on whether
// we have preacquired lock or not
//
468
#define SET_PRELOCK_FLAG(flg) ((flg) | (range_lock_grabbed ? (use_write_locks ? DB_PRELOCKED_WRITE : DB_PRELOCKED) : 0))
Zardosht Kasheff's avatar
Zardosht Kasheff committed
469 470 471 472 473 474 475 476 477 478 479 480 481

//
// This method retrieves the value of the auto increment column of a record in MySQL format
// This was basically taken from MyISAM
// Parameters:
//              type - the type of the auto increment column (e.g. int, float, double...)
//              offset - offset into the record where the auto increment column is stored
//      [in]    record - MySQL row whose auto increment value we want to extract
// Returns:
//      The value of the auto increment column in record
//
ulonglong retrieve_auto_increment(uint16 type, uint32 offset,const uchar *record)
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
482
    const uchar *key;     /* Key */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
483 484
    ulonglong   unsigned_autoinc = 0;  /* Unsigned auto-increment */
    longlong      signed_autoinc = 0;  /* Signed auto-increment */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528
    enum { unsigned_type, signed_type } autoinc_type;
    float float_tmp;   /* Temporary variable */
    double double_tmp; /* Temporary variable */

    key = ((uchar *) record) + offset;

    /* Set default autoincrement type */
    autoinc_type = unsigned_type;

    switch (type) {
    case HA_KEYTYPE_INT8:
        signed_autoinc   = (longlong) *(char*)key;
        autoinc_type     = signed_type;
        break;

    case HA_KEYTYPE_BINARY:
        unsigned_autoinc = (ulonglong) *(uchar*) key;
        break;

    case HA_KEYTYPE_SHORT_INT:
        signed_autoinc   = (longlong) sint2korr(key);
        autoinc_type     = signed_type;
        break;

    case HA_KEYTYPE_USHORT_INT:
        unsigned_autoinc = (ulonglong) uint2korr(key);
        break;

    case HA_KEYTYPE_LONG_INT:
        signed_autoinc   = (longlong) sint4korr(key);
        autoinc_type     = signed_type;
        break;

    case HA_KEYTYPE_ULONG_INT:
        unsigned_autoinc = (ulonglong) uint4korr(key);
        break;

    case HA_KEYTYPE_INT24:
        signed_autoinc   = (longlong) sint3korr(key);
        autoinc_type     = signed_type;
        break;

    case HA_KEYTYPE_UINT24:
        unsigned_autoinc = (ulonglong) uint3korr(key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
529
    break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564

    case HA_KEYTYPE_LONGLONG:
        signed_autoinc   = sint8korr(key);
        autoinc_type     = signed_type;
        break;

    case HA_KEYTYPE_ULONGLONG:
        unsigned_autoinc = uint8korr(key);
        break;

    /* The remaining two cases should not be used but are included for 
       compatibility */
    case HA_KEYTYPE_FLOAT:                      
        float4get(float_tmp, key);  /* Note: float4get is a macro */
        signed_autoinc   = (longlong) float_tmp;
        autoinc_type     = signed_type;
        break;

    case HA_KEYTYPE_DOUBLE:
        float8get(double_tmp, key); /* Note: float8get is a macro */
        signed_autoinc   = (longlong) double_tmp;
        autoinc_type     = signed_type;
        break;

    default:
        DBUG_ASSERT(0);
        unsigned_autoinc = 0;
    }

    if (signed_autoinc < 0) {
        signed_autoinc = 0;
    }

    return autoinc_type == unsigned_type ?  
           unsigned_autoinc : (ulonglong) signed_autoinc;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
565 566
}

567
static inline bool
Zardosht Kasheff's avatar
Zardosht Kasheff committed
568
is_null_field( TABLE* table, Field* field, const uchar* record) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
569
    uint null_offset;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
570
    bool ret_val;
571
    if (!field->real_maybe_null()) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
572
        ret_val = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
573
        goto exitpt;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
574 575
    }
    null_offset = get_null_offset(table,field);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
576 577 578
    ret_val = (record[null_offset] & field->null_bit) ? true: false;

exitpt:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
579 580 581
    return ret_val;
}

582
static inline ulong field_offset(Field* field, TABLE* table) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
583
    return((ulong) (field->ptr - table->record[0]));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
584 585
}

586
static inline HA_TOKU_ISO_LEVEL tx_to_toku_iso(ulong tx_isolation) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
587 588 589
    if (tx_isolation == ISO_READ_UNCOMMITTED) {
        return hatoku_iso_read_uncommitted;
    }
590 591 592
    else if (tx_isolation == ISO_READ_COMMITTED) {
        return hatoku_iso_read_committed;
    }
593 594 595
    else if (tx_isolation == ISO_REPEATABLE_READ) {
        return hatoku_iso_repeatable_read;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
596 597 598 599 600
    else {
        return hatoku_iso_serializable;
    }
}

Yoni Fogel's avatar
Yoni Fogel committed
601
static inline uint32_t toku_iso_to_txn_flag (HA_TOKU_ISO_LEVEL lvl) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
602
    if (lvl == hatoku_iso_read_uncommitted) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
603
        return DB_READ_UNCOMMITTED;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
604
    }
605 606 607
    else if (lvl == hatoku_iso_read_committed) {
        return DB_READ_COMMITTED;
    }
608 609 610
    else if (lvl == hatoku_iso_repeatable_read) {
        return DB_TXN_SNAPSHOT;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
611 612 613 614 615
    else {
        return 0;
    }
}

616
static int filter_key_part_compare (const void* left, const void* right) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685
    FILTER_KEY_PART_INFO* left_part= (FILTER_KEY_PART_INFO *)left;
    FILTER_KEY_PART_INFO* right_part = (FILTER_KEY_PART_INFO *)right;
    return left_part->offset - right_part->offset;
}

//
// Be very careful with parameters passed to this function. Who knows
// if key, table have proper info set. I had to verify by checking
// in the debugger.
//
void set_key_filter(MY_BITMAP* key_filter, KEY* key, TABLE* table, bool get_offset_from_keypart) {
    FILTER_KEY_PART_INFO parts[MAX_REF_PARTS];
    uint curr_skip_index = 0;

    for (uint i = 0; i < key->key_parts; i++) {
        //
        // horrendous hack due to bugs in mysql, basically
        // we cannot always reliably get the offset from the same source
        //
        parts[i].offset = get_offset_from_keypart ? key->key_part[i].offset : field_offset(key->key_part[i].field, table);
        parts[i].part_index = i;
    }
    qsort(
        parts, // start of array
        key->key_parts, //num elements
        sizeof(*parts), //size of each element
        filter_key_part_compare
        );

    for (uint i = 0; i < table->s->fields; i++) {
        Field* field = table->field[i];
        uint curr_field_offset = field_offset(field, table);
        if (curr_skip_index < key->key_parts) {
            uint curr_skip_offset = 0;
            curr_skip_offset = parts[curr_skip_index].offset;
            if (curr_skip_offset == curr_field_offset) {
                //
                // we have hit a field that is a portion of the primary key
                //
                uint curr_key_index = parts[curr_skip_index].part_index;
                curr_skip_index++;
                //
                // only choose to continue over the key if the key's length matches the field's length
                // otherwise, we may have a situation where the column is a varchar(10), the
                // key is only the first 3 characters, and we end up losing the last 7 bytes of the
                // column
                //
                TOKU_TYPE toku_type;
                toku_type = mysql_to_toku_type(field);
                switch(toku_type) {
                case(toku_type_blob):
                    break;
                case(toku_type_varbinary):
                case(toku_type_varstring):
                case(toku_type_fixbinary):
                case(toku_type_fixstring):
                    if (key->key_part[curr_key_index].length == field->field_length) {
                        bitmap_set_bit(key_filter,i);
                    }
                    break;
                default:
                    bitmap_set_bit(key_filter,i);
                    break;
                }
            }
        }
    }
}

686
static inline uchar* pack_fixed_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
687 688
    uchar* to_tokudb,
    const uchar* from_mysql,
Yoni Fogel's avatar
Yoni Fogel committed
689
    uint32_t num_bytes
Zardosht Kasheff's avatar
Zardosht Kasheff committed
690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714
    )
{
    switch (num_bytes) {
    case (1):
        memcpy(to_tokudb, from_mysql, 1);
        break;
    case (2):
        memcpy(to_tokudb, from_mysql, 2);
        break;
    case (3):
        memcpy(to_tokudb, from_mysql, 3);
        break;
    case (4):
        memcpy(to_tokudb, from_mysql, 4);
        break;
    case (8):
        memcpy(to_tokudb, from_mysql, 8);
        break;
    default:
        memcpy(to_tokudb, from_mysql, num_bytes);
        break;
    }
    return to_tokudb+num_bytes;
}

715
static inline const uchar* unpack_fixed_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
716 717
    uchar* to_mysql,
    const uchar* from_tokudb,
Yoni Fogel's avatar
Yoni Fogel committed
718
    uint32_t num_bytes
Zardosht Kasheff's avatar
Zardosht Kasheff committed
719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743
    )
{
    switch (num_bytes) {
    case (1):
        memcpy(to_mysql, from_tokudb, 1);
        break;
    case (2):
        memcpy(to_mysql, from_tokudb, 2);
        break;
    case (3):
        memcpy(to_mysql, from_tokudb, 3);
        break;
    case (4):
        memcpy(to_mysql, from_tokudb, 4);
        break;
    case (8):
        memcpy(to_mysql, from_tokudb, 8);
        break;
    default:
        memcpy(to_mysql, from_tokudb, num_bytes);
        break;
    }
    return from_tokudb+num_bytes;
}

744
static inline uchar* write_var_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
745
    uchar* to_tokudb_offset_ptr, //location where offset data is going to be written
746 747 748
    uchar* to_tokudb_data, // location where data is going to be written
    uchar* to_tokudb_offset_start, //location where offset starts, IS THIS A BAD NAME????
    const uchar * data, // the data to write
Yoni Fogel's avatar
Yoni Fogel committed
749 750
    uint32_t data_length, // length of data to write
    uint32_t offset_bytes // number of offset bytes
Zardosht Kasheff's avatar
Zardosht Kasheff committed
751 752
    )
{
753
    memcpy(to_tokudb_data, data, data_length);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
754 755 756
    //
    // for offset, we pack the offset where the data ENDS!
    //
Yoni Fogel's avatar
Yoni Fogel committed
757
    uint32_t offset = to_tokudb_data + data_length - to_tokudb_offset_start;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
758 759 760 761 762 763 764 765 766 767 768 769 770 771
    switch(offset_bytes) {
    case (1):
        to_tokudb_offset_ptr[0] = (uchar)offset;
        break;
    case (2):
        int2store(to_tokudb_offset_ptr,offset);
        break;
    default:
        assert(false);
        break;
    }
    return to_tokudb_data + data_length;
}

Yoni Fogel's avatar
Yoni Fogel committed
772
static inline uint32_t get_var_data_length(
773
    const uchar * from_mysql, 
Yoni Fogel's avatar
Yoni Fogel committed
774
    uint32_t mysql_length_bytes 
775 776
    ) 
{
Yoni Fogel's avatar
Yoni Fogel committed
777
    uint32_t data_length;
778 779 780 781 782 783 784 785 786 787 788 789 790 791
    switch(mysql_length_bytes) {
    case(1):
        data_length = from_mysql[0];
        break;
    case(2):
        data_length = uint2korr(from_mysql);
        break;
    default:
        assert(false);
        break;
    }
    return data_length;
}

792
static inline uchar* pack_var_field(
793 794 795 796
    uchar* to_tokudb_offset_ptr, //location where offset data is going to be written
    uchar* to_tokudb_data, // pointer to where tokudb data should be written
    uchar* to_tokudb_offset_start, //location where data starts, IS THIS A BAD NAME????
    const uchar * from_mysql, // mysql data
Yoni Fogel's avatar
Yoni Fogel committed
797 798
    uint32_t mysql_length_bytes, //number of bytes used to store length in from_mysql
    uint32_t offset_bytes //number of offset_bytes used in tokudb row
799 800 801 802 803 804 805 806 807 808 809 810 811
    )
{
    uint data_length = get_var_data_length(from_mysql, mysql_length_bytes);    
    return write_var_field(
        to_tokudb_offset_ptr,
        to_tokudb_data,
        to_tokudb_offset_start,
        from_mysql + mysql_length_bytes,
        data_length,
        offset_bytes
        );
}

812
static inline void unpack_var_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
813 814
    uchar* to_mysql,
    const uchar* from_tokudb_data,
Yoni Fogel's avatar
Yoni Fogel committed
815 816
    uint32_t from_tokudb_data_len,
    uint32_t mysql_length_bytes
Zardosht Kasheff's avatar
Zardosht Kasheff committed
817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838
    )
{
    //
    // store the length
    //
    switch (mysql_length_bytes) {
    case(1):
        to_mysql[0] = (uchar)from_tokudb_data_len;
        break;
    case(2):
        int2store(to_mysql, from_tokudb_data_len);
        break;
    default:
        assert(false);
        break;
    }
    //
    // store the data
    //
    memcpy(to_mysql+mysql_length_bytes, from_tokudb_data, from_tokudb_data_len);
}

839
static uchar* pack_toku_field_blob(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
840 841 842 843 844
    uchar* to_tokudb,
    const uchar* from_mysql,
    Field* field
    )
{
Yoni Fogel's avatar
Yoni Fogel committed
845 846
    uint32_t len_bytes = field->row_pack_length();
    uint32_t length = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
847 848 849 850 851
    uchar* data_ptr = NULL;
    memcpy(to_tokudb, from_mysql, len_bytes);

    switch (len_bytes) {
    case (1):
Yoni Fogel's avatar
Yoni Fogel committed
852
        length = (uint32_t)(*from_mysql);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873
        break;
    case (2):
        length = uint2korr(from_mysql);
        break;
    case (3):
        length = uint3korr(from_mysql);
        break;
    case (4):
        length = uint4korr(from_mysql);
        break;
    default:
        assert(false);
    }

    if (length > 0) {
        memcpy_fixed((uchar *)(&data_ptr), from_mysql + len_bytes, sizeof(uchar*));
        memcpy(to_tokudb + len_bytes, data_ptr, length);
    }
    return (to_tokudb + len_bytes + length);
}

874

875
static int add_table_to_metadata(const char *name, TABLE* table, DB_TXN* txn) {
876 877 878 879
    int error = 0;
    DBT key;
    DBT val;
    uchar hidden_primary_key = (table->s->primary_key >= MAX_KEY);
880
    assert(txn);
881
    
882 883
    memset((void *)&key, 0, sizeof(key));
    memset((void *)&val, 0, sizeof(val));
884 885 886 887 888 889 890 891 892
    key.data = (void *)name;
    key.size = strlen(name) + 1;
    val.data = &hidden_primary_key;
    val.size = sizeof(hidden_primary_key);
    error = metadata_db->put(
        metadata_db,
        txn,
        &key,
        &val,
893
        0
894 895 896 897
        );
    return error;
}

898
static int drop_table_from_metadata(const char *name, DB_TXN* txn) {
899 900 901
    int error = 0;
    DBT key;
    DBT data;
902
    assert(txn);
903 904
    memset((void *)&key, 0, sizeof(key));
    memset((void *)&data, 0, sizeof(data));
905 906 907 908 909 910 911 912 913 914 915
    key.data = (void *)name;
    key.size = strlen(name) + 1;
    error = metadata_db->del(
        metadata_db, 
        txn, 
        &key , 
        DB_DELETE_ANY
        );
    return error;
}

916
static int rename_table_in_metadata(const char *from, const char *to, DB_TXN* txn) {
917 918
    int error = 0;
    DBT from_key;
919 920
    DBT to_key;
    DBT val;
921
    assert(txn);
922
    
923 924 925
    memset((void *)&from_key, 0, sizeof(from_key));
    memset((void *)&to_key, 0, sizeof(to_key));
    memset((void *)&val, 0, sizeof(val));
926 927 928 929 930 931 932 933 934 935 936 937 938 939 940
    from_key.data = (void *)from;
    from_key.size = strlen(from) + 1;
    to_key.data = (void *)to;
    to_key.size = strlen(to) + 1;
    
    error = metadata_db->getf_set(
        metadata_db, 
        txn, 
        0, 
        &from_key, 
        smart_dbt_metacallback, 
        &val
        );

    if (error) {
941
        goto cleanup;
942 943 944 945 946 947 948
    }

    error = metadata_db->put(
        metadata_db,
        txn,
        &to_key,
        &val,
949
        0
950
        );
951 952 953
    if (error) {
        goto cleanup;
    }
954

955
    error = metadata_db->del(
956 957 958 959 960
        metadata_db, 
        txn, 
        &from_key, 
        DB_DELETE_ANY
        );
961 962 963
    if (error) {
        goto cleanup;
    }
964 965 966 967 968 969 970 971 972 973

    error = 0;

cleanup:
    my_free(val.data, MYF(MY_ALLOW_ZERO_PTR));

    return error;
}


974
static int check_table_in_metadata(const char *name, bool* table_found, DB_TXN* txn) {
975 976 977
    int error = 0;
    DBT key;
    pthread_mutex_lock(&tokudb_meta_mutex);
978
    memset((void *)&key, 0, sizeof(key));
979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002
    key.data = (void *)name;
    key.size = strlen(name) + 1;
    
    error = metadata_db->getf_set(
        metadata_db, 
        txn, 
        0, 
        &key, 
        smart_dbt_do_nothing, 
        NULL
        );

    if (error == 0) {
        *table_found = true;
    }
    else if (error == DB_NOTFOUND){
        *table_found = false;
        error = 0;
    }

    pthread_mutex_unlock(&tokudb_meta_mutex);
    return error;
}

1003
static int create_tokudb_trx_data_instance(tokudb_trx_data** out_trx) {
1004 1005 1006 1007 1008 1009 1010 1011 1012
    int error;
    tokudb_trx_data* trx = NULL;
    trx = (tokudb_trx_data *) my_malloc(sizeof(*trx), MYF(MY_ZEROFILL));
    if (!trx) {
        error = ENOMEM;
        goto cleanup;
    }

    *out_trx = trx;
1013
    error = 0;
1014 1015 1016 1017
cleanup:
    return error;
}

1018

1019
static inline int tokudb_generate_row(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1020 1021 1022 1023 1024
    DB *dest_db, 
    DB *src_db,
    DBT *dest_key, 
    DBT *dest_val,
    const DBT *src_key, 
1025
    const DBT *src_val
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1026 1027
    ) 
{
1028 1029
    int error;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1030 1031
    DB* curr_db = dest_db;
    uchar* row_desc = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
1032
    uint32_t desc_size;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1033
    uchar* buff = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
1034
    uint32_t max_key_len = 0;
1035
    
1036
    row_desc = (uchar *)curr_db->descriptor->dbt.data;
Yoni Fogel's avatar
Yoni Fogel committed
1037 1038
    row_desc += (*(uint32_t *)row_desc);
    desc_size = (*(uint32_t *)row_desc) - 4;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1039 1040 1041 1042 1043 1044
    row_desc += 4;
    
    if (is_key_pk(row_desc, desc_size)) {
        if (dest_key->flags == DB_DBT_REALLOC && dest_key->data != NULL) {
            free(dest_key->data);
        }
1045
        if (dest_val != NULL) {
1046 1047 1048
            if (dest_val->flags == DB_DBT_REALLOC && dest_val->data != NULL) {
                free(dest_val->data);
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1049 1050 1051
        }
        dest_key->data = src_key->data;
        dest_key->size = src_key->size;
1052
        dest_key->flags = 0;
1053
        if (dest_val != NULL) {
1054 1055 1056 1057
            dest_val->data = src_val->data;
            dest_val->size = src_val->size;
            dest_val->flags = 0;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1058 1059 1060
        error = 0;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1061 1062 1063 1064 1065 1066 1067 1068 1069
    // at this point, we need to create the key/val and set it
    // in the DBTs
    if (dest_key->flags == 0) {
        dest_key->ulen = 0;
        dest_key->size = 0;
        dest_key->data = NULL;
        dest_key->flags = DB_DBT_REALLOC;
    }
    if (dest_key->flags == DB_DBT_REALLOC) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1070 1071
        max_key_len = max_key_size_from_desc(row_desc, desc_size);
        max_key_len += src_key->size;
1072
        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1073 1074 1075 1076 1077 1078 1079
        if (max_key_len > dest_key->ulen) {
            void* old_ptr = dest_key->data;
            void* new_ptr = NULL;
            new_ptr = realloc(old_ptr, max_key_len);
            assert(new_ptr);
            dest_key->data = new_ptr;
            dest_key->ulen = max_key_len;
1080
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105

        buff = (uchar *)dest_key->data;
        assert(buff != NULL && max_key_len > 0);
    }
    else {
        assert(false);
    }

    dest_key->size = pack_key_from_desc(
        buff,
        row_desc,
        desc_size,
        src_key,
        src_val
        );
    assert(dest_key->ulen >= dest_key->size);
    if (tokudb_debug & TOKUDB_DEBUG_CHECK_KEY && !max_key_len) {
        max_key_len = max_key_size_from_desc(row_desc, desc_size);
        max_key_len += src_key->size;
    }
    if (max_key_len) {
        assert(max_key_len >= dest_key->size);
    }

    row_desc += desc_size;
Yoni Fogel's avatar
Yoni Fogel committed
1106
    desc_size = (*(uint32_t *)row_desc) - 4;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1107
    row_desc += 4;
1108
    if (dest_val != NULL) {
1109
        if (!is_key_clustering(row_desc, desc_size) || src_val->size == 0) {
1110
            dest_val->size = 0;
1111 1112
        }
        else {
1113
            uchar* buff = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1114 1115 1116 1117 1118
            if (dest_val->flags == 0) {
                dest_val->ulen = 0;
                dest_val->size = 0;
                dest_val->data = NULL;
                dest_val->flags = DB_DBT_REALLOC;
1119
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1120
            if (dest_val->flags == DB_DBT_REALLOC){
1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
                if (dest_val->ulen < src_val->size) {
                    void* old_ptr = dest_val->data;
                    void* new_ptr = NULL;
                    new_ptr = realloc(old_ptr, src_val->size);
                    assert(new_ptr);
                    dest_val->data = new_ptr;
                    dest_val->ulen = src_val->size;
                }
                buff = (uchar *)dest_val->data;
                assert(buff != NULL);
            }
            else {
                assert(false);
            }
            dest_val->size = pack_clustering_val_from_desc(
                buff,
                row_desc,
                desc_size,
                src_val
                );
            assert(dest_val->ulen >= dest_val->size);
1142 1143 1144
        }
    }
    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1145
cleanup:
1146 1147 1148
    return error;
}

1149
static int generate_row_for_del(
1150 1151 1152 1153
    DB *dest_db, 
    DB *src_db,
    DBT *dest_key,
    const DBT *src_key, 
1154 1155
    const DBT *src_val
    )
1156 1157 1158 1159 1160 1161 1162
{
    return tokudb_generate_row(
        dest_db,
        src_db,
        dest_key,
        NULL,
        src_key,
1163
        src_val
1164 1165 1166 1167
        );
}


1168
static int generate_row_for_put(
1169 1170 1171 1172 1173
    DB *dest_db, 
    DB *src_db,
    DBT *dest_key, 
    DBT *dest_val,
    const DBT *src_key, 
1174
    const DBT *src_val
1175 1176 1177 1178 1179 1180 1181 1182
    ) 
{
    return tokudb_generate_row(
        dest_db,
        src_db,
        dest_key,
        dest_val,
        src_key,
1183
        src_val
1184 1185 1186
        );
}

1187

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1188 1189 1190
ha_tokudb::ha_tokudb(handlerton * hton, TABLE_SHARE * table_arg):handler(hton, table_arg) 
    // flags defined in sql\handler.h
{
1191
    share = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1192
    int_table_flags = HA_REC_NOT_IN_SEQ  | HA_NULL_IN_KEY | HA_CAN_INDEX_BLOBS | HA_PRIMARY_KEY_IN_READ_INDEX | 
1193
                    HA_FILE_BASED | HA_AUTO_PART_KEY | HA_TABLE_SCAN_ON_INDEX | HA_CAN_WRITE_DURING_OPTIMIZE;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1194 1195
    alloc_ptr = NULL;
    rec_buff = NULL;
1196
    rec_update_buff = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1197
    transaction = NULL;
1198
    is_fast_alter_running = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212
    cursor = NULL;
    fixed_cols_for_query = NULL;
    var_cols_for_query = NULL;
    num_fixed_cols_for_query = 0;
    num_var_cols_for_query = 0;
    unpack_entire_row = true;
    read_blobs = false;
    read_key = false;
    added_rows = 0;
    deleted_rows = 0;
    last_dup_key = UINT_MAX;
    using_ignore = 0;
    last_cursor_error = 0;
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1213 1214
    blob_buff = NULL;
    num_blob_bytes = 0;
1215 1216
    delay_updating_ai_metadata = false;
    ai_metadata_update_required = false;
1217 1218
    memset(mult_key_dbt, 0, sizeof(mult_key_dbt));
    memset(mult_rec_dbt, 0, sizeof(mult_rec_dbt));
1219 1220
    loader = NULL;
    abort_loader = false;
1221
    memset(&lc, 0, sizeof(lc));
1222
    lock.type = TL_IGNORE;
Yoni Fogel's avatar
Yoni Fogel committed
1223
    for (uint32_t i = 0; i < MAX_KEY+1; i++) {
1224
        mult_put_flags[i] = 0;
1225
        mult_del_flags[i] = DB_DELETE_ANY;
1226 1227
        mult_dbt_flags[i] = DB_DBT_REALLOC;
    }
1228 1229
    num_DBs_locked_in_bulk = false;
    lock_count = 0;
1230
    use_write_locks = false;
1231 1232 1233 1234 1235 1236 1237
    range_query_buff = NULL;
    size_range_query_buff = 0;
    bytes_used_in_range_query_buff = 0;
    curr_range_query_buff_offset = 0;
    doing_bulk_fetch = false;
    prelocked_left_range_size = 0;
    prelocked_right_range_size = 0;
1238
    tokudb_active_index = MAX_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1239 1240
}

1241 1242 1243
ha_tokudb::~ha_tokudb() {
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
//
// states if table has an auto increment column, if so, sets index where auto inc column is to index
// Parameters:
//      [out]   index - if auto inc exists, then this param is set to where it exists in table, if not, then unchanged
// Returns:
//      true if auto inc column exists, false otherwise
//
bool ha_tokudb::has_auto_increment_flag(uint* index) {
    //
    // check to see if we have auto increment field
    //
    bool ai_found = false;
    uint ai_index = 0;
    for (uint i = 0; i < table_share->fields; i++, ai_index++) {
        Field* field = table->field[i];
        if (field->flags & AUTO_INCREMENT_FLAG) {
            ai_found = true;
            *index = ai_index;
            break;
        }
    }
    return ai_found;
}

1268 1269
#define status_dict_pagesize 1024

1270
static int open_status_dictionary(DB** ptr, const char* name, DB_TXN* txn) {
1271 1272 1273
    int error;
    char* newname = NULL;
    uint open_mode = DB_THREAD;
Yoni Fogel's avatar
Yoni Fogel committed
1274
    uint32_t pagesize = 0;
1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293
    newname = (char *)my_malloc(
        get_max_dict_name_path_length(name), 
        MYF(MY_WME)
        );
    if (newname == NULL) {
        error = ENOMEM;
        goto cleanup;
    }
    make_name(newname, name, "status");
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
        TOKUDB_TRACE("open:%s\n", newname);
    }
    error = db_create(ptr, db_env, 0);
    if (error) { goto cleanup; }
    
    error = (*ptr)->open((*ptr), txn, newname, NULL, DB_BTREE, open_mode, 0);
    if (error) { 
        goto cleanup; 
    }
1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304

    error = (*ptr)->get_pagesize(*ptr, &pagesize);
    if (error) { 
        goto cleanup; 
    }

    if (pagesize > status_dict_pagesize) {
        error = (*ptr)->change_pagesize(*ptr, status_dict_pagesize);
        if (error) { goto cleanup; }
    }
    
1305 1306 1307
cleanup:
    if (error) {
        if (*ptr) {
1308 1309
            int r = (*ptr)->close(*ptr, 0);
            assert(r==0);
1310 1311 1312 1313 1314 1315 1316
            *ptr = NULL;
        }
    }
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1317
int ha_tokudb::open_main_dictionary(const char* name, bool is_read_only, DB_TXN* txn) {
1318 1319
    int error;    
    char* newname = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1320
    uint open_flags = (is_read_only ? DB_RDONLY : 0) | DB_THREAD;
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339

    assert(share->file == NULL);
    assert(share->key_file[primary_key] == NULL);

    newname = (char *)my_malloc(
        get_max_dict_name_path_length(name),
        MYF(MY_WME|MY_ZEROFILL)
        );
    if (newname == NULL) { 
        error = ENOMEM;
        goto exit;
    }
    make_name(newname, name, "main");

    error = db_create(&share->file, db_env, 0);
    if (error) {
        goto exit;
    }
    share->key_file[primary_key] = share->file;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1340

1341 1342 1343 1344 1345 1346 1347 1348
    error = share->file->open(share->file, txn, newname, NULL, DB_BTREE, open_flags, 0);
    if (error) {
        goto exit;
    }
    
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
        TOKUDB_TRACE("open:%s:file=%p\n", newname, share->file);
    }
1349 1350

    error = 0;
1351 1352 1353
exit:
    if (error) {
        if (share->file) {
1354
            int r = share->file->close(
1355 1356 1357
                share->file,
                0
                );
1358
            assert(r==0);
1359 1360 1361 1362 1363 1364 1365
            share->file = NULL;
            share->key_file[primary_key] = NULL;
        }
    }
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1366

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1367 1368 1369
//
// Open a secondary table, the key will be a secondary index, the data will be a primary key
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1370
int ha_tokudb::open_secondary_dictionary(DB** ptr, KEY* key_info, const char* name, bool is_read_only, DB_TXN* txn) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1371
    int error = ENOSYS;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1372
    char dict_name[MAX_DICT_NAME_LEN];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1373
    uint open_flags = (is_read_only ? DB_RDONLY : 0) | DB_THREAD;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1374
    char* newname = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1375 1376 1377 1378 1379 1380
    uint newname_len = 0;
    
    sprintf(dict_name, "key-%s", key_info->name);

    newname_len = get_max_dict_name_path_length(name);
    newname = (char *)my_malloc(newname_len, MYF(MY_WME|MY_ZEROFILL));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1381 1382 1383 1384
    if (newname == NULL) {
        error = ENOMEM;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1385
    make_name(newname, name, dict_name);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1386 1387 1388 1389 1390 1391


    if ((error = db_create(ptr, db_env, 0))) {
        my_errno = error;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1392

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1393

1394
    if ((error = (*ptr)->open(*ptr, txn, newname, NULL, DB_BTREE, open_flags, 0))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1395 1396 1397 1398 1399 1400 1401
        my_errno = error;
        goto cleanup;
    }
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
        TOKUDB_TRACE("open:%s:file=%p\n", newname, *ptr);
    }
cleanup:
1402 1403
    if (error) {
        if (*ptr) {
1404 1405
            int r = (*ptr)->close(*ptr, 0);
            assert(r==0);
1406 1407 1408
            *ptr = NULL;
        }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1409
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1410 1411 1412
    return error;
}

1413
static int initialize_col_pack_info(KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1414 1415 1416 1417
    int error = ENOSYS;
    //
    // set up the cp_info
    //
1418 1419
    assert(kc_info->cp_info[keynr] == NULL);
    kc_info->cp_info[keynr] = (COL_PACK_INFO *)my_malloc(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1420 1421 1422
        table_share->fields*sizeof(COL_PACK_INFO), 
        MYF(MY_WME | MY_ZEROFILL)
        );
1423
    if (kc_info->cp_info[keynr] == NULL) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1424 1425 1426 1427
        error = ENOMEM;
        goto exit;
    }
    {
Yoni Fogel's avatar
Yoni Fogel committed
1428 1429
    uint32_t curr_fixed_offset = 0;
    uint32_t curr_var_index = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1430
    for (uint j = 0; j < table_share->fields; j++) {
1431
        COL_PACK_INFO* curr = &kc_info->cp_info[keynr][j];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1432 1433 1434 1435
        //
        // need to set the offsets / indexes
        // offsets are calculated AFTER the NULL bytes
        //
1436 1437
        if (!bitmap_is_set(&kc_info->key_filters[keynr],j)) {
            if (kc_info->field_lengths[j]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1438
                curr->col_pack_val = curr_fixed_offset;
1439
                curr_fixed_offset += kc_info->field_lengths[j];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1440
            }
1441
            else if (kc_info->length_bytes[j]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1442 1443 1444 1445 1446 1447 1448 1449 1450
                curr->col_pack_val = curr_var_index;
                curr_var_index++;
            }
        }
    }
    
    //
    // set up the mcp_info
    //
1451
    kc_info->mcp_info[keynr].fixed_field_size = get_fixed_field_size(
1452
        kc_info,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1453 1454 1455
        table_share,
        keynr
        );
1456 1457
    kc_info->mcp_info[keynr].len_of_offsets = get_len_of_offsets(
        kc_info,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1458 1459 1460 1461 1462 1463 1464 1465 1466 1467
        table_share,
        keynr
        );

    error = 0;
    }
exit:
    return error;
}

1468 1469 1470 1471 1472 1473 1474 1475
// reset the kc_info state at keynr
static void reset_key_and_col_info(KEY_AND_COL_INFO *kc_info, uint keynr) {
    bitmap_clear_all(&kc_info->key_filters[keynr]);
    my_free(kc_info->cp_info[keynr], MYF(MY_ALLOW_ZERO_PTR));
    kc_info->cp_info[keynr] = NULL;
    kc_info->mcp_info[keynr] = (MULTI_COL_PACK_INFO) { 0, 0 };
}

1476
static int initialize_key_and_col_info(TABLE_SHARE* table_share, TABLE* table, KEY_AND_COL_INFO* kc_info, uint hidden_primary_key, uint primary_key) {
1477
    int error = 0;
Yoni Fogel's avatar
Yoni Fogel committed
1478 1479
    uint32_t curr_blob_field_index = 0;
    uint32_t max_var_bytes = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494
    //
    // fill in the field lengths. 0 means it is a variable sized field length
    // fill in length_bytes, 0 means it is fixed or blob
    //
    for (uint i = 0; i < table_share->fields; i++) {
        Field* field = table_share->field[i];
        TOKU_TYPE toku_type = mysql_to_toku_type(field);
        uint32 pack_length = 0;
        switch (toku_type) {
        case toku_type_int:
        case toku_type_double:
        case toku_type_float:
        case toku_type_fixbinary:
        case toku_type_fixstring:
            pack_length = field->pack_length();
1495
            assert(pack_length < 1<<16);
Yoni Fogel's avatar
Yoni Fogel committed
1496
            kc_info->field_lengths[i] = (uint16_t)pack_length;
1497
            kc_info->length_bytes[i] = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1498 1499
            break;
        case toku_type_blob:
1500 1501 1502
            kc_info->field_lengths[i] = 0;
            kc_info->length_bytes[i] = 0;
            kc_info->blob_fields[curr_blob_field_index] = i;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1503 1504 1505 1506 1507 1508 1509
            curr_blob_field_index++;
            break;
        case toku_type_varstring:
        case toku_type_varbinary:
            //
            // meaning it is variable sized
            //
1510 1511
            kc_info->field_lengths[i] = 0;
            kc_info->length_bytes[i] = (uchar)((Field_varstring *)field)->length_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1512
            max_var_bytes += field->field_length;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1513 1514 1515 1516 1517
            break;
        default:
            assert(false);
        }
    }
1518
    kc_info->num_blobs = curr_blob_field_index;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1519

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1520 1521 1522 1523 1524 1525
    //
    // initialize share->num_offset_bytes
    // because MAX_REF_LENGTH is 65536, we
    // can safely set num_offset_bytes to 1 or 2
    //
    if (max_var_bytes < 256) {
1526
        kc_info->num_offset_bytes = 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1527 1528
    }
    else {
1529
        kc_info->num_offset_bytes = 2;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1530 1531 1532
    }


Zardosht Kasheff's avatar
Zardosht Kasheff committed
1533 1534 1535 1536 1537 1538 1539
    for (uint i = 0; i < table_share->keys + test(hidden_primary_key); i++) {
        //
        // do the cluster/primary key filtering calculations
        //
        if (! (i==primary_key && hidden_primary_key) ){        
            if ( i == primary_key ) {
                set_key_filter(
1540
                    &kc_info->key_filters[primary_key],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1541 1542 1543 1544 1545
                    &table_share->key_info[primary_key],
                    table,
                    true
                    );
            }
1546
            else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1547
                set_key_filter(
1548
                    &kc_info->key_filters[i],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1549 1550 1551 1552 1553 1554
                    &table_share->key_info[i],
                    table,
                    true
                    );
                if (!hidden_primary_key) {
                    set_key_filter(
1555
                        &kc_info->key_filters[i],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1556 1557 1558 1559 1560 1561 1562 1563
                        &table_share->key_info[primary_key],
                        table,
                        true
                        );
                }
            }
        }
        if (i == primary_key || table_share->key_info[i].flags & HA_CLUSTERING) {
1564
            error = initialize_col_pack_info(kc_info,table_share,i);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1565 1566 1567 1568 1569 1570
            if (error) {
                goto exit;
            }
        }

    }
1571 1572 1573
exit:
    return error;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1574

1575 1576
bool ha_tokudb::can_replace_into_be_fast(TABLE_SHARE* table_share, KEY_AND_COL_INFO* kc_info, uint pk) {
    uint curr_num_DBs = table_share->keys + test(hidden_primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1577 1578 1579 1580 1581 1582
    bool ret_val;
    if (curr_num_DBs == 1) {
        ret_val = true;
        goto exit;
    }
    ret_val = true;
1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598
    for (uint curr_index = 0; curr_index < table_share->keys; curr_index++) {
        if (curr_index == pk) continue;
        KEY* curr_key_info = &table_share->key_info[curr_index];
        for (uint i = 0; i < curr_key_info->key_parts; i++) {
            uint16 curr_field_index = curr_key_info->key_part[i].field->field_index;
            if (!bitmap_is_set(&kc_info->key_filters[curr_index],curr_field_index)) {
                ret_val = false;
                goto exit;
            }
            if (bitmap_is_set(&kc_info->key_filters[curr_index], curr_field_index) &&
                !bitmap_is_set(&kc_info->key_filters[pk], curr_field_index)) {
                ret_val = false;
                goto exit;
            }
            
        }
1599
    }
1600 1601 1602
exit:
    return ret_val;
}
1603 1604 1605 1606 1607 1608 1609

int ha_tokudb::initialize_share(
    const char* name,
    int mode
    )
{
    int error = 0;
Yoni Fogel's avatar
Yoni Fogel committed
1610
    uint64_t num_rows = 0;
1611
    bool table_exists;
1612 1613
    DB_TXN* txn = NULL;
    bool do_commit = false;
1614
    THD* thd = ha_thd();
1615 1616
    tokudb_trx_data *trx = NULL;
    trx = (tokudb_trx_data *) thd_data_get(ha_thd(), tokudb_hton->slot);
1617
    if (thd_sql_command(thd) == SQLCOM_CREATE_TABLE && trx && trx->sub_sp_level) {
1618 1619 1620 1621 1622 1623 1624 1625
        txn = trx->sub_sp_level;
    }
    else {
        do_commit = true;
        error = db_env->txn_begin(db_env, 0, &txn, 0);
        if (error) { goto exit; }
    }

1626 1627 1628
    DBUG_PRINT("info", ("share->use_count %u", share->use_count));

    table_exists = true;
1629
    error = check_table_in_metadata(name, &table_exists, txn);
1630 1631 1632 1633 1634 1635 1636 1637 1638

    if (error) {
        goto exit;
    }
    if (!table_exists) {
        sql_print_error("table %s does not exist in metadata, was it moved from someplace else? Not opening table", name);
        error = HA_ADMIN_FAILED;
        goto exit;
    }
1639

1640
    error = get_status(txn);
1641 1642 1643
    if (error) {
        goto exit;
    }
1644
    if (share->version != HA_TOKU_VERSION) {
1645 1646 1647 1648
        error = ENOSYS;
        goto exit;
    }

1649 1650
    if (TOKU_PARTITION_WRITE_FRM_DATA) {
        // verify frm data for all tables
1651
        error = verify_frm_data(table->s->path.str, txn);
1652
        if (error)
1653
            goto exit;
1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664
    } else {
        // verify frm data for non-partitioned tables
        if (table->part_info == NULL) {
            error = verify_frm_data(table->s->path.str, txn);
            if (error)
                goto exit;
        } else {
            // remove the frm data for partitions since we are not maintaining it
            error = remove_frm_data(share->status_block, txn);
            if (error)
                goto exit;
1665 1666
        }
    }
1667

1668 1669 1670 1671 1672 1673 1674
    error = initialize_key_and_col_info(
        table_share,
        table, 
        &share->kc_info,
        hidden_primary_key,
        primary_key
        );
1675
    if (error) { goto exit; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1676
    
1677
    error = open_main_dictionary(name, mode == O_RDONLY, txn);
1678 1679 1680
    if (error) { goto exit; }

    share->has_unique_keys = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1681
    /* Open other keys;  These are part of the share structure */
1682
    for (uint i = 0; i < table_share->keys; i++) {
1683 1684 1685
        if (table_share->key_info[i].flags & HA_NOSAME) {
            share->has_unique_keys = true;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1686
        if (i != primary_key) {
1687
            error = open_secondary_dictionary(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1688 1689 1690
                &share->key_file[i],
                &table_share->key_info[i],
                name,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1691
                mode == O_RDONLY,
1692
                txn
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1693 1694 1695 1696 1697 1698
                );
            if (error) {
                goto exit;
            }
        }
    }
1699 1700 1701 1702 1703 1704
    share->replace_into_fast = can_replace_into_be_fast(
        table_share, 
        &share->kc_info, 
        primary_key
        );
        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1705 1706 1707 1708 1709
    if (!hidden_primary_key) {
        //
        // We need to set the ref_length to start at 5, to account for
        // the "infinity byte" in keys, and for placing the DBT size in the first four bytes
        //
Yoni Fogel's avatar
Yoni Fogel committed
1710
        ref_length = sizeof(uint32_t) + sizeof(uchar);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1711 1712 1713 1714 1715 1716 1717 1718 1719
        KEY_PART_INFO *key_part = table->key_info[primary_key].key_part;
        KEY_PART_INFO *end = key_part + table->key_info[primary_key].key_parts;
        for (; key_part != end; key_part++) {
            ref_length += key_part->field->max_packed_col_length(key_part->length);
        }
        share->status |= STATUS_PRIMARY_KEY_INIT;
    }
    share->ref_length = ref_length;

1720
    error = estimate_num_rows(share->file,&num_rows, txn);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737
    //
    // estimate_num_rows should not fail under normal conditions
    //
    if (error == 0) {
        share->rows = num_rows;
    }
    else {
        goto exit;
    }
    //
    // initialize auto increment data
    //
    share->has_auto_inc = has_auto_increment_flag(&share->ai_field_index);
    if (share->has_auto_inc) {
        init_auto_increment();
    }

1738
    if (may_table_be_empty(NULL)) {
1739 1740 1741 1742 1743 1744
        share->try_table_lock = true;
    }
    else {
        share->try_table_lock = false;
    }

1745 1746
    share->num_DBs = table_share->keys + test(hidden_primary_key);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1747 1748
    error = 0;
exit:
1749 1750 1751
    if (do_commit && txn) {
        commit_txn(txn,0);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1752 1753 1754
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1755 1756


Zardosht Kasheff's avatar
Zardosht Kasheff committed
1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767
//
// Creates and opens a handle to a table which already exists in a tokudb
// database.
// Parameters:
//      [in]   name - table name
//             mode - seems to specify if table is read only
//             test_if_locked - unused
// Returns:
//      0 on success
//      1 on error
//
1768
int ha_tokudb::open(const char *name, int mode, uint test_if_locked) {
1769
    TOKUDB_DBUG_ENTER("ha_tokudb::open %p %s", this, name);
1770
    THD* thd = ha_thd();
1771

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1772 1773
    int error = 0;
    int ret_val = 0;
1774
    uint curr_num_DBs = 0;
1775

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1776 1777 1778
    transaction = NULL;
    cursor = NULL;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1779

1780 1781 1782 1783 1784 1785
    /* Open primary key */
    hidden_primary_key = 0;
    if ((primary_key = table_share->primary_key) >= MAX_KEY) {
        // No primary key
        primary_key = table_share->keys;
        key_used_on_scan = MAX_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1786
        hidden_primary_key = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
Yoni Fogel's avatar
Yoni Fogel committed
1787
        ref_length = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH + sizeof(uint32_t);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1788
    } 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1789
    else {
1790
        key_used_on_scan = primary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1791
    }
1792
    curr_num_DBs = table_share->keys + test(hidden_primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1793

1794
    /* Need some extra memory in case of packed keys */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1795
    // the "+ 1" is for the first byte that states +/- infinity
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1796 1797
    // multiply everything by 2 to account for clustered keys having a key and primary key together
    max_key_length = 2*(table_share->max_key_length + MAX_REF_PARTS * 3 + sizeof(uchar));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1798 1799 1800
    alloc_ptr = my_multi_malloc(MYF(MY_WME),
        &key_buff, max_key_length, 
        &key_buff2, max_key_length, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1801
        &key_buff3, max_key_length, 
1802 1803
        &prelocked_left_range, max_key_length, 
        &prelocked_right_range, max_key_length, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1804
        &primary_key_buff, (hidden_primary_key ? 0 : max_key_length),
Yoni Fogel's avatar
Yoni Fogel committed
1805 1806
        &fixed_cols_for_query, table_share->fields*sizeof(uint32_t),
        &var_cols_for_query, table_share->fields*sizeof(uint32_t),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1807 1808 1809 1810 1811
        NullS
        );
    if (alloc_ptr == NULL) {
        ret_val = 1;
        goto exit;
1812 1813
    }

1814 1815 1816 1817 1818 1819 1820
    size_range_query_buff = get_tokudb_read_buf_size(thd);
    range_query_buff = (uchar *)my_malloc(size_range_query_buff, MYF(MY_WME));
    if (range_query_buff == NULL) {
        ret_val = 1;
        goto exit;
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1821
    alloced_rec_buff_length = table_share->rec_buff_length + table_share->fields;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1822
    rec_buff = (uchar *) my_malloc(alloced_rec_buff_length, MYF(MY_WME));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1823 1824 1825
    if (rec_buff == NULL) {
        ret_val = 1;
        goto exit;
1826
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1827

1828 1829 1830 1831 1832 1833 1834
    alloced_update_rec_buff_length = alloced_rec_buff_length;
    rec_update_buff = (uchar *) my_malloc(alloced_update_rec_buff_length, MYF(MY_WME));
    if (rec_update_buff == NULL) {
        ret_val = 1;
        goto exit;
    }

Yoni Fogel's avatar
Yoni Fogel committed
1835
    for (uint32_t i = 0; i < sizeof(mult_key_dbt)/sizeof(mult_key_dbt[0]); i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1836
        mult_key_dbt[i].flags = DB_DBT_REALLOC;
1837 1838
    }

Yoni Fogel's avatar
Yoni Fogel committed
1839
    for (uint32_t i = 0; i < curr_num_DBs; i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1840
        mult_rec_dbt[i].flags = DB_DBT_REALLOC;
1841 1842
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1843 1844 1845 1846 1847
    /* Init shared structure */
    share = get_share(name, table_share);
    if (share == NULL) {
        ret_val = 1;
        goto exit;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1848 1849
    }

1850 1851 1852 1853
    thr_lock_data_init(&share->lock, &lock, NULL);

    /* Fill in shared structure, if needed */
    pthread_mutex_lock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1854
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
1855 1856
        TOKUDB_TRACE("tokudbopen:%p:share=%p:file=%p:table=%p:table->s=%p:%d\n", 
                     this, share, share->file, table, table->s, share->use_count);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1857
    }
1858
    if (!share->use_count++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1859 1860 1861 1862 1863
        ret_val = initialize_share(
            name,
            mode
            );
        if (ret_val) {
1864
            free_share(share, 1);
1865
            goto exit;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1866
        }
1867 1868 1869 1870
    }
    ref_length = share->ref_length;     // If second open
    pthread_mutex_unlock(&share->mutex);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1871
    key_read = false;
1872
    stats.block_size = 1<<20;    // QQQ Tokudb DB block size
1873

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1874 1875
    init_hidden_prim_key_info();

1876 1877
    info(HA_STATUS_NO_LOCK | HA_STATUS_VARIABLE | HA_STATUS_CONST);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1878 1879
exit:
    if (ret_val) {
1880 1881
        my_free(range_query_buff, MYF(MY_ALLOW_ZERO_PTR));
        range_query_buff = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1882 1883 1884 1885
        my_free(alloc_ptr, MYF(MY_ALLOW_ZERO_PTR));
        alloc_ptr = NULL;
        my_free(rec_buff, MYF(MY_ALLOW_ZERO_PTR));
        rec_buff = NULL;
1886 1887
        my_free(rec_update_buff, MYF(MY_ALLOW_ZERO_PTR));
        rec_update_buff = NULL;
1888
        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1889 1890 1891 1892 1893
        if (error) {
            my_errno = error;
        }
    }
    TOKUDB_DBUG_RETURN(ret_val);
1894 1895
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1896 1897 1898 1899 1900 1901 1902 1903 1904
//
// estimate the number of rows in a DB
// Parameters:
//      [in]    db - DB whose number of rows will be estimated
//      [out]   num_rows - number of estimated rows in db
// Returns:
//      0 on success
//      error otherwise
//
Yoni Fogel's avatar
Yoni Fogel committed
1905
int ha_tokudb::estimate_num_rows(DB* db, uint64_t* num_rows, DB_TXN* txn) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1906 1907 1908
    int error = ENOSYS;
    DBC* crsr = NULL;
    bool do_commit = false;
1909
    DB_BTREE_STAT64 dict_stats;
1910
    DB_TXN* txn_to_use = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1911

1912 1913
    if (txn == NULL) {
        error = db_env->txn_begin(db_env, 0, &txn_to_use, DB_READ_UNCOMMITTED);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1914 1915 1916
        if (error) goto cleanup;
        do_commit = true;
    }
1917 1918 1919
    else {
        txn_to_use = txn;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1920

1921 1922
    error = db->stat64(
        share->file, 
1923
        txn_to_use, 
1924
        &dict_stats
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1925
        );
1926
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1927

1928
    *num_rows = dict_stats.bt_ndata;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1929 1930
    error = 0;
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1931
    if (crsr != NULL) {
1932 1933
        int r = crsr->c_close(crsr);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1934 1935
        crsr = NULL;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1936
    if (do_commit) {
1937 1938
        commit_txn(txn_to_use, 0);
        txn_to_use = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1939
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1940 1941 1942
    return error;
}

1943

1944
int ha_tokudb::write_to_status(DB* db, HA_METADATA_KEY curr_key_data, void* data, uint size, DB_TXN* txn ){
1945
    return write_metadata(db, &curr_key_data, sizeof curr_key_data, data, size, txn);
1946 1947
}

1948 1949 1950
int ha_tokudb::remove_from_status(DB *db, HA_METADATA_KEY curr_key_data, DB_TXN *txn) {
    return remove_metadata(db, &curr_key_data, sizeof curr_key_data, txn);
}
1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970

int ha_tokudb::remove_metadata(DB* db, void* key_data, uint key_size, DB_TXN* transaction){
    int error;
    DBT key;
    DB_TXN* txn = NULL;
    bool do_commit = false;
    //
    // transaction to be used for putting metadata into status.tokudb
    //
    if (transaction == NULL) {
        error = db_env->txn_begin(db_env, 0, &txn, 0);
        if (error) { 
            goto cleanup;
        }
        do_commit = true;
    }
    else {
        txn = transaction;
    }

1971
    memset(&key, 0, sizeof(key));
1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
    key.data = key_data;
    key.size = key_size;
    error = db->del(db, txn, &key, DB_DELETE_ANY);
    if (error) { 
        goto cleanup; 
    }
    
    error = 0;
cleanup:
    if (do_commit && txn) {
        if (!error) {
            commit_txn(txn, DB_TXN_NOSYNC);
        }
        else {
            abort_txn(txn);
        }
    }
    return error;
1990 1991
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1992 1993 1994
//
// helper function to write a piece of metadata in to status.tokudb
//
1995
int ha_tokudb::write_metadata(DB* db, void* key_data, uint key_size, void* val_data, uint val_size, DB_TXN* transaction ){
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1996 1997 1998 1999
    int error;
    DBT key;
    DBT value;
    DB_TXN* txn = NULL;
2000
    bool do_commit = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2001 2002 2003
    //
    // transaction to be used for putting metadata into status.tokudb
    //
2004 2005 2006 2007 2008 2009 2010 2011 2012
    if (transaction == NULL) {
        error = db_env->txn_begin(db_env, 0, &txn, 0);
        if (error) { 
            goto cleanup;
        }
        do_commit = true;
    }
    else {
        txn = transaction;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2013 2014
    }

2015 2016
    memset(&key, 0, sizeof(key));
    memset(&value, 0, sizeof(value));
2017 2018 2019 2020
    key.data = key_data;
    key.size = key_size;
    value.data = val_data;
    value.size = val_size;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2021 2022 2023 2024 2025 2026 2027
    error = db->put(db, txn, &key, &value, 0);
    if (error) { 
        goto cleanup; 
    }
    
    error = 0;
cleanup:
2028
    if (do_commit && txn) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2029
        if (!error) {
2030
            commit_txn(txn, DB_TXN_NOSYNC);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2031 2032
        }
        else {
2033
            abort_txn(txn);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2034 2035 2036 2037 2038
        }
    }
    return error;
}

2039
int ha_tokudb::write_frm_data(DB* db, DB_TXN* txn, const char* frm_name) {
2040
    TOKUDB_DBUG_ENTER("ha_tokudb::write_frm_data %p %p %p %s", this, db, txn, frm_name);
2041

2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057
    uchar* frm_data = NULL;
    size_t frm_len = 0;
    int error = 0;

    error = readfrm(frm_name,&frm_data,&frm_len);
    if (error) { goto cleanup; }
    
    error = write_to_status(db,hatoku_frm_data,frm_data,(uint)frm_len, txn);
    if (error) { goto cleanup; }

    error = 0;
cleanup:
    my_free(frm_data, MYF(MY_ALLOW_ZERO_PTR));
    TOKUDB_DBUG_RETURN(error);
}

2058 2059 2060 2061
int ha_tokudb::remove_frm_data(DB *db, DB_TXN *txn) {
    return remove_from_status(db, hatoku_frm_data, txn);
}

2062 2063 2064 2065
static int
smart_dbt_callback_verify_frm (DBT const *key, DBT  const *row, void *context) {
    DBT* stored_frm = (DBT *)context;
    stored_frm->size = row->size;
2066 2067 2068
    stored_frm->data = (uchar *)my_malloc(row->size, MYF(MY_WME));
    assert(stored_frm->data);
    memcpy(stored_frm->data, row->data, row->size);
2069 2070 2071
    return 0;
}

2072
int ha_tokudb::verify_frm_data(const char* frm_name, DB_TXN* txn) {
2073
    TOKUDB_DBUG_ENTER("ha_tokudb::verify_frm_data %s", frm_name);
2074 2075
    uchar* mysql_frm_data = NULL;
    size_t mysql_frm_len = 0;
2076
    DBT key, stored_frm;
2077 2078 2079
    int error = 0;
    HA_METADATA_KEY curr_key = hatoku_frm_data;

2080 2081
    memset(&key, 0, sizeof(key));
    memset(&stored_frm, 0, sizeof(&stored_frm));
2082 2083 2084 2085 2086 2087
    // get the frm data from MySQL
    error = readfrm(frm_name,&mysql_frm_data,&mysql_frm_len);
    if (error) { goto cleanup; }

    key.data = &curr_key;
    key.size = sizeof(curr_key);
2088
    error = share->status_block->getf_set(
2089
        share->status_block, 
2090 2091
        txn,
        0,
2092
        &key, 
2093 2094
        smart_dbt_callback_verify_frm, 
        &stored_frm
2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108
        );
    if (error == DB_NOTFOUND) {
        // if not found, write it
        error = write_frm_data(
            share->status_block,
            txn,
            frm_name
            );
        goto cleanup;
    }
    else if (error) {
        goto cleanup;
    }

2109 2110
    if (stored_frm.size != mysql_frm_len || 
        memcmp(stored_frm.data, mysql_frm_data, stored_frm.size))
2111 2112 2113 2114
    {
        error = HA_ERR_TABLE_DEF_CHANGED;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2115

2116 2117 2118
    error = 0;
cleanup:
    my_free(mysql_frm_data, MYF(MY_ALLOW_ZERO_PTR));
2119
    my_free(stored_frm.data, MYF(MY_ALLOW_ZERO_PTR));
2120 2121
    TOKUDB_DBUG_RETURN(error);
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132

//
// Updates status.tokudb with a new max value used for the auto increment column
// Parameters:
//      [in]    db - this will always be status.tokudb
//              val - value to store
//  Returns:
//      0 on success, error otherwise
//
//
int ha_tokudb::update_max_auto_inc(DB* db, ulonglong val){
2133
    return write_to_status(db,hatoku_max_ai,&val,sizeof(val), NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146
}

//
// Writes the initial auto increment value, as specified by create table
// so if a user does "create table t1 (a int auto_increment, primary key (a)) auto_increment=100",
// then the value 100 will be stored here in val
// Parameters:
//      [in]    db - this will always be status.tokudb
//              val - value to store
//  Returns:
//      0 on success, error otherwise
//
//
2147 2148
int ha_tokudb::write_auto_inc_create(DB* db, ulonglong val, DB_TXN* txn){
    return write_to_status(db,hatoku_ai_create_value,&val,sizeof(val), txn);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2149
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2150 2151


Zardosht Kasheff's avatar
Zardosht Kasheff committed
2152 2153 2154
//
// Closes a handle to a table. 
//
2155
int ha_tokudb::close(void) {
2156
    TOKUDB_DBUG_ENTER("ha_tokudb::close %p", this);
2157
    TOKUDB_DBUG_RETURN(__close(0));
2158 2159 2160
}

int ha_tokudb::__close(int mutex_is_locked) {
2161
    TOKUDB_DBUG_ENTER("ha_tokudb::__close %p", this);
2162
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) 
2163
        TOKUDB_TRACE("close:%p\n", this);
2164
    my_free(rec_buff, MYF(MY_ALLOW_ZERO_PTR));
2165
    my_free(rec_update_buff, MYF(MY_ALLOW_ZERO_PTR));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2166
    my_free(blob_buff, MYF(MY_ALLOW_ZERO_PTR));
2167
    my_free(alloc_ptr, MYF(MY_ALLOW_ZERO_PTR));
2168
    my_free(range_query_buff, MYF(MY_ALLOW_ZERO_PTR));
Yoni Fogel's avatar
Yoni Fogel committed
2169
    for (uint32_t i = 0; i < sizeof(mult_rec_dbt)/sizeof(mult_rec_dbt[0]); i++) {
2170 2171 2172 2173
        if (mult_rec_dbt[i].flags == DB_DBT_REALLOC &&
                mult_rec_dbt[i].data != NULL) {
            free(mult_rec_dbt[i].data);
        }
2174
    }
Yoni Fogel's avatar
Yoni Fogel committed
2175
    for (uint32_t i = 0; i < sizeof(mult_key_dbt)/sizeof(mult_key_dbt[0]); i++) {
2176 2177 2178 2179
        if (mult_key_dbt[i].flags == DB_DBT_REALLOC &&
                mult_key_dbt[i].data != NULL) {
            free(mult_key_dbt[i].data);
        }
2180
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2181
    rec_buff = NULL;
2182
    rec_update_buff = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2183
    alloc_ptr = NULL;
2184
    ha_tokudb::reset();
2185
    TOKUDB_DBUG_RETURN(free_share(share, mutex_is_locked));
2186 2187
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2188 2189 2190 2191 2192 2193
//
// Reallocate record buffer (rec_buff) if needed
// If not needed, does nothing
// Parameters:
//          length - size of buffer required for rec_buff
//
2194
bool ha_tokudb::fix_rec_buff_for_blob(ulong length) {
2195
    if (!rec_buff || (length > alloced_rec_buff_length)) {
2196
        uchar *newptr;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2197
        if (!(newptr = (uchar *) my_realloc((void *) rec_buff, length, MYF(MY_ALLOW_ZERO_PTR))))
2198 2199 2200 2201 2202
            return 1;
        rec_buff = newptr;
        alloced_rec_buff_length = length;
    }
    return 0;
2203 2204
}

2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221
//
// Reallocate record buffer (rec_buff) if needed
// If not needed, does nothing
// Parameters:
//          length - size of buffer required for rec_buff
//
bool ha_tokudb::fix_rec_update_buff_for_blob(ulong length) {
    if (!rec_update_buff || (length > alloced_update_rec_buff_length)) {
        uchar *newptr;
        if (!(newptr = (uchar *) my_realloc((void *) rec_update_buff, length, MYF(MY_ALLOW_ZERO_PTR))))
            return 1;
        rec_update_buff= newptr;
        alloced_update_rec_buff_length = length;
    }
    return 0;
}

2222
/* Calculate max length needed for row */
2223 2224 2225 2226 2227
ulong ha_tokudb::max_row_length(const uchar * buf) {
    ulong length = table_share->reclength + table_share->fields * 2;
    uint *ptr, *end;
    for (ptr = table_share->blob_field, end = ptr + table_share->blob_fields; ptr != end; ptr++) {
        Field_blob *blob = ((Field_blob *) table->field[*ptr]);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2228
        length += blob->get_length((uchar *) (buf + field_offset(blob, table))) + 2;
2229 2230
    }
    return length;
2231 2232 2233 2234
}

/*
*/
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2235 2236 2237 2238 2239 2240 2241 2242 2243 2244
//
// take the row passed in as a DBT*, and convert it into a row in MySQL format in record
// Pack a row for storage.
// If the row is of fixed length, just store the  row 'as is'.
// If not, we will generate a packed row suitable for storage.
// This will only fail if we don't have enough memory to pack the row,
// which may only happen in rows with blobs, as the default row length is
// pre-allocated.
// Parameters:
//      [out]   row - row stored in DBT to be converted
2245
//      [out]   buf - buffer where row is packed
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2246 2247
//      [in]    record - row in MySQL format
//
2248

2249
int ha_tokudb::pack_row_in_buff(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2250
    DBT * row, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2251
    const uchar* record,
2252 2253
    uint index,
    uchar* row_buff
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2254 2255
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2256 2257 2258 2259
    uchar* fixed_field_ptr = NULL;
    uchar* var_field_offset_ptr = NULL;
    uchar* start_field_data_ptr = NULL;
    uchar* var_field_data_ptr = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2260
    int r = ENOSYS;
2261
    memset((void *) row, 0, sizeof(*row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2262 2263 2264

    my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);
    
2265
    // Copy null bytes
2266 2267
    memcpy(row_buff, record, table_share->null_bytes);
    fixed_field_ptr = row_buff + table_share->null_bytes;
2268
    var_field_offset_ptr = fixed_field_ptr + share->kc_info.mcp_info[index].fixed_field_size;
2269 2270
    start_field_data_ptr = var_field_offset_ptr + share->kc_info.mcp_info[index].len_of_offsets;
    var_field_data_ptr = var_field_offset_ptr + share->kc_info.mcp_info[index].len_of_offsets;
2271

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2272
    // assert that when the hidden primary key exists, primary_key_offsets is NULL
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2273 2274 2275
    for (uint i = 0; i < table_share->fields; i++) {
        Field* field = table->field[i];
        uint curr_field_offset = field_offset(field, table);
2276
        if (bitmap_is_set(&share->kc_info.key_filters[index],i)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2277 2278
            continue;
        }
2279
        if (share->kc_info.field_lengths[i]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2280 2281 2282
            fixed_field_ptr = pack_fixed_field(
                fixed_field_ptr,
                record + curr_field_offset, 
2283
                share->kc_info.field_lengths[i]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2284 2285
                );
        }
2286
        else if (share->kc_info.length_bytes[i]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2287 2288 2289 2290 2291
            var_field_data_ptr = pack_var_field(
                var_field_offset_ptr,
                var_field_data_ptr,
                start_field_data_ptr,
                record + curr_field_offset,
2292 2293
                share->kc_info.length_bytes[i],
                share->kc_info.num_offset_bytes
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2294
                );
2295
            var_field_offset_ptr += share->kc_info.num_offset_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2296 2297 2298
        }
    }

2299 2300
    for (uint i = 0; i < share->kc_info.num_blobs; i++) {
        Field* field = table->field[share->kc_info.blob_fields[i]];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2301
        var_field_data_ptr = pack_toku_field_blob(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2302
            var_field_data_ptr,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2303 2304
            record + field_offset(field, table),
            field
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2305
            );
2306
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2307

2308 2309
    row->data = row_buff;
    row->size = (size_t) (var_field_data_ptr - row_buff);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2310 2311 2312 2313
    r = 0;

    dbug_tmp_restore_column_map(table->write_set, old_map);
    return r;
2314 2315
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2316

2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335
int ha_tokudb::pack_row(
    DBT * row, 
    const uchar* record,
    uint index
    )
{
    return pack_row_in_buff(row,record,index,rec_buff);
}

int ha_tokudb::pack_old_row_for_update(
    DBT * row, 
    const uchar* record,
    uint index
    )
{
    return pack_row_in_buff(row,record,index,rec_update_buff);
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
2336
int ha_tokudb::unpack_blobs(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2337 2338
    uchar* record,
    const uchar* from_tokudb_blob,
Yoni Fogel's avatar
Yoni Fogel committed
2339
    uint32_t num_bytes,
2340
    bool check_bitmap
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2341 2342
    )
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2343 2344 2345
    uint error = 0;
    uchar* ptr = NULL;
    const uchar* buff = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2346
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2347
    // assert that num_bytes > 0 iff share->num_blobs > 0
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2348
    //
2349
    assert( !((share->kc_info.num_blobs == 0) && (num_bytes > 0)) );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361
    if (num_bytes > num_blob_bytes) {
        ptr = (uchar *)my_realloc((void *)blob_buff, num_bytes, MYF(MY_ALLOW_ZERO_PTR));
        if (ptr == NULL) {
            error = ENOMEM;
            goto exit;
        }
        blob_buff = ptr;
        num_blob_bytes = num_bytes;
    }
    
    memcpy(blob_buff, from_tokudb_blob, num_bytes);
    buff= blob_buff;
2362
    for (uint i = 0; i < share->kc_info.num_blobs; i++) {
Yoni Fogel's avatar
Yoni Fogel committed
2363
        uint32_t curr_field_index = share->kc_info.blob_fields[i]; 
2364 2365 2366 2367 2368
        bool skip = check_bitmap ? 
            !(bitmap_is_set(table->read_set,curr_field_index) || 
                bitmap_is_set(table->write_set,curr_field_index)) : 
            false;
        Field* field = table->field[curr_field_index];
Yoni Fogel's avatar
Yoni Fogel committed
2369
        uint32_t len_bytes = field->row_pack_length();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2370
        buff = unpack_toku_field_blob(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2371
            record + field_offset(field, table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2372
            buff,
2373
            len_bytes,
2374
            skip
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2375 2376 2377
            );
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2378 2379 2380
    error = 0;
exit:
    return error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2381 2382
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2383 2384 2385 2386 2387 2388
//
// take the row passed in as a DBT*, and convert it into a row in MySQL format in record
// Parameters:
//      [out]   record - row in MySQL format
//      [in]    row - row stored in DBT to be converted
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2389
int ha_tokudb::unpack_row(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2390 2391
    uchar* record, 
    DBT const *row, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2392 2393
    DBT const *key,
    uint index
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2394 2395
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2396 2397 2398 2399
    //
    // two cases, fixed length row, and variable length row
    // fixed length row is first below
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2400
    /* Copy null bits */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2401
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2402 2403 2404
    const uchar* fixed_field_ptr = (const uchar *) row->data;
    const uchar* var_field_offset_ptr = NULL;
    const uchar* var_field_data_ptr = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
2405
    uint32_t data_end_offset = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2406 2407
    memcpy(record, fixed_field_ptr, table_share->null_bytes);
    fixed_field_ptr += table_share->null_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2408

2409
    var_field_offset_ptr = fixed_field_ptr + share->kc_info.mcp_info[index].fixed_field_size;
2410
    var_field_data_ptr = var_field_offset_ptr + share->kc_info.mcp_info[index].len_of_offsets;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2411

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2412 2413 2414 2415 2416 2417 2418
    //
    // unpack the key, if necessary
    //
    if (!(hidden_primary_key && index == primary_key)) {
        unpack_key(record,key,index);
    }

Yoni Fogel's avatar
Yoni Fogel committed
2419
    uint32_t last_offset = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430
    //
    // we have two methods of unpacking, one if we need to unpack the entire row
    // the second if we unpack a subset of the entire row
    // first method here is if we unpack the entire row
    //
    if (unpack_entire_row) {
        //
        // fill in parts of record that are not part of the key
        //
        for (uint i = 0; i < table_share->fields; i++) {
            Field* field = table->field[i];
2431
            if (bitmap_is_set(&share->kc_info.key_filters[index],i)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2432 2433
                continue;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2434

2435
            if (share->kc_info.field_lengths[i]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2436 2437 2438
                fixed_field_ptr = unpack_fixed_field(
                    record + field_offset(field, table),
                    fixed_field_ptr,
2439
                    share->kc_info.field_lengths[i]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2440 2441
                    );
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2442
            //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2443 2444
            // here, we DO modify var_field_data_ptr or var_field_offset_ptr
            // as we unpack variable sized fields
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2445
            //
2446 2447
            else if (share->kc_info.length_bytes[i]) {
                switch (share->kc_info.num_offset_bytes) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461
                case (1):
                    data_end_offset = var_field_offset_ptr[0];
                    break;
                case (2):
                    data_end_offset = uint2korr(var_field_offset_ptr);
                    break;
                default:
                    assert(false);
                    break;
                }
                unpack_var_field(
                    record + field_offset(field, table),
                    var_field_data_ptr,
                    data_end_offset - last_offset,
2462
                    share->kc_info.length_bytes[i]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2463
                    );
2464
                var_field_offset_ptr += share->kc_info.num_offset_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2465 2466
                var_field_data_ptr += data_end_offset - last_offset;
                last_offset = data_end_offset;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2467
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2468
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2469
        error = unpack_blobs(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2470 2471
            record,
            var_field_data_ptr,
Yoni Fogel's avatar
Yoni Fogel committed
2472
            row->size - (uint32_t)(var_field_data_ptr - (const uchar *)row->data),
2473
            false
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2474
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2475 2476 2477
        if (error) {
            goto exit;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2478
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2479 2480 2481 2482
    //
    // in this case, we unpack only what is specified 
    // in fixed_cols_for_query and var_cols_for_query
    //
2483
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2484 2485 2486
        //
        // first the fixed fields
        //
Yoni Fogel's avatar
Yoni Fogel committed
2487
        for (uint32_t i = 0; i < num_fixed_cols_for_query; i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2488 2489 2490 2491
            uint field_index = fixed_cols_for_query[i];
            Field* field = table->field[field_index];
            unpack_fixed_field(
                record + field_offset(field, table),
2492 2493
                fixed_field_ptr + share->kc_info.cp_info[index][field_index].col_pack_val,
                share->kc_info.field_lengths[field_index]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2494
                );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2495 2496 2497
        }

        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2498 2499
        // now the var fields
        // here, we do NOT modify var_field_data_ptr or var_field_offset_ptr
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2500
        //
Yoni Fogel's avatar
Yoni Fogel committed
2501
        for (uint32_t i = 0; i < num_var_cols_for_query; i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2502 2503
            uint field_index = var_cols_for_query[i];
            Field* field = table->field[field_index];
Yoni Fogel's avatar
Yoni Fogel committed
2504 2505 2506
            uint32_t var_field_index = share->kc_info.cp_info[index][field_index].col_pack_val;
            uint32_t data_start_offset;
            uint32_t field_len;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2507
            
2508 2509 2510 2511 2512 2513 2514
            get_var_field_info(
                &field_len, 
                &data_start_offset, 
                var_field_index, 
                var_field_offset_ptr, 
                share->kc_info.num_offset_bytes
                );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2515 2516 2517 2518

            unpack_var_field(
                record + field_offset(field, table),
                var_field_data_ptr + data_start_offset,
2519 2520
                field_len,
                share->kc_info.length_bytes[field_index]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2521 2522 2523
                );
        }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2524 2525 2526 2527
        if (read_blobs) {
            //
            // now the blobs
            //
2528 2529 2530 2531 2532 2533
            get_blob_field_info(
                &data_end_offset, 
                share->kc_info.mcp_info[index].len_of_offsets,
                var_field_data_ptr, 
                share->kc_info.num_offset_bytes
                );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2534

2535
            var_field_data_ptr += data_end_offset;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2536
            error = unpack_blobs(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2537 2538
                record,
                var_field_data_ptr,
Yoni Fogel's avatar
Yoni Fogel committed
2539
                row->size - (uint32_t)(var_field_data_ptr - (const uchar *)row->data),
2540
                true
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2541
                );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2542 2543 2544
            if (error) {
                goto exit;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2545
        }
2546
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2547 2548 2549
    error = 0;
exit:
    return error;
2550 2551
}

Yoni Fogel's avatar
Yoni Fogel committed
2552
uint32_t ha_tokudb::place_key_into_mysql_buff(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2553 2554 2555 2556 2557
    KEY* key_info, 
    uchar * record, 
    uchar* data
    ) 
{
2558
    KEY_PART_INFO *key_part = key_info->key_part, *end = key_part + key_info->key_parts;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2559
    uchar *pos = data;
2560 2561

    for (; key_part != end; key_part++) {
2562 2563
        if (key_part->field->null_bit) {
            uint null_offset = get_null_offset(table, key_part->field);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2564
            if (*pos++ == NULL_COL_VAL) { // Null value
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2565 2566 2567 2568
                //
                // We don't need to reset the record data as we will not access it
                // if the null data is set
                //            
2569
                record[null_offset] |= key_part->field->null_bit;
2570 2571
                continue;
            }
2572
            record[null_offset] &= ~key_part->field->null_bit;
2573
        }
2574
#if !defined(MARIADB_BASE_VERSION)
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2575 2576 2577 2578
        //
        // HOPEFULLY TEMPORARY
        //
        assert(table->s->db_low_byte_first);
2579
#endif
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2580
        pos = unpack_toku_key_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2581
            record + field_offset(key_part->field, table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2582
            pos,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2583 2584
            key_part->field,
            key_part->length
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2585
            );
2586
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2587
    return pos-data;
2588 2589
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2590
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2591
// Store the key and the primary key into the row
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2592
// Parameters:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2593 2594 2595 2596
//      [out]   record - key stored in MySQL format
//      [in]    key - key stored in DBT to be converted
//              index -index into key_file that represents the DB 
//                  unpacking a key of
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2597
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2598
void ha_tokudb::unpack_key(uchar * record, DBT const *key, uint index) {
Yoni Fogel's avatar
Yoni Fogel committed
2599
    uint32_t bytes_read;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2600
    uchar *pos = (uchar *) key->data + 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2601 2602 2603 2604 2605
    bytes_read = place_key_into_mysql_buff(
        &table->key_info[index], 
        record, 
        pos
        );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2606
    if( (index != primary_key) && !hidden_primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2607 2608 2609
        //
        // also unpack primary key
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2610 2611 2612 2613 2614
        place_key_into_mysql_buff(
            &table->key_info[primary_key], 
            record, 
            pos+bytes_read
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2615 2616
    }
}
2617

Yoni Fogel's avatar
Yoni Fogel committed
2618
uint32_t ha_tokudb::place_key_into_dbt_buff(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2619 2620 2621 2622 2623 2624 2625
    KEY* key_info, 
    uchar * buff, 
    const uchar * record, 
    bool* has_null, 
    int key_length
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2626 2627 2628
    KEY_PART_INFO *key_part = key_info->key_part;
    KEY_PART_INFO *end = key_part + key_info->key_parts;
    uchar* curr_buff = buff;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2629
    *has_null = false;
2630
    for (; key_part != end && key_length > 0; key_part++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2631 2632 2633 2634 2635 2636
        //
        // accessing key_part->field->null_bit instead off key_part->null_bit
        // because key_part->null_bit is not set in add_index
        // filed ticket 862 to look into this
        //
        if (key_part->field->null_bit) {
2637
            /* Store 0 if the key part is a NULL part */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2638
            uint null_offset = get_null_offset(table, key_part->field);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2639
            if (record[null_offset] & key_part->field->null_bit) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2640
                *curr_buff++ = NULL_COL_VAL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2641
                *has_null = true;
2642 2643
                continue;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2644
            *curr_buff++ = NONNULL_COL_VAL;        // Store NOT NULL marker
2645
        }
2646
#if !defined(MARIADB_BASE_VERSION)
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2647
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2648 2649 2650
        // HOPEFULLY TEMPORARY
        //
        assert(table->s->db_low_byte_first);
2651
#endif
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2652
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2653 2654 2655 2656
        // accessing field_offset(key_part->field) instead off key_part->offset
        // because key_part->offset is SET INCORRECTLY in add_index
        // filed ticket 862 to look into this
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2657
        curr_buff = pack_toku_key_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2658
            curr_buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2659
            (uchar *) (record + field_offset(key_part->field, table)),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2660 2661
            key_part->field,
            key_part->length
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2662
            );
2663 2664
        key_length -= key_part->length;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683
    return curr_buff - buff;
}



//
// Create a packed key from a row. This key will be written as such
// to the index tree.  This will never fail as the key buffer is pre-allocated.
// Parameters:
//      [out]   key - DBT that holds the key
//      [in]    key_info - holds data about the key, such as it's length and offset into record
//      [out]   buff - buffer that will hold the data for key (unless 
//                  we have a hidden primary key)
//      [in]    record - row from which to create the key
//              key_length - currently set to MAX_KEY_LENGTH, is it size of buff?
// Returns:
//      the parameter key
//

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2684
DBT* ha_tokudb::create_dbt_key_from_key(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2685
    DBT * key,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2686
    KEY* key_info, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2687
    uchar * buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2688
    const uchar * record, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2689 2690
    bool* has_null,
    bool dont_pack_pk,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2691 2692 2693
    int key_length
    ) 
{
Yoni Fogel's avatar
Yoni Fogel committed
2694
    uint32_t size = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2695 2696 2697 2698 2699 2700 2701
    uchar* tmp_buff = buff;
    my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);

    key->data = buff;

    //
    // first put the "infinity" byte at beginning. States if missing columns are implicitly
2702
    // positive infinity or negative infinity or zero. For this, because we are creating key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2703 2704 2705
    // from a row, there is no way that columns can be missing, so in practice,
    // this will be meaningless. Might as well put in a value
    //
2706
    *tmp_buff++ = COL_ZERO;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2707
    size++;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2708 2709 2710 2711 2712 2713 2714
    size += place_key_into_dbt_buff(
        key_info, 
        tmp_buff, 
        record, 
        has_null, 
        key_length
        );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2715
    if (!dont_pack_pk) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733
        tmp_buff = buff + size;
        if (hidden_primary_key) {
            memcpy_fixed(tmp_buff, current_ident, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
            size += TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
        }
        else {
            bool tmp_bool = false;
            size += place_key_into_dbt_buff(
                &table->key_info[primary_key], 
                tmp_buff, 
                record, 
                &tmp_bool, 
                MAX_KEY_LENGTH //this parameter does not matter
                );
        }
    }

    key->size = size;
2734 2735
    DBUG_DUMP("key", (uchar *) key->data, key->size);
    dbug_tmp_restore_column_map(table->write_set, old_map);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
    return key;
}


//
// Create a packed key from a row. This key will be written as such
// to the index tree.  This will never fail as the key buffer is pre-allocated.
// Parameters:
//      [out]   key - DBT that holds the key
//              keynr - index for which to create the key
//      [out]   buff - buffer that will hold the data for key (unless 
//                  we have a hidden primary key)
//      [in]    record - row from which to create the key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2749
//      [out]   has_null - says if the key has a NULL value for one of its columns
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2750 2751 2752 2753
//              key_length - currently set to MAX_KEY_LENGTH, is it size of buff?
// Returns:
//      the parameter key
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2754 2755 2756 2757 2758 2759 2760 2761 2762
DBT *ha_tokudb::create_dbt_key_from_table(
    DBT * key, 
    uint keynr, 
    uchar * buff, 
    const uchar * record, 
    bool* has_null, 
    int key_length
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2763
    TOKUDB_DBUG_ENTER("ha_tokudb::create_dbt_key_from_table");
2764
    memset((void *) key, 0, sizeof(*key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2765
    if (hidden_primary_key && keynr == primary_key) {
2766 2767
        key->data = buff;
        memcpy(buff, &current_ident, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2768
        key->size = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2769
        *has_null = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2770 2771
        DBUG_RETURN(key);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2772
    DBUG_RETURN(create_dbt_key_from_key(key, &table->key_info[keynr],buff,record, has_null, (keynr == primary_key), key_length));
2773 2774
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786
DBT* ha_tokudb::create_dbt_key_for_lookup(
    DBT * key, 
    KEY* key_info, 
    uchar * buff, 
    const uchar * record, 
    bool* has_null, 
    int key_length
    )
{
    TOKUDB_DBUG_ENTER("ha_tokudb::create_dbt_key_from_lookup");
    DBUG_RETURN(create_dbt_key_from_key(key, key_info, buff, record, has_null, true, key_length));    
}
2787

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799
//
// Create a packed key from from a MySQL unpacked key (like the one that is
// sent from the index_read() This key is to be used to read a row
// Parameters:
//      [out]   key - DBT that holds the key
//              keynr - index for which to pack the key
//      [out]   buff - buffer that will hold the data for key
//      [in]    key_ptr - MySQL unpacked key
//              key_length - length of key_ptr
// Returns:
//      the parameter key
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2800 2801 2802 2803 2804 2805
DBT *ha_tokudb::pack_key(
    DBT * key, 
    uint keynr, 
    uchar * buff, 
    const uchar * key_ptr, 
    uint key_length, 
2806
    int8_t inf_byte
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2807 2808
    ) 
{
2809
    TOKUDB_DBUG_ENTER("ha_tokudb::pack_key");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2810
    KEY *key_info = &table->key_info[keynr];
2811 2812 2813 2814
    KEY_PART_INFO *key_part = key_info->key_part;
    KEY_PART_INFO *end = key_part + key_info->key_parts;
    my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);

2815
    memset((void *) key, 0, sizeof(*key));
2816 2817
    key->data = buff;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2818 2819 2820 2821
    //
    // first put the "infinity" byte at beginning. States if missing columns are implicitly
    // positive infinity or negative infinity
    //
2822
    *buff++ = (uchar)inf_byte;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2823

2824 2825 2826
    for (; key_part != end && (int) key_length > 0; key_part++) {
        uint offset = 0;
        if (key_part->null_bit) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2827 2828
            if (!(*key_ptr == 0)) {
                *buff++ = NULL_COL_VAL;
2829 2830 2831 2832
                key_length -= key_part->store_length;
                key_ptr += key_part->store_length;
                continue;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2833
            *buff++ = NONNULL_COL_VAL;
2834 2835
            offset = 1;         // Data is at key_ptr+1
        }
2836
#if !defined(MARIADB_BASE_VERSION)
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2837
        assert(table->s->db_low_byte_first);
2838
#endif
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2839
        buff = pack_key_toku_key_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2840
            buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2841
            (uchar *) key_ptr + offset,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2842 2843
            key_part->field,
            key_part->length
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2844
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2845
        
2846 2847 2848 2849 2850 2851 2852
        key_ptr += key_part->store_length;
        key_length -= key_part->store_length;
    }
    key->size = (buff - (uchar *) key->data);
    DBUG_DUMP("key", (uchar *) key->data, key->size);
    dbug_tmp_restore_column_map(table->write_set, old_map);
    DBUG_RETURN(key);
2853 2854
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2855 2856 2857 2858 2859
//
// get max used hidden primary key value
//
void ha_tokudb::init_hidden_prim_key_info() {
    TOKUDB_DBUG_ENTER("ha_tokudb::init_prim_key_info");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2860
    pthread_mutex_lock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2861
    if (!(share->status & STATUS_PRIMARY_KEY_INIT)) {
2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885
        int error = 0;
        THD* thd = ha_thd();
        DB_TXN* txn = NULL;
        DBC* c = NULL;
        tokudb_trx_data *trx = NULL;
        trx = (tokudb_trx_data *) thd_data_get(ha_thd(), tokudb_hton->slot);
        bool do_commit = false;
        if (thd_sql_command(thd) == SQLCOM_CREATE_TABLE && trx && trx->sub_sp_level) {
            txn = trx->sub_sp_level;
        }
        else {
            do_commit = true;
            error = db_env->txn_begin(db_env, 0, &txn, 0);
            assert(error == 0);
        }
        
        error = share->key_file[primary_key]->cursor(
            share->key_file[primary_key],
            txn,
            &c,
            0
            );
        assert(error == 0);
        DBT key,val;        
2886 2887
        memset(&key, 0, sizeof(key));
        memset(&val, 0, sizeof(val));
2888
        error = c->c_get(c, &key, &val, DB_LAST);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2889
        if (error == 0) {
2890 2891 2892 2893 2894 2895 2896
            assert(key.size == TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
            share->auto_ident = hpk_char_to_num((uchar *)key.data);
        }
        error = c->c_close(c);
        assert(error == 0);
        if (do_commit) {
            commit_txn(txn, 0);
2897
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2898
        share->status |= STATUS_PRIMARY_KEY_INIT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2899 2900
    }
    pthread_mutex_unlock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2901 2902
    DBUG_VOID_RETURN;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2903

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2904 2905 2906 2907 2908


/** @brief
    Get metadata info stored in status.tokudb
    */
2909
int ha_tokudb::get_status(DB_TXN* txn) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2910
    TOKUDB_DBUG_ENTER("ha_tokudb::get_status");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2911 2912 2913
    DBT key, value;
    HA_METADATA_KEY curr_key;
    int error;
2914

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2915 2916 2917 2918
    //
    // open status.tokudb
    //
    if (!share->status_block) {
2919 2920 2921
        error = open_status_dictionary(
            &share->status_block, 
            share->table_name, 
2922
            txn
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2923
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2924 2925 2926
        if (error) { 
            goto cleanup; 
        }
2927
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2928
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2929 2930 2931
    //
    // transaction to be used for putting metadata into status.tokudb
    //
2932 2933
    memset(&key, 0, sizeof(key));
    memset(&value, 0, sizeof(value));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2934 2935
    key.data = &curr_key;
    key.size = sizeof(curr_key);
2936
    value.flags = DB_DBT_USERMEM;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2937

2938 2939 2940 2941 2942 2943
    assert(share->status_block);
    //
    // get version
    //
    value.ulen = sizeof(share->version);
    value.data = &share->version;
2944
    curr_key = hatoku_new_version;
2945 2946 2947 2948 2949 2950 2951 2952
    error = share->status_block->get(
        share->status_block, 
        txn, 
        &key, 
        &value, 
        0
        );
    if (error == DB_NOTFOUND) {
2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985
        //
        // hack to keep handle the issues of going back and forth
        // between 5.0.3 to 5.0.4
        // the problem with going back and forth
        // is with storing the frm file, 5.0.4 stores it, 5.0.3 does not
        // so, if a user goes back and forth and alters the schema
        // the frm stored can get out of sync with the schema of the table
        // This can cause issues.
        // To take care of this, we are doing this versioning work here.
        // We change the key that stores the version. 
        // In 5.0.3, it is hatoku_old_version, in 5.0.4 it is hatoku_new_version
        // When we encounter a table that does not have hatoku_new_version
        // set, we give it the right one, and overwrite the old one with zero.
        // This ensures that 5.0.3 cannot open the table. Once it has been opened by 5.0.4
        //
        uint dummy_version = 0;
        share->version = HA_TOKU_ORIG_VERSION;
        error = write_to_status(
            share->status_block, 
            hatoku_new_version,
            &share->version,
            sizeof(share->version), 
            txn
            );
        if (error) { goto cleanup; }
        error = write_to_status(
            share->status_block, 
            hatoku_old_version,
            &dummy_version,
            sizeof(dummy_version), 
            txn
            );
        if (error) { goto cleanup; }
2986 2987 2988 2989
    }
    else if (error || value.size != sizeof(share->version)) {
        if (error == 0) {
            error = HA_ERR_INTERNAL_ERROR;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2990
        }
2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011
        goto cleanup;
    }
    //
    // get capabilities
    //
    curr_key = hatoku_capabilities;
    value.ulen = sizeof(share->capabilities);
    value.data = &share->capabilities;
    error = share->status_block->get(
        share->status_block, 
        txn, 
        &key, 
        &value, 
        0
        );
    if (error == DB_NOTFOUND) {
        share->capabilities= 0;
    }
    else if (error || value.size != sizeof(share->version)) {
        if (error == 0) {
            error = HA_ERR_INTERNAL_ERROR;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3012
        }
3013
        goto cleanup;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3014
    }
3015
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3016 3017 3018
    error = 0;
cleanup:
    TOKUDB_DBUG_RETURN(error);
3019 3020
}

3021 3022 3023
/** @brief
    Return an estimated of the number of rows in the table.
    Used when sorting to allocate buffers and by the optimizer.
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3024
    This is used in filesort.cc. 
3025
*/
3026
ha_rows ha_tokudb::estimate_rows_upper_bound() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3027
    TOKUDB_DBUG_ENTER("ha_tokudb::estimate_rows_upper_bound");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3028
    DBUG_RETURN(share->rows + HA_TOKUDB_EXTRA_ROWS);
3029
}
3030

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3031 3032 3033 3034
//
// Function that compares two primary keys that were saved as part of rnd_pos
// and ::position
//
3035
int ha_tokudb::cmp_ref(const uchar * ref1, const uchar * ref2) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3036 3037
    int ret_val = 0;
    ret_val = tokudb_compare_two_keys(
Yoni Fogel's avatar
Yoni Fogel committed
3038 3039 3040 3041
        ref1 + sizeof(uint32_t),
        *(uint32_t *)ref1,
        ref2 + sizeof(uint32_t),
        *(uint32_t *)ref2,
3042
        (uchar *)share->file->descriptor->dbt.data + 4,
Yoni Fogel's avatar
Yoni Fogel committed
3043
        *(uint32_t *)share->file->descriptor->dbt.data - 4,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3044
        false
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3045 3046
        );
    return ret_val;
3047 3048
}

3049
bool ha_tokudb::check_if_incompatible_data(HA_CREATE_INFO * info, uint table_changes) {
3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062
  //
  // This is a horrendous hack for now, as copied by InnoDB.
  // This states that if the auto increment create field has changed,
  // via a "alter table foo auto_increment=new_val", that this
  // change is incompatible, and to rebuild the entire table
  // This will need to be fixed
  //
  if ((info->used_fields & HA_CREATE_USED_AUTO) &&
      info->auto_increment_value != 0) {

    return COMPATIBLE_DATA_NO;
  }
  if (table_changes != IS_EQUAL_YES)
3063
    return COMPATIBLE_DATA_NO;
3064
  return COMPATIBLE_DATA_YES;
3065
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3066

3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078
//
// Method that is called before the beginning of many calls
// to insert rows (ha_tokudb::write_row). There is no guarantee
// that start_bulk_insert is called, however there is a guarantee
// that if start_bulk_insert is called, then end_bulk_insert may be
// called as well.
// Parameters:
//      [in]    rows - an estimate of the number of rows that will be inserted
//                     if number of rows is unknown (such as if doing 
//                     "insert into foo select * from bar), then rows 
//                     will be 0
//
3079 3080 3081 3082 3083
//
// This function returns true if the table MAY be empty.
// It is NOT meant to be a 100% check for emptiness.
// This is used for a bulk load optimization.
//
3084
bool ha_tokudb::may_table_be_empty(DB_TXN *txn) {
3085 3086 3087
    int error;
    bool ret_val = false;
    DBC* tmp_cursor = NULL;
3088
    DB_TXN* tmp_txn = NULL;
3089

3090 3091 3092 3093 3094 3095
    if (txn == NULL) {
        error = db_env->txn_begin(db_env, 0, &tmp_txn, 0);
        if (error) {
            goto cleanup;
        }
        txn = tmp_txn;
3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111
    }

    error = share->file->cursor(share->file, txn, &tmp_cursor, 0);
    if (error) {
        goto cleanup;
    }
    error = tmp_cursor->c_getf_next(tmp_cursor, 0, smart_dbt_do_nothing, NULL);
    if (error == DB_NOTFOUND) {
        ret_val = true;
    }
    else {
        ret_val = false;
    }
    error = 0;
cleanup:
    if (tmp_cursor) {
3112 3113
        int r = tmp_cursor->c_close(tmp_cursor);
        assert(r==0);
3114 3115
        tmp_cursor = NULL;
    }
3116 3117 3118
    if (tmp_txn) {
        commit_txn(tmp_txn, 0);
        tmp_txn = NULL;
3119 3120 3121 3122
    }
    return ret_val;
}

3123
void ha_tokudb::start_bulk_insert(ha_rows rows) {
3124 3125
    TOKUDB_DBUG_ENTER("ha_tokudb::start_bulk_insert");
    THD* thd = ha_thd();
3126
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
3127
    delay_updating_ai_metadata = true;
3128
    ai_metadata_update_required = false;
3129
    abort_loader = false;
3130 3131 3132 3133 3134 3135
    
    rw_rdlock(&share->num_DBs_lock);
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
    num_DBs_locked_in_bulk = true;
    lock_count = 0;
    
3136
    if (share->try_table_lock) {
3137
        if (get_prelock_empty(thd) && may_table_be_empty(transaction)) {
3138
            if (using_ignore || is_insert_ignore(thd) || thd->lex->duplicates != DUP_ERROR) {
3139 3140 3141 3142 3143 3144 3145
                acquire_table_lock(transaction, lock_write);
            }
            else {
                mult_dbt_flags[primary_key] = 0;
                if (!thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS) && !hidden_primary_key) {
                    mult_put_flags[primary_key] = DB_NOOVERWRITE;
                }
Yoni Fogel's avatar
Yoni Fogel committed
3146
                uint32_t loader_flags = (get_load_save_space(thd)) ? 
3147
                    LOADER_USE_PUTS : 0;
3148

3149 3150 3151 3152 3153 3154 3155 3156 3157
                int error = db_env->create_loader(
                    db_env, 
                    transaction, 
                    &loader, 
                    NULL, // no src_db needed
                    curr_num_DBs, 
                    share->key_file, 
                    mult_put_flags,
                    mult_dbt_flags,
3158
                    loader_flags
3159
                    );
3160 3161 3162 3163
                if (error) { 
                    assert(loader == NULL);
                    goto exit_try_table_lock;
                }
3164 3165 3166 3167

                lc.thd = thd;
                lc.ha = this;
                
3168
                error = loader->set_poll_function(loader, loader_poll_fun, &lc);
3169 3170 3171 3172
                assert(!error);

                error = loader->set_error_callback(loader, loader_dup_fun, &lc);
                assert(!error);
3173 3174

                trx->stmt_progress.using_loader = true;
3175
            }
3176
        }
3177
    exit_try_table_lock:
3178
        pthread_mutex_lock(&share->mutex);
3179
        share->try_table_lock = false;
3180 3181
        pthread_mutex_unlock(&share->mutex);
    }
3182
    DBUG_VOID_RETURN;
3183 3184 3185 3186 3187 3188 3189
}

//
// Method that is called at the end of many calls to insert rows
// (ha_tokudb::write_row). If start_bulk_insert is called, then
// this is guaranteed to be called.
//
3190
int ha_tokudb::end_bulk_insert(bool abort) {
3191
    TOKUDB_DBUG_ENTER("ha_tokudb::end_bulk_insert");
3192
    int error = 0;
3193
    THD* thd = ha_thd();
3194
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
3195
    bool using_loader = (loader != NULL);
3196 3197 3198 3199
    if (ai_metadata_update_required) {
        pthread_mutex_lock(&share->mutex);
        error = update_max_auto_inc(share->status_block, share->last_auto_increment);
        pthread_mutex_unlock(&share->mutex);
3200
        if (error) { goto cleanup; }
3201 3202 3203
    }
    delay_updating_ai_metadata = false;
    ai_metadata_update_required = false;
3204 3205
    loader_error = 0;
    if (loader) {
3206
        if (!abort_loader && !thd->killed) {
3207 3208
            error = loader->close(loader);
            loader = NULL;
3209 3210 3211 3212 3213 3214
            if (error) { 
                if (thd->killed) {
                    my_error(ER_QUERY_INTERRUPTED, MYF(0));
                }
                goto cleanup; 
            }
3215 3216 3217 3218

            for (uint i = 0; i < table_share->keys; i++) {
                if (table_share->key_info[i].flags & HA_NOSAME) {
                    bool is_unique;
3219 3220 3221
                    if (i == primary_key) {
                        continue;
                    }
3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237
                    error = is_index_unique(
                        &is_unique, 
                        transaction, 
                        share->key_file[i], 
                        &table->key_info[i]
                        );
                    if (error) goto cleanup;
                    if (!is_unique) {
                        error = HA_ERR_FOUND_DUPP_KEY;
                        last_dup_key = i;
                        goto cleanup;
                    }
                }
            }
        }
        else {
3238 3239
            error = sprintf(write_status_msg, "aborting bulk load"); 
            thd_proc_info(thd, write_status_msg);
3240 3241
            loader->abort(loader);
            loader = NULL;
3242
            share->try_table_lock = true;
3243 3244 3245 3246
        }
    }

cleanup:
3247 3248 3249 3250 3251
    if (num_DBs_locked_in_bulk) {
        rw_unlock(&share->num_DBs_lock);
    }
    num_DBs_locked_in_bulk = false;
    lock_count = 0;
3252
    if (loader) {
3253 3254
        error = sprintf(write_status_msg, "aborting bulk load"); 
        thd_proc_info(thd, write_status_msg);
3255 3256 3257 3258
        loader->abort(loader);
        loader = NULL;
    }
    abort_loader = false;
3259
    memset(&lc, 0, sizeof(lc));
3260 3261
    if (error || loader_error) {
        my_errno = error ? error : loader_error;
3262 3263 3264
        if (using_loader) {
            share->try_table_lock = true;
        }
3265
    }
3266
    trx->stmt_progress.using_loader = false;
3267
    TOKUDB_DBUG_RETURN(error ? error : loader_error);
3268 3269
}

3270 3271 3272
int ha_tokudb::end_bulk_insert() {
    return end_bulk_insert( false );
}
3273

3274 3275 3276 3277 3278
int ha_tokudb::is_index_unique(bool* is_unique, DB_TXN* txn, DB* db, KEY* key_info) {
    int error;
    DBC* tmp_cursor1 = NULL;
    DBC* tmp_cursor2 = NULL;
    DBT key1, key2, val, packed_key1, packed_key2;
Yoni Fogel's avatar
Yoni Fogel committed
3279
    uint64_t cnt = 0;
3280
    char status_msg[MAX_ALIAS_NAME + 200]; //buffer of 200 should be a good upper bound.
3281
    THD* thd = ha_thd();
3282 3283 3284 3285 3286
    memset(&key1, 0, sizeof(key1));
    memset(&key2, 0, sizeof(key2));
    memset(&val, 0, sizeof(val));
    memset(&packed_key1, 0, sizeof(packed_key1));
    memset(&packed_key2, 0, sizeof(packed_key2));
3287 3288 3289 3290 3291 3292
    *is_unique = true;
    
    error = db->cursor(
        db, 
        txn, 
        &tmp_cursor1, 
3293
        DB_SERIALIZABLE
3294 3295 3296 3297 3298 3299 3300
        );
    if (error) { goto cleanup; }

    error = db->cursor(
        db, 
        txn, 
        &tmp_cursor2,
3301
        DB_SERIALIZABLE
3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371
        );
    if (error) { goto cleanup; }

    
    error = tmp_cursor1->c_get(
        tmp_cursor1, 
        &key1, 
        &val, 
        DB_NEXT
        );
    if (error == DB_NOTFOUND) {
        *is_unique = true;
        error = 0;
        goto cleanup;
    }
    else if (error) { goto cleanup; }
    error = tmp_cursor2->c_get(
        tmp_cursor2, 
        &key2, 
        &val, 
        DB_NEXT
        );
    if (error) { goto cleanup; }

    error = tmp_cursor2->c_get(
        tmp_cursor2, 
        &key2, 
        &val, 
        DB_NEXT
        );
    if (error == DB_NOTFOUND) {
        *is_unique = true;
        error = 0;
        goto cleanup;
    }
    else if (error) { goto cleanup; }

    while (error != DB_NOTFOUND) {
        bool has_null1;
        bool has_null2;
        int cmp;
        place_key_into_mysql_buff(
            key_info,
            table->record[0], 
            (uchar *) key1.data + 1
            );
        place_key_into_mysql_buff(
            key_info,
            table->record[1], 
            (uchar *) key2.data + 1
            );
        
        create_dbt_key_for_lookup(
            &packed_key1,
            key_info,
            key_buff,
            table->record[0],
            &has_null1
            );
        create_dbt_key_for_lookup(
            &packed_key2,
            key_info,
            key_buff2,
            table->record[1],
            &has_null2
            );

        if (!has_null1 && !has_null2) {
            cmp = tokudb_prefix_cmp_dbt_key(db, &packed_key1, &packed_key2);
            if (cmp == 0) {
3372 3373 3374 3375 3376 3377
                memcpy(key_buff, key1.data, key1.size);
                place_key_into_mysql_buff(
                    key_info,
                    table->record[0], 
                    (uchar *) key_buff + 1
                    );
3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396
                *is_unique = false;
                break;
            }
        }

        error = tmp_cursor1->c_get(
            tmp_cursor1, 
            &key1, 
            &val, 
            DB_NEXT
            );
        if (error) { goto cleanup; }
        error = tmp_cursor2->c_get(
            tmp_cursor2, 
            &key2, 
            &val, 
            DB_NEXT
            );
        if (error && (error != DB_NOTFOUND)) { goto cleanup; }
3397 3398

        cnt++;
3399 3400 3401 3402
        if ((cnt % 10000) == 0) {
            sprintf(
                status_msg, 
                "Verifying index uniqueness: Checked %llu of %llu rows in key-%s.", 
3403
                (long long unsigned) cnt, 
3404 3405 3406
                share->rows, 
                key_info->name);
            thd_proc_info(thd, status_msg);
3407 3408 3409 3410 3411 3412
            if (thd->killed) {
                my_error(ER_QUERY_INTERRUPTED, MYF(0));
                error = ER_QUERY_INTERRUPTED;
                goto cleanup;
            }
        }
3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428
    }

    error = 0;

cleanup:
    if (tmp_cursor1) {
        tmp_cursor1->c_close(tmp_cursor1);
        tmp_cursor1 = NULL;
    }
    if (tmp_cursor2) {
        tmp_cursor2->c_close(tmp_cursor2);
        tmp_cursor2 = NULL;
    }
    return error;
}

3429 3430 3431 3432 3433 3434 3435
int ha_tokudb::is_val_unique(bool* is_unique, uchar* record, KEY* key_info, uint dict_index, DB_TXN* txn) {
    DBT key;
    int error = 0;
    bool has_null;
    DBC* tmp_cursor = NULL;
    struct index_read_info ir_info;
    struct smart_dbt_info info;
3436
    memset((void *)&key, 0, sizeof(key));
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461
    info.ha = this;
    info.buf = NULL;
    info.keynr = dict_index;

    ir_info.smart_dbt_info = info;
    
    create_dbt_key_for_lookup(
        &key,
        key_info,
        key_buff3,
        record,
        &has_null
        );
    ir_info.orig_key = &key;

    if (has_null) {
        error = 0;
        *is_unique = true;
        goto cleanup;
    }
    
    error = share->key_file[dict_index]->cursor(
        share->key_file[dict_index], 
        txn, 
        &tmp_cursor, 
3462
        DB_SERIALIZABLE
3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506
        );
    if (error) { goto cleanup; }

    error = tmp_cursor->c_getf_set_range(
        tmp_cursor, 
        0, 
        &key, 
        smart_dbt_callback_lookup, 
        &ir_info
        );
    if (error == DB_NOTFOUND) {
        *is_unique = true;
        error = 0;
        goto cleanup;
    }
    else if (error) {
        goto cleanup;
    }
    if (ir_info.cmp) {
        *is_unique = true;
    }
    else {
        *is_unique = false;
    }
    error = 0;

cleanup:
    if (tmp_cursor) {
        int r = tmp_cursor->c_close(tmp_cursor);
        assert(r==0);
        tmp_cursor = NULL;
    }
    return error;
}

int ha_tokudb::do_uniqueness_checks(uchar* record, DB_TXN* txn, THD* thd) {
    int error;
    //
    // first do uniqueness checks
    //
    if (share->has_unique_keys && !thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS)) {
        for (uint keynr = 0; keynr < table_share->keys; keynr++) {
            bool is_unique_key = table->key_info[keynr].flags & HA_NOSAME;
            bool is_unique = false;
3507 3508 3509 3510 3511 3512
            //
            // don't need to do check for primary key
            //
            if (keynr == primary_key) {
                continue;
            }
3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534
            if (!is_unique_key) {
                continue;
            }
            //
            // if unique key, check uniqueness constraint
            // but, we do not need to check it if the key has a null
            // and we do not need to check it if unique_checks is off
            //
            error = is_val_unique(&is_unique, record, &table->key_info[keynr], keynr, txn);
            if (error) { goto cleanup; }
            if (!is_unique) {
                error = DB_KEYEXIST;
                last_dup_key = keynr;
                goto cleanup;
            }
        }
    }    
    error = 0;
cleanup:
    return error;
}

3535
void ha_tokudb::test_row_packing(uchar* record, DBT* pk_key, DBT* pk_val) {
3536 3537 3538 3539 3540 3541 3542 3543 3544 3545
    int error;
    DBT row, key;
    //
    // variables for testing key packing, only used in some debug modes
    //
    uchar* tmp_pk_key_data = NULL;
    uchar* tmp_pk_val_data = NULL;
    DBT tmp_pk_key;
    DBT tmp_pk_val;
    bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3546
    int cmp;
3547

3548 3549
    memset(&tmp_pk_key, 0, sizeof(DBT));
    memset(&tmp_pk_val, 0, sizeof(DBT));
3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565

    //
    //use for testing the packing of keys
    //
    tmp_pk_key_data = (uchar *)my_malloc(pk_key->size, MYF(MY_WME));
    assert(tmp_pk_key_data);
    tmp_pk_val_data = (uchar *)my_malloc(pk_val->size, MYF(MY_WME));
    assert(tmp_pk_val_data);
    memcpy(tmp_pk_key_data, pk_key->data, pk_key->size);
    memcpy(tmp_pk_val_data, pk_val->data, pk_val->size);
    tmp_pk_key.data = tmp_pk_key_data;
    tmp_pk_key.size = pk_key->size;
    tmp_pk_val.data = tmp_pk_val_data;
    tmp_pk_val.size = pk_val->size;

    for (uint keynr = 0; keynr < table_share->keys; keynr++) {
Yoni Fogel's avatar
Yoni Fogel committed
3566
        uint32_t tmp_num_bytes = 0;
3567
        uchar* row_desc = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
3568
        uint32_t desc_size = 0;
3569 3570 3571 3572 3573
        
        if (keynr == primary_key) {
            continue;
        }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3574
        create_dbt_key_from_table(&key, keynr, key_buff2, record, &has_null); 
3575 3576 3577 3578

        //
        // TEST
        //
3579
        row_desc = (uchar *)share->key_file[keynr]->descriptor->dbt.data;
Yoni Fogel's avatar
Yoni Fogel committed
3580 3581
        row_desc += (*(uint32_t *)row_desc);
        desc_size = (*(uint32_t *)row_desc) - 4;
3582 3583 3584 3585 3586 3587 3588 3589 3590
        row_desc += 4;
        tmp_num_bytes = pack_key_from_desc(
            key_buff3,
            row_desc,
            desc_size,
            &tmp_pk_key,
            &tmp_pk_val
            );
        assert(tmp_num_bytes == key.size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3591
        cmp = memcmp(key_buff3,key_buff2,tmp_num_bytes);
3592 3593 3594 3595 3596 3597
        assert(cmp == 0);

        //
        // test key packing of clustering keys
        //
        if (table->key_info[keynr].flags & HA_CLUSTERING) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3598
            error = pack_row(&row, (const uchar *) record, keynr);
3599
            assert(error == 0);
3600 3601 3602
            uchar* tmp_buff = NULL;
            tmp_buff = (uchar *)my_malloc(alloced_rec_buff_length,MYF(MY_WME));
            assert(tmp_buff);
3603
            row_desc = (uchar *)share->key_file[keynr]->descriptor->dbt.data;
Yoni Fogel's avatar
Yoni Fogel committed
3604 3605 3606
            row_desc += (*(uint32_t *)row_desc);
            row_desc += (*(uint32_t *)row_desc);
            desc_size = (*(uint32_t *)row_desc) - 4;
3607 3608 3609 3610 3611 3612 3613 3614
            row_desc += 4;
            tmp_num_bytes = pack_clustering_val_from_desc(
                tmp_buff,
                row_desc,
                desc_size,
                &tmp_pk_val
                );
            assert(tmp_num_bytes == row.size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3615
            cmp = memcmp(tmp_buff,rec_buff,tmp_num_bytes);
3616 3617 3618 3619 3620
            assert(cmp == 0);
            my_free(tmp_buff,MYF(MY_ALLOW_ZERO_PTR));
        }
    }

3621 3622 3623 3624
    //
    // copy stuff back out
    //
    error = pack_row(pk_val, (const uchar *) record, primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3625 3626 3627
    assert(pk_val->size == tmp_pk_val.size);
    cmp = memcmp(pk_val->data, tmp_pk_val_data, pk_val->size);    
    assert( cmp == 0);
3628

3629 3630 3631 3632
    my_free(tmp_pk_key_data,MYF(MY_ALLOW_ZERO_PTR));
    my_free(tmp_pk_val_data,MYF(MY_ALLOW_ZERO_PTR));
}

3633 3634 3635
//
// set the put flags for the main dictionary
//
3636 3637
void ha_tokudb::set_main_dict_put_flags(
    THD* thd, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3638
    bool opt_eligible,
Yoni Fogel's avatar
Yoni Fogel committed
3639
    uint32_t* put_flags
3640 3641
    ) 
{
Yoni Fogel's avatar
Yoni Fogel committed
3642
    uint32_t old_prelock_flags = 0;
3643 3644
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
    bool in_hot_index = share->num_DBs > curr_num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3645 3646
    bool using_ignore_flag_opt = do_ignore_flag_optimization(
            thd, table, share->replace_into_fast);
3647
    //
3648
    // optimization for "REPLACE INTO..." (and "INSERT IGNORE") command
3649
    // if the command is "REPLACE INTO" and the only table
3650 3651
    // is the main table (or all indexes are a subset of the pk), 
    // then we can simply insert the element
3652 3653 3654
    // with DB_YESOVERWRITE. If the element does not exist,
    // it will act as a normal insert, and if it does exist, it 
    // will act as a replace, which is exactly what REPLACE INTO is supposed
3655
    // to do. We cannot do this if otherwise, because then we lose
3656 3657
    // consistency between indexes
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3658 3659
    if (hidden_primary_key) 
    {
3660
        *put_flags = old_prelock_flags;
3661
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3662 3663
    else if (thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS)
            && !is_replace_into(thd) && !is_insert_ignore(thd))
3664
    {
3665
        *put_flags = old_prelock_flags;
3666
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3667 3668
    else if (using_ignore_flag_opt && is_replace_into(thd) 
            && !in_hot_index)
3669
    {
3670
        *put_flags = old_prelock_flags;
3671
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3672
    else if (opt_eligible && using_ignore_flag_opt && is_insert_ignore(thd) 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3673
            && !in_hot_index)
3674
    {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3675
        *put_flags = DB_NOOVERWRITE_NO_ERROR | old_prelock_flags;
3676 3677 3678
    }
    else 
    {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3679
        *put_flags = DB_NOOVERWRITE | old_prelock_flags;
3680
    }
3681 3682 3683 3684
}

int ha_tokudb::insert_row_to_main_dictionary(uchar* record, DBT* pk_key, DBT* pk_val, DB_TXN* txn) {
    int error = 0;
Yoni Fogel's avatar
Yoni Fogel committed
3685
    uint32_t put_flags = mult_put_flags[primary_key];
3686 3687 3688 3689 3690
    THD *thd = ha_thd();
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);

    assert(curr_num_DBs == 1);
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3691
    set_main_dict_put_flags(thd, true, &put_flags);
3692

3693 3694 3695 3696 3697 3698 3699
    error = share->file->put(
        share->file, 
        txn, 
        pk_key,
        pk_val, 
        put_flags
        );
3700 3701 3702 3703 3704 3705 3706

    if (error) {
        last_dup_key = primary_key;
        goto cleanup;
    }

cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3707 3708 3709
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3710
int ha_tokudb::insert_rows_to_dictionaries_mult(DBT* pk_key, DBT* pk_val, DB_TXN* txn, THD* thd) {
3711
    int error = 0;
3712
    uint curr_num_DBs = share->num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3713
    set_main_dict_put_flags(thd, true, &mult_put_flags[primary_key]);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760
    uint32_t i, flags = mult_put_flags[primary_key];

    // the insert ignore optimization uses DB_NOOVERWRITE_NO_ERROR, 
    // which is not allowed with env->put_multiple. 
    // we have to insert the rows one by one in this case.
    if (flags & DB_NOOVERWRITE_NO_ERROR) {
        DB * src_db = share->key_file[primary_key];
        for (i = 0; i < curr_num_DBs; i++) {
            DB * db = share->key_file[i];
            if (i == primary_key) {
                // if it's the primary key, insert the rows
                // as they are.
                error = db->put(db, txn, pk_key, pk_val, flags);
            } else {
                // generate a row for secondary keys.
                // use our multi put key/rec buffers
                // just as the ydb layer would have in
                // env->put_multiple(), except that
                // we will just do a put() right away.
                error = tokudb_generate_row(db, src_db,
                        &mult_key_dbt[i], &mult_rec_dbt[i], 
                        pk_key, pk_val);
                if (error != 0) {
                    goto out;
                }
                error = db->put(db, txn, &mult_key_dbt[i], 
                        &mult_rec_dbt[i], flags);
            }
            if (error != 0) {
                goto out;
            }
        }
    } else {
        // not insert ignore, so we can use put multiple
        error = db_env->put_multiple(
            db_env, 
            share->key_file[primary_key], 
            txn, 
            pk_key, 
            pk_val,
            curr_num_DBs, 
            share->key_file, 
            mult_key_dbt,
            mult_rec_dbt,
            mult_put_flags
            );
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3761

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3762
out:
3763 3764 3765 3766 3767 3768
    //
    // We break if we hit an error, unless it is a dup key error
    // and MySQL told us to ignore duplicate key errors
    //
    if (error) {
        last_dup_key = primary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3769 3770 3771
    }
    return error;
}
3772

3773 3774
volatile int ha_tokudb_write_row_wait = 0; // debug

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3775 3776 3777 3778 3779 3780 3781 3782
//
// Stores a row in the table, called when handling an INSERT query
// Parameters:
//      [in]    record - a row in MySQL format
// Returns:
//      0 on success
//      error otherwise
//
3783
int ha_tokudb::write_row(uchar * record) {
3784
    TOKUDB_DBUG_ENTER("ha_tokudb::write_row");
3785 3786
    while (ha_tokudb_write_row_wait) sleep(1); // debug

3787
    DBT row, prim_key;
3788
    int error;
3789
    THD *thd = ha_thd();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3790
    bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3791 3792
    DB_TXN* sub_trans = NULL;
    DB_TXN* txn = NULL;
3793
    tokudb_trx_data *trx = NULL;
3794
    uint curr_num_DBs;
3795
    bool create_sub_trans = false;
3796

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3797 3798 3799 3800
    //
    // some crap that needs to be done because MySQL does not properly abstract
    // this work away from us, namely filling in auto increment and setting auto timestamp
    //
3801
    statistic_increment(table->in_use->status_var.ha_write_count, &LOCK_status);
3802
#if MYSQL_VERSION_ID < 50600
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3803
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT) {
3804
        table->timestamp_field->set_time();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3805
    }
3806
#endif
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3807
    if (table->next_number_field && record == table->record[0]) {
3808
        update_auto_increment();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3809
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820

    //
    // check to see if some value for the auto increment column that is bigger
    // than anything else til now is being used. If so, update the metadata to reflect it
    // the goal here is we never want to have a dup key error due to a bad increment
    // of the auto inc field.
    //
    if (share->has_auto_inc && record == table->record[0]) {
        pthread_mutex_lock(&share->mutex);
        ulonglong curr_auto_inc = retrieve_auto_increment(
            table->field[share->ai_field_index]->key_type(), 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3821
            field_offset(table->field[share->ai_field_index], table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3822 3823 3824
            record
            );
        if (curr_auto_inc > share->last_auto_increment) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3825
            share->last_auto_increment = curr_auto_inc;
3826 3827
            if (delay_updating_ai_metadata) {
                ai_metadata_update_required = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3828
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3829 3830 3831 3832
            else {
                update_max_auto_inc(share->status_block, share->last_auto_increment);
            }
         }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3833 3834 3835
        pthread_mutex_unlock(&share->mutex);
    }

3836 3837 3838
    //
    // grab reader lock on numDBs_lock
    //
3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849
    if (!num_DBs_locked_in_bulk) {
        rw_rdlock(&share->num_DBs_lock);
    }
    else {
        lock_count++;
        if (lock_count >= 2000) {
            rw_unlock(&share->num_DBs_lock);
            rw_rdlock(&share->num_DBs_lock);
            lock_count = 0;
        }
    }
3850
    curr_num_DBs = share->num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3851 3852 3853 3854
    
    if (hidden_primary_key) {
        get_auto_primary_key(current_ident);
    }
3855

3856 3857 3858
    if (table_share->blob_fields) {
        if (fix_rec_buff_for_blob(max_row_length(record))) {
            error = HA_ERR_OUT_OF_MEM;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3859 3860 3861
            goto cleanup;
        }
    }
3862

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3863 3864
    create_dbt_key_from_table(&prim_key, primary_key, primary_key_buff, record, &has_null);
    if ((error = pack_row(&row, (const uchar *) record, primary_key))){
3865
        goto cleanup;
3866 3867
    }

3868
    create_sub_trans = (using_ignore && !(do_ignore_flag_optimization(thd,table,share->replace_into_fast)));
3869
    if (create_sub_trans) {
3870
        error = db_env->txn_begin(db_env, transaction, &sub_trans, DB_INHERIT_ISOLATION);
3871 3872 3873
        if (error) {
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3874
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3875
    
3876
    txn = create_sub_trans ? sub_trans : transaction;
3877 3878

    if (tokudb_debug & TOKUDB_DEBUG_CHECK_KEY) {
3879
        test_row_packing(record,&prim_key,&row);
3880
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3881

3882 3883 3884 3885 3886 3887
    if (loader) {
        error = loader->put(loader, &prim_key, &row);
        if (error) {
            abort_loader = true;
            goto cleanup;
        }
3888 3889
    }
    else {
3890 3891 3892 3893 3894 3895
        if (curr_num_DBs == 1) {
            error = insert_row_to_main_dictionary(record,&prim_key, &row, txn);
            if (error) { goto cleanup; }
        }
        else {
            error = do_uniqueness_checks(record, txn, thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921
            if (error) {
                // for #4633
                // if we have a duplicate key error, let's check the primary key to see
                // if there is a duplicate there. If so, set last_dup_key to the pk
                if (error == DB_KEYEXIST && !test(hidden_primary_key)) {
                    int r = share->file->getf_set(
                        share->file, 
                        txn, 
                        0, 
                        &prim_key, 
                        smart_dbt_do_nothing, 
                        NULL
                        );
                    if (r == 0) {
                        // if we get no error, that means the row
                        // was found and this is a duplicate key,
                        // so we set last_dup_key
                        last_dup_key = primary_key;
                    }
                    else if (r != DB_NOTFOUND) {
                        // if some other error is returned, return that to the user.
                        error = r;
                    }
                }
                goto cleanup; 
            }
3922

3923 3924 3925
            error = insert_rows_to_dictionaries_mult(&prim_key, &row, txn, thd);
            if (error) { goto cleanup; }
        }
3926
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3927

3928
    trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3929 3930
    if (!error) {
        added_rows++;
3931 3932
        trx->stmt_progress.inserted++;
        track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3933 3934
    }
cleanup:
3935 3936 3937
    if (!num_DBs_locked_in_bulk) {
       rw_unlock(&share->num_DBs_lock);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3938
    if (error == DB_KEYEXIST) {
3939
        error = HA_ERR_FOUND_DUPP_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3940
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3941 3942 3943 3944 3945
    if (sub_trans) {
        // no point in recording error value of abort.
        // nothing we can do about it anyway and it is not what
        // we want to return.
        if (error) {
3946
            abort_txn(sub_trans);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3947 3948
        }
        else {
3949
            commit_txn(sub_trans, DB_TXN_NOSYNC);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3950 3951
        }
    }
3952
    TOKUDB_DBUG_RETURN(error);
3953 3954 3955
}

/* Compare if a key in a row has changed */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3956 3957 3958
bool ha_tokudb::key_changed(uint keynr, const uchar * old_row, const uchar * new_row) {
    DBT old_key;
    DBT new_key;
3959 3960
    memset((void *) &old_key, 0, sizeof(old_key));
    memset((void *) &new_key, 0, sizeof(new_key));
3961

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3962 3963 3964 3965
    bool has_null;
    create_dbt_key_from_table(&new_key, keynr, key_buff2, new_row, &has_null);
    create_dbt_key_for_lookup(&old_key,&table->key_info[keynr], key_buff3, old_row, &has_null);
    return tokudb_prefix_cmp_dbt_key(share->key_file[keynr], &old_key, &new_key);
3966 3967
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3968 3969 3970 3971 3972 3973 3974 3975 3976
//
// Updates a row in the table, called when handling an UPDATE query
// Parameters:
//      [in]    old_row - row to be updated, in MySQL format
//      [in]    new_row - new row, in MySQL format
// Returns:
//      0 on success
//      error otherwise
//
3977
int ha_tokudb::update_row(const uchar * old_row, uchar * new_row) {
3978
    TOKUDB_DBUG_ENTER("update_row");
3979
    DBT prim_key, old_prim_key, prim_row, old_prim_row;
3980
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3981 3982
    bool has_null;
    THD* thd = ha_thd();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3983 3984
    DB_TXN* sub_trans = NULL;
    DB_TXN* txn = NULL;
3985
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
3986
    uint curr_num_DBs;
3987 3988

    LINT_INIT(error);
3989 3990 3991 3992
    memset((void *) &prim_key, 0, sizeof(prim_key));
    memset((void *) &old_prim_key, 0, sizeof(old_prim_key));
    memset((void *) &prim_row, 0, sizeof(prim_row));
    memset((void *) &old_prim_row, 0, sizeof(old_prim_row));
3993 3994


3995
    statistic_increment(table->in_use->status_var.ha_update_count, &LOCK_status);
3996
#if MYSQL_VERSION_ID < 50600
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3997
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_UPDATE) {
3998
        table->timestamp_field->set_time();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3999
    }
4000
#endif
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4001 4002 4003 4004 4005 4006 4007 4008 4009 4010
    //
    // check to see if some value for the auto increment column that is bigger
    // than anything else til now is being used. If so, update the metadata to reflect it
    // the goal here is we never want to have a dup key error due to a bad increment
    // of the auto inc field.
    //
    if (share->has_auto_inc && new_row == table->record[0]) {
        pthread_mutex_lock(&share->mutex);
        ulonglong curr_auto_inc = retrieve_auto_increment(
            table->field[share->ai_field_index]->key_type(), 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4011
            field_offset(table->field[share->ai_field_index], table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022
            new_row
            );
        if (curr_auto_inc > share->last_auto_increment) {
            error = update_max_auto_inc(share->status_block, curr_auto_inc);
            if (!error) {
                share->last_auto_increment = curr_auto_inc;
            }
        }
        pthread_mutex_unlock(&share->mutex);
    }

4023 4024 4025 4026 4027 4028
    //
    // grab reader lock on numDBs_lock
    //
    rw_rdlock(&share->num_DBs_lock);
    curr_num_DBs = share->num_DBs;

4029 4030 4031 4032 4033
    if (using_ignore) {
        error = db_env->txn_begin(db_env, transaction, &sub_trans, DB_INHERIT_ISOLATION);
        if (error) {
            goto cleanup;
        }
4034
    }
4035
    txn = using_ignore ? sub_trans : transaction;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4036 4037


4038
    if (hidden_primary_key) {
4039
        memset((void *) &prim_key, 0, sizeof(prim_key));
4040 4041 4042
        prim_key.data = (void *) current_ident;
        prim_key.size = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
        old_prim_key = prim_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4043 4044
    } 
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4045
        create_dbt_key_from_table(&prim_key, primary_key, key_buff, new_row, &has_null);
4046
        create_dbt_key_from_table(&old_prim_key, primary_key, primary_key_buff, old_row, &has_null);
4047 4048
    }

4049 4050 4051 4052 4053
    //
    // do uniqueness checks
    //
    if (share->has_unique_keys && !thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS)) {
        for (uint keynr = 0; keynr < table_share->keys; keynr++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4054
            bool is_unique_key = table->key_info[keynr].flags & HA_NOSAME;
4055 4056
            if (keynr == primary_key) {
                continue;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4057
            }
4058
            if (is_unique_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4059 4060
                bool key_ch = key_changed(keynr, old_row, new_row);
                if (key_ch) {
4061 4062 4063 4064 4065 4066 4067 4068
                    bool is_unique;
                    error = is_val_unique(&is_unique, new_row, &table->key_info[keynr], keynr, txn);
                    if (error) goto cleanup;
                    if (!is_unique) {
                        error = DB_KEYEXIST;
                        last_dup_key = keynr;
                        goto cleanup;
                    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4069
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4070
            }
4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081
        }
    }
    
    if (table_share->blob_fields) {
        if (fix_rec_buff_for_blob(max_row_length(new_row))) {
            error = HA_ERR_OUT_OF_MEM;
            goto cleanup;
        }
        if (fix_rec_update_buff_for_blob(max_row_length(old_row))) {
            error = HA_ERR_OUT_OF_MEM;
            goto cleanup;
4082 4083
        }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4084

4085 4086 4087 4088 4089 4090
    error = pack_row(&prim_row, new_row, primary_key);
    if (error) { goto cleanup; }

    error = pack_old_row_for_update(&old_prim_row, old_row, primary_key);
    if (error) { goto cleanup; }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4091
    set_main_dict_put_flags(thd, false, &mult_put_flags[primary_key]);
4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109

    error = db_env->update_multiple(
        db_env, 
        share->key_file[primary_key], 
        txn,
        &old_prim_key, 
        &old_prim_row,
        &prim_key, 
        &prim_row,
        curr_num_DBs, 
        share->key_file,
        mult_put_flags,
        2*curr_num_DBs, 
        mult_key_dbt,
        curr_num_DBs, 
        mult_rec_dbt
        );
    
4110 4111 4112 4113
    if (error == DB_KEYEXIST) {
        last_dup_key = primary_key;
    }    
    else if (!error) {
4114 4115
        trx->stmt_progress.updated++;
        track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4116
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4117 4118


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4119
cleanup:
4120
    rw_unlock(&share->num_DBs_lock);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4121
    if (error == DB_KEYEXIST) {
4122
        error = HA_ERR_FOUND_DUPP_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4123
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4124 4125 4126 4127 4128
    if (sub_trans) {
        // no point in recording error value of abort.
        // nothing we can do about it anyway and it is not what
        // we want to return.
        if (error) {
4129
            abort_txn(sub_trans);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4130 4131
        }
        else {
4132
            commit_txn(sub_trans, DB_TXN_NOSYNC);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4133 4134
        }
    }
4135
    TOKUDB_DBUG_RETURN(error);
4136 4137
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4138
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4139
// Deletes a row in the table, called when handling a DELETE query
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4140 4141 4142 4143 4144 4145
// Parameters:
//      [in]    record - row to be deleted, in MySQL format
// Returns:
//      0 on success
//      error otherwise
//
4146
int ha_tokudb::delete_row(const uchar * record) {
4147
    TOKUDB_DBUG_ENTER("ha_tokudb::delete_row");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4148
    int error = ENOSYS;
4149
    DBT row, prim_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4150
    bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4151
    THD* thd = ha_thd();
4152
    uint curr_num_DBs;
4153
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4154

4155 4156
    statistic_increment(table->in_use->status_var.ha_delete_count, &LOCK_status);

4157 4158 4159 4160 4161 4162
    //
    // grab reader lock on numDBs_lock
    //
    rw_rdlock(&share->num_DBs_lock);
    curr_num_DBs = share->num_DBs;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4163
    create_dbt_key_from_table(&prim_key, primary_key, key_buff, record, &has_null);
4164 4165 4166 4167 4168
    if (table_share->blob_fields) {
        if (fix_rec_buff_for_blob(max_row_length(record))) {
            error = HA_ERR_OUT_OF_MEM;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4169
    }
4170 4171 4172
    if ((error = pack_row(&row, (const uchar *) record, primary_key))){
        goto cleanup;
    }
4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184

    error = db_env->del_multiple(
        db_env, 
        share->key_file[primary_key], 
        transaction, 
        &prim_key, 
        &row,
        curr_num_DBs, 
        share->key_file, 
        mult_key_dbt,
        mult_del_flags
        );
4185

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4186 4187
    if (error) {
        DBUG_PRINT("error", ("Got error %d", error));
4188
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4189
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4190
        deleted_rows++;
4191 4192
        trx->stmt_progress.deleted++;
        track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4193
    }
4194
cleanup:
4195
    rw_unlock(&share->num_DBs_lock);
4196
    TOKUDB_DBUG_RETURN(error);
4197 4198
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4199 4200 4201 4202 4203 4204 4205
//
// takes as input table->read_set and table->write_set
// and puts list of field indexes that need to be read in
// unpack_row in the member variables fixed_cols_for_query
// and var_cols_for_query
//
void ha_tokudb::set_query_columns(uint keynr) {
Yoni Fogel's avatar
Yoni Fogel committed
4206 4207
    uint32_t curr_fixed_col_index = 0;
    uint32_t curr_var_col_index = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4208 4209 4210 4211 4212 4213 4214
    read_key = false;
    read_blobs = false;
    //
    // i know this is probably confusing and will need to be explained better
    //
    uint key_index = 0;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4215
    if (keynr == primary_key || keynr == MAX_KEY) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4216 4217 4218 4219 4220 4221 4222 4223 4224 4225
        key_index = primary_key;
    }
    else {
        key_index = (table->key_info[keynr].flags & HA_CLUSTERING ? keynr : primary_key);
    }
    for (uint i = 0; i < table_share->fields; i++) {
        if (bitmap_is_set(table->read_set,i) || 
            bitmap_is_set(table->write_set,i)
            ) 
        {
4226
            if (bitmap_is_set(&share->kc_info.key_filters[key_index],i)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4227 4228 4229 4230 4231 4232
                read_key = true;
            }
            else {
                //
                // if fixed field length
                //
4233
                if (share->kc_info.field_lengths[i] != 0) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4234 4235 4236 4237 4238 4239 4240 4241 4242
                    //
                    // save the offset into the list
                    //
                    fixed_cols_for_query[curr_fixed_col_index] = i;
                    curr_fixed_col_index++;
                }
                //
                // varchar or varbinary
                //
4243
                else if (share->kc_info.length_bytes[i] != 0) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259
                    var_cols_for_query[curr_var_col_index] = i;
                    curr_var_col_index++;
                }
                //
                // it is a blob
                //
                else {
                    read_blobs = true;
                }
            }
        }
    }
    num_fixed_cols_for_query = curr_fixed_col_index;
    num_var_cols_for_query = curr_var_col_index;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4260
void ha_tokudb::column_bitmaps_signal() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4261 4262 4263
    //
    // if we have max number of indexes, then MAX_KEY == primary_key
    //
4264 4265
    if (tokudb_active_index != MAX_KEY || tokudb_active_index == primary_key) {
        set_query_columns(tokudb_active_index);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4266 4267
    }
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4268

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4269 4270 4271 4272 4273 4274 4275 4276
//
// Notification that a scan of entire secondary table is about
// to take place. Will pre acquire table read lock
// Returns:
//      0 on success
//      error otherwise
//
int ha_tokudb::prepare_index_scan() {
4277
    int error = 0;
4278
    HANDLE_INVALID_CURSOR();
4279
    error = prelock_range(NULL, NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4280 4281 4282 4283 4284 4285 4286 4287 4288
    if (error) { last_cursor_error = error; goto cleanup; }

    range_lock_grabbed = true;
    error = 0;
cleanup:
    return error;
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4289 4290 4291 4292 4293 4294 4295
//
// Notification that a range query getting all elements that equal a key
//  to take place. Will pre acquire read lock
// Returns:
//      0 on success
//      error otherwise
//
4296
int ha_tokudb::prepare_index_key_scan(const uchar * key, uint key_len) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4297 4298
    int error = 0;
    DBT start_key, end_key;
4299
    THD* thd = ha_thd();
4300
    HANDLE_INVALID_CURSOR();
4301
    pack_key(&start_key, tokudb_active_index, prelocked_left_range, key, key_len, COL_NEG_INF);
4302
    prelocked_left_range_size = start_key.size;
4303
    pack_key(&end_key, tokudb_active_index, prelocked_right_range, key, key_len, COL_POS_INF);
4304
    prelocked_right_range_size = end_key.size;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4305

4306 4307 4308 4309 4310 4311
    error = cursor->c_pre_acquire_range_lock(
        cursor, 
        &start_key, 
        &end_key 
        );

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4312 4313 4314 4315 4316
    if (error){ 
        goto cleanup; 
    }

    range_lock_grabbed = true;
4317
    doing_bulk_fetch = (thd_sql_command(thd) == SQLCOM_SELECT);
4318 4319
    bulk_fetch_iteration = 0;
    rows_fetched_using_bulk_fetch = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4320 4321 4322
    error = 0;
cleanup:
    if (error) {
4323 4324 4325
        if (error == DB_LOCK_NOTGRANTED) {
            error = HA_ERR_LOCK_WAIT_TIMEOUT;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4326 4327
        last_cursor_error = error;
        //
4328 4329
        // cursor should be initialized here, but in case it is not, 
        // we still check
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4330 4331
        //
        if (cursor) {
4332 4333
            int r = cursor->c_close(cursor);
            assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4334 4335 4336 4337 4338 4339
            cursor = NULL;
        }
    }
    return error;
}

4340 4341 4342 4343
void ha_tokudb::invalidate_bulk_fetch() {
    bytes_used_in_range_query_buff= 0;
    curr_range_query_buff_offset = 0;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4344

4345 4346
volatile int ha_tokudb_index_init_wait = 0; // debug

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4347 4348 4349 4350 4351 4352 4353 4354 4355
//
// Initializes local cursor on DB with index keynr
// Parameters:
//          keynr - key (index) number
//          sorted - 1 if result MUST be sorted according to index
// Returns:
//      0 on success
//      error otherwise
//
4356
int ha_tokudb::index_init(uint keynr, bool sorted) {
4357
    TOKUDB_DBUG_ENTER("ha_tokudb::index_init %p %d", this, keynr);
4358 4359
    while (ha_tokudb_index_init_wait) sleep(1); // debug

4360
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4361
    THD* thd = ha_thd(); 
4362 4363 4364 4365 4366 4367 4368 4369
    DBUG_PRINT("enter", ("table: '%s'  key: %d", table_share->table_name.str, keynr));

    /*
       Under some very rare conditions (like full joins) we may already have
       an active cursor at this point
     */
    if (cursor) {
        DBUG_PRINT("note", ("Closing active cursor"));
4370 4371
        int r = cursor->c_close(cursor);
        assert(r==0);
4372 4373
    }
    active_index = keynr;
4374 4375 4376 4377 4378 4379 4380 4381 4382

    if (active_index < MAX_KEY) {
        DBUG_ASSERT(keynr <= table->s->keys);
    } else {
        DBUG_ASSERT(active_index == MAX_KEY);
        keynr = primary_key;
    }
    tokudb_active_index = keynr;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4383
    last_cursor_error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4384
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4385
    DBUG_ASSERT(share->key_file[keynr]);
4386
    cursor_flags = get_cursor_isolation_flags(lock.type, thd);
4387
    if (use_write_locks) {
4388
        cursor_flags |= DB_RMW;
4389 4390 4391 4392
    }
    if (get_disable_prefetching(thd)) {
        cursor_flags |= DBC_DISABLE_PREFETCHING;
    }
4393 4394
    if ((error = share->key_file[keynr]->cursor(share->key_file[keynr], transaction, &cursor, cursor_flags))) {
        if (error == TOKUDB_MVCC_DICTIONARY_TOO_NEW) {
4395
            error = HA_ERR_TABLE_DEF_CHANGED;
4396 4397
            my_error(ER_TABLE_DEF_CHANGED, MYF(0));
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4398 4399 4400 4401
        if (error == DB_LOCK_NOTGRANTED) {
            error = HA_ERR_LOCK_WAIT_TIMEOUT;
            my_error(ER_LOCK_WAIT_TIMEOUT, MYF(0));
        }
4402
        table->status = STATUS_NOT_FOUND;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4403
        last_cursor_error = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4404
        cursor = NULL;             // Safety
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4405
        goto exit;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4406
    }
4407
    memset((void *) &last_key, 0, sizeof(last_key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4408 4409 4410 4411 4412 4413 4414 4415

    if (thd_sql_command(thd) == SQLCOM_SELECT) {
        set_query_columns(keynr);
        unpack_entire_row = false;
    }
    else {
        unpack_entire_row = true;
    }
4416 4417
    invalidate_bulk_fetch();
    doing_bulk_fetch = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4418 4419
    error = 0;
exit:
4420
    TOKUDB_DBUG_RETURN(error);
4421 4422
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4423 4424 4425
//
// closes the local cursor
//
4426
int ha_tokudb::index_end() {
4427
    TOKUDB_DBUG_ENTER("ha_tokudb::index_end %p", this);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4428
    range_lock_grabbed = false;
4429 4430
    if (cursor) {
        DBUG_PRINT("enter", ("table: '%s'", table_share->table_name.str));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4431 4432
        int r = cursor->c_close(cursor);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4433
        cursor = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4434
        last_cursor_error = 0;
4435
    }
4436
    active_index = tokudb_active_index = MAX_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4437 4438 4439 4440 4441 4442 4443 4444 4445

    //
    // reset query variables
    //
    unpack_entire_row = true;
    read_blobs = true;
    read_key = true;
    num_fixed_cols_for_query = 0;
    num_var_cols_for_query = 0;
4446 4447 4448 4449

    invalidate_bulk_fetch();
    doing_bulk_fetch = false;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4450
    TOKUDB_DBUG_RETURN(0);
4451 4452
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4453 4454 4455

int ha_tokudb::handle_cursor_error(int error, int err_to_return, uint keynr) {
    TOKUDB_DBUG_ENTER("ha_tokudb::handle_cursor_error");
4456
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4457 4458 4459
        if (error == DB_LOCK_NOTGRANTED) {
            error = HA_ERR_LOCK_WAIT_TIMEOUT;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4460
        last_cursor_error = error;
4461
        table->status = STATUS_NOT_FOUND;
4462 4463
        int r = cursor->c_close(cursor);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4464
        cursor = NULL;
Barry Perlman's avatar
Barry Perlman committed
4465
        if (error == DB_NOTFOUND) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4466
            error = err_to_return;
4467
            if ((share->key_file[keynr]->cursor(share->key_file[keynr], transaction, &cursor, cursor_flags))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4468 4469 4470
                cursor = NULL;             // Safety
            }
        }
4471
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4472 4473 4474 4475 4476 4477 4478 4479 4480 4481
    TOKUDB_DBUG_RETURN(error);
}


//
// Helper function for read_row and smart_dbt_callback_xxx functions
// When using a hidden primary key, upon reading a row, 
// we set the current_ident field to whatever the primary key we retrieved
// was
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4482
void ha_tokudb::extract_hidden_primary_key(uint keynr, DBT const *found_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4483
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4484
    // extract hidden primary key to current_ident
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4485 4486
    //
    if (hidden_primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4487 4488 4489
        if (keynr == primary_key) {
            memcpy_fixed(current_ident, (char *) found_key->data, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4490
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4491
        // if secondary key, hidden primary key is at end of found_key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4492
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4493
        else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4494 4495 4496 4497 4498 4499
            memcpy_fixed(
                current_ident, 
                (char *) found_key->data + found_key->size - TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH, 
                TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH
                );
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4500
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4501 4502
}

4503

4504
int ha_tokudb::read_row_callback (uchar * buf, uint keynr, DBT const *row, DBT const *found_key) {
4505
    assert(keynr == primary_key);
4506
    return unpack_row(buf, row,found_key, keynr);
4507 4508
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4509 4510 4511 4512 4513 4514 4515 4516 4517 4518
//
// Reads the contents of row and found_key, DBT's retrieved from the DB associated to keynr, into buf
// This function assumes that we are using a covering index, as a result, if keynr is the primary key,
// we do not read row into buf
// Parameters:
//      [out]   buf - buffer for the row, in MySQL format
//              keynr - index into key_file that represents DB we are currently operating on.
//      [in]    row - the row that has been read from the preceding DB call
//      [in]    found_key - key used to retrieve the row
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4519
void ha_tokudb::read_key_only(uchar * buf, uint keynr, DBT const *found_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4520 4521
    TOKUDB_DBUG_ENTER("ha_tokudb::read_key_only");
    table->status = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4522 4523 4524 4525
    //
    // only case when we do not unpack the key is if we are dealing with the main dictionary
    // of a table with a hidden primary key
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4526 4527 4528
    if (!(hidden_primary_key && keynr == primary_key)) {
        unpack_key(buf, found_key, keynr);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541
    DBUG_VOID_RETURN;
}

//
// Helper function used to try to retrieve the entire row
// If keynr is associated with the main table, reads contents of found_key and row into buf, otherwise,
// makes copy of primary key and saves it to last_key. This can later be used to retrieve the entire row
// Parameters:
//      [out]   buf - buffer for the row, in MySQL format
//              keynr - index into key_file that represents DB we are currently operating on.
//      [in]    row - the row that has been read from the preceding DB call
//      [in]    found_key - key used to retrieve the row
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4542
int ha_tokudb::read_primary_key(uchar * buf, uint keynr, DBT const *row, DBT const *found_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4543
    TOKUDB_DBUG_ENTER("ha_tokudb::read_primary_key");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4544
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4545
    table->status = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4546 4547 4548 4549
    //
    // case where we read from secondary table that is not clustered
    //
    if (keynr != primary_key && !(table->key_info[keynr].flags & HA_CLUSTERING)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4550
        bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4551
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4552 4553
        // create a DBT that has the same data as row, this is inefficient
        // extract_hidden_primary_key MUST have been called before this
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4554
        //
4555
        memset((void *) &last_key, 0, sizeof(last_key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4556 4557 4558 4559 4560 4561 4562 4563 4564 4565
        if (!hidden_primary_key) {
            unpack_key(buf, found_key, keynr);
        }
        create_dbt_key_from_table(
            &last_key, 
            primary_key,
            key_buff,
            buf,
            &has_null
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4566
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4567
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4568
    // else read from clustered/primary key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4569
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4570
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4571 4572
        error = unpack_row(buf, row, found_key, keynr);
        if (error) { goto exit; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4573 4574
    }
    if (found_key) { DBUG_DUMP("read row key", (uchar *) found_key->data, found_key->size); }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4575 4576 4577
    error = 0;
exit:
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589
}

//
// This function reads an entire row into buf. This function also assumes that
// the key needed to retrieve the row is stored in the member variable last_key
// Parameters:
//      [out]   buf - buffer for the row, in MySQL format
// Returns:
//      0 on success, error otherwise
//
int ha_tokudb::read_full_row(uchar * buf) {
    TOKUDB_DBUG_ENTER("ha_tokudb::read_full_row");
4590
    int error = 0;
4591 4592 4593 4594
    struct smart_dbt_info info;
    info.ha = this;
    info.buf = buf;
    info.keynr = primary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4595
    //
4596
    // assumes key is stored in this->last_key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4597
    //
4598 4599 4600 4601 4602 4603 4604 4605 4606 4607

    error = share->file->getf_set(
        share->file, 
        transaction, 
        cursor_flags, 
        &last_key, 
        smart_dbt_callback_rowread_ptquery, 
        &info
        );

4608
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4609 4610 4611
        if (error == DB_LOCK_NOTGRANTED) {
            error = HA_ERR_LOCK_WAIT_TIMEOUT;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4612 4613 4614 4615
        table->status = STATUS_NOT_FOUND;
        TOKUDB_DBUG_RETURN(error == DB_NOTFOUND ? HA_ERR_CRASHED : error);
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4616
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4617 4618 4619
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631
// 
// Reads the next row matching to the key, on success, advances cursor 
// Parameters: 
//      [out]   buf - buffer for the next row, in MySQL format 
//      [in]     key - key value 
//                keylen - length of key 
// Returns: 
//      0 on success 
//      HA_ERR_END_OF_FILE if not found 
//      error otherwise 
// 
int ha_tokudb::index_next_same(uchar * buf, const uchar * key, uint keylen) { 
4632 4633 4634
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next_same");
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4635 4636 4637 4638
    DBT curr_key;
    DBT found_key;
    bool has_null;
    int cmp;
4639
    int error = get_next(buf, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4640 4641 4642 4643 4644 4645
    if (error) {
        goto cleanup;
    }
    //
    // now do the comparison
    //
4646 4647 4648
    pack_key(&curr_key, tokudb_active_index, key_buff2, key, keylen, COL_ZERO);
    create_dbt_key_from_table(&found_key,tokudb_active_index,key_buff3,buf,&has_null);
    cmp = tokudb_prefix_cmp_dbt_key(share->key_file[tokudb_active_index], &curr_key, &found_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4649 4650 4651
    if (cmp) {
        error = HA_ERR_END_OF_FILE; 
    }
4652 4653 4654 4655

cleanup: 
    TOKUDB_DBUG_RETURN(error);
} 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4656 4657


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674
//
// According to InnoDB handlerton: Positions an index cursor to the index 
// specified in keynr. Fetches the row if any
// Parameters:
//      [out]       buf - buffer for the  returned row
//      [in]         key - key value, according to InnoDB, if NULL, 
//                              position cursor at start or end of index,
//                              not sure if this is done now
//                    key_len - length of key
//                    find_flag - according to InnoDB, search flags from my_base.h
// Returns:
//      0 on success
//      HA_ERR_KEY_NOT_FOUND if not found (per InnoDB), 
//          we seem to return HA_ERR_END_OF_FILE if find_flag != HA_READ_KEY_EXACT
//          TODO: investigate this for correctness
//      error otherwise
//
4675 4676
int ha_tokudb::index_read(uchar * buf, const uchar * key, uint key_len, enum ha_rkey_function find_flag) {
    TOKUDB_DBUG_ENTER("ha_tokudb::index_read %p find %d", this, find_flag);
4677
    invalidate_bulk_fetch();
4678
    // TOKUDB_DBUG_DUMP("key=", key, key_len);
4679
    DBT row;
4680
    DBT lookup_key;
4681
    int error = 0;    
Yoni Fogel's avatar
Yoni Fogel committed
4682
    uint32_t flags = 0;
4683 4684
    THD* thd = ha_thd();
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4685
    struct smart_dbt_info info;
4686
    struct index_read_info ir_info;
4687

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4688
    HANDLE_INVALID_CURSOR();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4689

4690
    table->in_use->status_var.ha_read_key_count++;
4691
    memset((void *) &row, 0, sizeof(row));
4692

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4693 4694
    info.ha = this;
    info.buf = buf;
4695
    info.keynr = tokudb_active_index;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4696

4697
    ir_info.smart_dbt_info = info;
4698
    ir_info.cmp = 0;
4699

4700
    flags = SET_PRELOCK_FLAG(0);
4701
    switch (find_flag) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4702
    case HA_READ_KEY_EXACT: /* Find first record else error */
4703
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_NEG_INF);
4704
        ir_info.orig_key = &lookup_key;
4705 4706 4707

        error = cursor->c_getf_set_range(cursor, flags,
                &lookup_key, SMART_DBT_IR_CALLBACK, &ir_info);
4708 4709
        if (ir_info.cmp) {
            error = DB_NOTFOUND;
4710
        }
4711
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4712
    case HA_READ_AFTER_KEY: /* Find next rec. after key-record */
4713
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_POS_INF);
4714 4715
        error = cursor->c_getf_set_range(cursor, flags,
                &lookup_key, SMART_DBT_CALLBACK, &info);
4716
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4717
    case HA_READ_BEFORE_KEY: /* Find next rec. before key-record */
4718
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_NEG_INF);
4719 4720
        error = cursor->c_getf_set_range_reverse(cursor, flags, 
                &lookup_key, SMART_DBT_CALLBACK, &info);
4721
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4722
    case HA_READ_KEY_OR_NEXT: /* Record or next record */
4723
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_NEG_INF);
4724 4725
        error = cursor->c_getf_set_range(cursor, flags,
                &lookup_key, SMART_DBT_CALLBACK, &info);
4726
        break;
4727 4728 4729
    //
    // This case does not seem to ever be used, it is ok for it to be slow
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4730
    case HA_READ_KEY_OR_PREV: /* Record or previous */
4731
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_NEG_INF);
4732
        ir_info.orig_key = &lookup_key;
4733 4734
        error = cursor->c_getf_set_range(cursor, flags,
                &lookup_key, SMART_DBT_IR_CALLBACK, &ir_info);
4735
        if (error == DB_NOTFOUND) {
4736 4737
            error = cursor->c_getf_last(cursor, flags, SMART_DBT_CALLBACK, &info);
        }
4738 4739 4740
        else if (ir_info.cmp) {
            error = cursor->c_getf_prev(cursor, flags, SMART_DBT_CALLBACK, &info);
        }
4741
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4742
    case HA_READ_PREFIX_LAST_OR_PREV: /* Last or prev key with the same prefix */
4743
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_POS_INF);
4744 4745
        error = cursor->c_getf_set_range_reverse(cursor, flags, 
                    &lookup_key, SMART_DBT_CALLBACK, &info);
4746
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4747
    case HA_READ_PREFIX_LAST:
4748
        pack_key(&lookup_key, tokudb_active_index, key_buff3, key, key_len, COL_POS_INF);
4749
        ir_info.orig_key = &lookup_key;
4750
        error = cursor->c_getf_set_range_reverse(cursor, flags, &lookup_key, SMART_DBT_IR_CALLBACK, &ir_info);
4751 4752
        if (ir_info.cmp) {
            error = DB_NOTFOUND;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4753 4754
        }
        break;
4755
    default:
4756
        TOKUDB_TRACE("unsupported:%d\n", find_flag);
4757
        error = HA_ERR_UNSUPPORTED;
4758
        break;
4759
    }
4760 4761
    error = handle_cursor_error(error,HA_ERR_KEY_NOT_FOUND,tokudb_active_index);
    if (!error && !key_read && tokudb_active_index != primary_key && !(table->key_info[tokudb_active_index].flags & HA_CLUSTERING)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4762
        error = read_full_row(buf);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4763
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4764 4765 4766 4767
    
    if (error && (tokudb_debug & TOKUDB_DEBUG_ERROR)) {
        TOKUDB_TRACE("error:%d:%d\n", error, find_flag);
    }
4768 4769 4770
    trx->stmt_progress.queried++;
    track_progress(thd);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4771
cleanup:
4772
    TOKUDB_DBUG_RETURN(error);
4773 4774
}

4775

4776 4777 4778 4779 4780
int ha_tokudb::read_data_from_range_query_buff(uchar* buf, bool need_val) {
    // buffer has the next row, get it from there
    int error;
    uchar* curr_pos = range_query_buff+curr_range_query_buff_offset;
    DBT curr_key;
4781
    memset((void *) &curr_key, 0, sizeof(curr_key));
4782 4783
    
    // get key info
Yoni Fogel's avatar
Yoni Fogel committed
4784
    uint32_t key_size = *(uint32_t *)curr_pos;
4785 4786 4787 4788 4789 4790 4791 4792 4793 4794
    curr_pos += sizeof(key_size);
    uchar* curr_key_buff = curr_pos;
    curr_pos += key_size;
    
    curr_key.data = curr_key_buff;
    curr_key.size = key_size;
    
    // if this is a covering index, this is all we need
    if (this->key_read) {
        assert(!need_val);
4795 4796
        extract_hidden_primary_key(tokudb_active_index, &curr_key);
        read_key_only(buf, tokudb_active_index, &curr_key);
4797 4798 4799 4800 4801
        error = 0;
    }
    // we need to get more data
    else {
        DBT curr_val;
4802
        memset((void *) &curr_val, 0, sizeof(curr_val));
4803
        uchar* curr_val_buff = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
4804
        uint32_t val_size = 0;
4805 4806 4807 4808
        // in this case, we don't have a val, we are simply extracting the pk
        if (!need_val) {
            curr_val.data = curr_val_buff;
            curr_val.size = val_size;
4809 4810
            extract_hidden_primary_key(tokudb_active_index, &curr_key);
            error = read_primary_key( buf, tokudb_active_index, &curr_val, &curr_key);
4811 4812
        }
        else {
4813
            extract_hidden_primary_key(tokudb_active_index, &curr_key);
4814 4815 4816
            // need to extract a val and place it into buf
            if (unpack_entire_row) {
                // get val info
Yoni Fogel's avatar
Yoni Fogel committed
4817
                val_size = *(uint32_t *)curr_pos;
4818 4819 4820 4821 4822
                curr_pos += sizeof(val_size);
                curr_val_buff = curr_pos;
                curr_pos += val_size;
                curr_val.data = curr_val_buff;
                curr_val.size = val_size;
4823
                error = unpack_row(buf,&curr_val, &curr_key, tokudb_active_index);
4824 4825
            }
            else {
4826 4827
                if (!(hidden_primary_key && tokudb_active_index == primary_key)) {
                    unpack_key(buf,&curr_key,tokudb_active_index);
4828 4829 4830 4831 4832 4833 4834 4835
                }
                // read rows we care about

                // first the null bytes;
                memcpy(buf, curr_pos, table_share->null_bytes);
                curr_pos += table_share->null_bytes;

                // now the fixed sized rows                
Yoni Fogel's avatar
Yoni Fogel committed
4836
                for (uint32_t i = 0; i < num_fixed_cols_for_query; i++) {
4837 4838 4839 4840 4841 4842 4843 4844 4845 4846
                    uint field_index = fixed_cols_for_query[i];
                    Field* field = table->field[field_index];
                    unpack_fixed_field(
                        buf + field_offset(field, table),
                        curr_pos,
                        share->kc_info.field_lengths[field_index]
                        );
                    curr_pos += share->kc_info.field_lengths[field_index];
                }
                // now the variable sized rows
Yoni Fogel's avatar
Yoni Fogel committed
4847
                for (uint32_t i = 0; i < num_var_cols_for_query; i++) {
4848 4849
                    uint field_index = var_cols_for_query[i];
                    Field* field = table->field[field_index];
Yoni Fogel's avatar
Yoni Fogel committed
4850
                    uint32_t field_len = *(uint32_t *)curr_pos;
4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861
                    curr_pos += sizeof(field_len);
                    unpack_var_field(
                        buf + field_offset(field, table),
                        curr_pos,
                        field_len,
                        share->kc_info.length_bytes[field_index]
                        );
                    curr_pos += field_len;
                }
                // now the blobs
                if (read_blobs) {
Yoni Fogel's avatar
Yoni Fogel committed
4862
                    uint32_t blob_size = *(uint32_t *)curr_pos;
4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884
                    curr_pos += sizeof(blob_size);
                    error = unpack_blobs(
                        buf,
                        curr_pos,
                        blob_size,
                        true
                        );
                    curr_pos += blob_size;
                    if (error) {
                        invalidate_bulk_fetch();
                        goto exit;
                    }
                }
                error = 0;
            }
        }
    }
    
    curr_range_query_buff_offset = curr_pos - range_query_buff;
exit:
    return error;
}
4885

4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903
static int
smart_dbt_bf_callback(DBT const *key, DBT  const *row, void *context) {
    SMART_DBT_BF_INFO info = (SMART_DBT_BF_INFO)context;
    return info->ha->fill_range_query_buf(info->need_val, key, row, info->direction, info->thd);
}

// fill in the range query buf for bulk fetch
int ha_tokudb::fill_range_query_buf(
    bool need_val, 
    DBT const *key, 
    DBT  const *row, 
    int direction,
    THD* thd
    ) {
    int error;
    //
    // first put the value into range_query_buf
    //
Yoni Fogel's avatar
Yoni Fogel committed
4904 4905 4906
    uint32_t size_remaining = size_range_query_buff - bytes_used_in_range_query_buff;
    uint32_t size_needed;
    uint32_t user_defined_size = get_tokudb_read_buf_size(thd);
4907 4908 4909
    uchar* curr_pos = NULL;
    if (need_val) {
        if (unpack_entire_row) {
Yoni Fogel's avatar
Yoni Fogel committed
4910
            size_needed = 2*sizeof(uint32_t) + key->size + row->size;
4911 4912 4913
        }
        else {
            // this is an upper bound
Yoni Fogel's avatar
Yoni Fogel committed
4914
            size_needed = sizeof(uint32_t) + // size of key length
4915
                          key->size + row->size + //key and row
Yoni Fogel's avatar
Yoni Fogel committed
4916 4917
                          num_var_cols_for_query*(sizeof(uint32_t)) + //lengths of varchars stored
                          sizeof(uint32_t); //length of blobs
4918 4919 4920
        }
    }
    else {
Yoni Fogel's avatar
Yoni Fogel committed
4921
        size_needed = sizeof(uint32_t) + key->size;
4922 4923 4924 4925 4926 4927 4928 4929 4930
    }
    if (size_remaining < size_needed) {
        range_query_buff = (uchar *)my_realloc(
            (void *)range_query_buff, 
            bytes_used_in_range_query_buff+size_needed, 
            MYF(MY_WME)
            );
        if (range_query_buff == NULL) {
            error = ENOMEM;
4931
            invalidate_bulk_fetch();
4932 4933
            goto cleanup;
        }
4934
        size_range_query_buff = bytes_used_in_range_query_buff+size_needed;
4935 4936 4937 4938 4939 4940
    }
    //
    // now we know we have the size, let's fill the buffer, starting with the key
    //
    curr_pos = range_query_buff + bytes_used_in_range_query_buff;

Yoni Fogel's avatar
Yoni Fogel committed
4941 4942
    *(uint32_t *)curr_pos = key->size;
    curr_pos += sizeof(uint32_t);
4943 4944 4945 4946
    memcpy(curr_pos, key->data, key->size);
    curr_pos += key->size;
    if (need_val) {
        if (unpack_entire_row) {
Yoni Fogel's avatar
Yoni Fogel committed
4947 4948
            *(uint32_t *)curr_pos = row->size;
            curr_pos += sizeof(uint32_t);
4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959
            memcpy(curr_pos, row->data, row->size);
            curr_pos += row->size;
        }
        else {
            // need to unpack just the data we care about
            const uchar* fixed_field_ptr = (const uchar *) row->data;
            fixed_field_ptr += table_share->null_bytes;

            const uchar* var_field_offset_ptr = NULL;
            const uchar* var_field_data_ptr = NULL;
            
4960 4961
            var_field_offset_ptr = fixed_field_ptr + share->kc_info.mcp_info[tokudb_active_index].fixed_field_size;
            var_field_data_ptr = var_field_offset_ptr + share->kc_info.mcp_info[tokudb_active_index].len_of_offsets;
4962 4963 4964 4965 4966 4967 4968 4969

            // first the null bytes
            memcpy(curr_pos, row->data, table_share->null_bytes);
            curr_pos += table_share->null_bytes;
            // now the fixed fields
            //
            // first the fixed fields
            //
Yoni Fogel's avatar
Yoni Fogel committed
4970
            for (uint32_t i = 0; i < num_fixed_cols_for_query; i++) {
4971 4972 4973
                uint field_index = fixed_cols_for_query[i];
                memcpy(
                    curr_pos, 
4974
                    fixed_field_ptr + share->kc_info.cp_info[tokudb_active_index][field_index].col_pack_val,
4975 4976 4977 4978 4979 4980 4981 4982
                    share->kc_info.field_lengths[field_index]
                    );
                curr_pos += share->kc_info.field_lengths[field_index];
            }
            
            //
            // now the var fields
            //
Yoni Fogel's avatar
Yoni Fogel committed
4983
            for (uint32_t i = 0; i < num_var_cols_for_query; i++) {
4984
                uint field_index = var_cols_for_query[i];
4985
                uint32_t var_field_index = share->kc_info.cp_info[tokudb_active_index][field_index].col_pack_val;
Yoni Fogel's avatar
Yoni Fogel committed
4986 4987
                uint32_t data_start_offset;
                uint32_t field_len;
4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002
                
                get_var_field_info(
                    &field_len, 
                    &data_start_offset, 
                    var_field_index, 
                    var_field_offset_ptr, 
                    share->kc_info.num_offset_bytes
                    );
                memcpy(curr_pos, &field_len, sizeof(field_len));
                curr_pos += sizeof(field_len);
                memcpy(curr_pos, var_field_data_ptr + data_start_offset, field_len);
                curr_pos += field_len;
            }
            
            if (read_blobs) {
Yoni Fogel's avatar
Yoni Fogel committed
5003 5004
                uint32_t blob_offset = 0;
                uint32_t data_size = 0;
5005 5006 5007 5008 5009
                //
                // now the blobs
                //
                get_blob_field_info(
                    &blob_offset, 
5010
                    share->kc_info.mcp_info[tokudb_active_index].len_of_offsets,
5011 5012 5013
                    var_field_data_ptr, 
                    share->kc_info.num_offset_bytes
                    );
Yoni Fogel's avatar
Yoni Fogel committed
5014
                data_size = row->size - blob_offset - (uint32_t)(var_field_data_ptr - (const uchar *)row->data);
5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030
                memcpy(curr_pos, &data_size, sizeof(data_size));
                curr_pos += sizeof(data_size);
                memcpy(curr_pos, var_field_data_ptr + blob_offset, data_size);
                curr_pos += data_size;
            }
        }
    }

    bytes_used_in_range_query_buff = curr_pos - range_query_buff;
    assert(bytes_used_in_range_query_buff <= size_range_query_buff);

    //
    // now determine if we should continue with the bulk fetch
    // we want to stop under these conditions:
    //  - we overran the prelocked range
    //  - we are close to the end of the buffer
5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046
    //  - we have fetched an exponential amount of rows with
    //  respect to the bulk fetch iteration, which is initialized 
    //  to 0 in index_init() and prelock_range().

    rows_fetched_using_bulk_fetch++;
    // if the iteration is less than the number of possible shifts on
    // a 64 bit integer, check that we haven't exceeded this iterations
    // row fetch upper bound.
    if (bulk_fetch_iteration < HA_TOKU_BULK_FETCH_ITERATION_MAX) {
        uint64_t row_fetch_upper_bound = 1LLU << bulk_fetch_iteration;
        assert(row_fetch_upper_bound > 0);
        if (rows_fetched_using_bulk_fetch >= row_fetch_upper_bound) { 
            error = 0;
            goto cleanup;
        }
    }
5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059

    if (bytes_used_in_range_query_buff + table_share->rec_buff_length > user_defined_size) {
        error = 0;
        goto cleanup;
    }
    if (direction > 0) {
        // compare what we got to the right endpoint of prelocked range
        // because we are searching keys in ascending order
        if (prelocked_right_range_size == 0) {
            error = TOKUDB_CURSOR_CONTINUE;
            goto cleanup;
        }
        DBT right_range;
5060
        memset(&right_range, 0, sizeof(right_range));
5061 5062 5063
        right_range.size = prelocked_right_range_size;
        right_range.data = prelocked_right_range;
        int cmp = tokudb_cmp_dbt_key(
5064
            share->key_file[tokudb_active_index], 
5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077
            key, 
            &right_range
            );
        error = (cmp > 0) ? 0 : TOKUDB_CURSOR_CONTINUE;
    }
    else {
        // compare what we got to the left endpoint of prelocked range
        // because we are searching keys in descending order
        if (prelocked_left_range_size == 0) {
            error = TOKUDB_CURSOR_CONTINUE;
            goto cleanup;
        }
        DBT left_range;
5078
        memset(&left_range, 0, sizeof(left_range));
5079 5080 5081
        left_range.size = prelocked_left_range_size;
        left_range.data = prelocked_left_range;
        int cmp = tokudb_cmp_dbt_key(
5082
            share->key_file[tokudb_active_index], 
5083 5084 5085 5086 5087 5088 5089 5090 5091 5092
            key, 
            &left_range
            );
        error = (cmp < 0) ? 0 : TOKUDB_CURSOR_CONTINUE;
    }
cleanup:
    return error;
}

int ha_tokudb::get_next(uchar* buf, int direction) {
5093
    int error = 0; 
Yoni Fogel's avatar
Yoni Fogel committed
5094
    uint32_t flags = SET_PRELOCK_FLAG(0);
5095 5096
    THD* thd = ha_thd();
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);;
5097
    bool need_val;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5098 5099
    HANDLE_INVALID_CURSOR();

5100 5101 5102 5103
    // we need to read the val of what we retrieve if
    // we do NOT have a covering index AND we are using a clustering secondary
    // key
    need_val = (this->key_read == 0) && 
5104 5105
                (tokudb_active_index == primary_key || 
                 table->key_info[tokudb_active_index].flags & HA_CLUSTERING
5106
                       );
5107

5108 5109
    if ((bytes_used_in_range_query_buff - curr_range_query_buff_offset) > 0) {
        error = read_data_from_range_query_buff(buf, need_val);
5110
    }
5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122
    else {
        invalidate_bulk_fetch();
        if (doing_bulk_fetch) {
            struct smart_dbt_bf_info bf_info;
            bf_info.ha = this;
            // you need the val if you have a clustering index and key_read is not 0;
            bf_info.direction = direction;
            bf_info.thd = ha_thd();
            bf_info.need_val = need_val;
            //
            // call c_getf_next with purpose of filling in range_query_buff
            //
5123
            rows_fetched_using_bulk_fetch = 0;
5124
            if (direction > 0) {
5125 5126 5127
                error = cursor->c_getf_next(cursor, flags, smart_dbt_bf_callback, &bf_info);
            } else {
                error = cursor->c_getf_prev(cursor, flags, smart_dbt_bf_callback, &bf_info);
5128
            }
5129 5130 5131
            if (bulk_fetch_iteration < HA_TOKU_BULK_FETCH_ITERATION_MAX) {
                bulk_fetch_iteration++;
            }
5132

5133
            error = handle_cursor_error(error, HA_ERR_END_OF_FILE,tokudb_active_index);
5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144
            if (error) { goto cleanup; }
            
            //
            // now that range_query_buff is filled, read an element
            //
            error = read_data_from_range_query_buff(buf, need_val);
        }
        else {
            struct smart_dbt_info info;
            info.ha = this;
            info.buf = buf;
5145
            info.keynr = tokudb_active_index;
5146 5147 5148 5149 5150 5151

            if (direction > 0) {
                error = cursor->c_getf_next(cursor, flags, SMART_DBT_CALLBACK, &info);
            } else {
                error = cursor->c_getf_prev(cursor, flags, SMART_DBT_CALLBACK, &info);
            }
5152
            error = handle_cursor_error(error, HA_ERR_END_OF_FILE, tokudb_active_index);
5153 5154 5155
        }
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5156
    //
5157 5158 5159 5160 5161 5162
    // at this point, one of two things has happened
    // either we have unpacked the data into buf, and we 
    // are done, or we have unpacked the primary key
    // into last_key, and we use the code below to
    // read the full row by doing a point query into the 
    // main table.
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5163
    //
5164
    
5165
    if (!error && !key_read && (tokudb_active_index != primary_key) && !(table->key_info[tokudb_active_index].flags & HA_CLUSTERING) ) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5166 5167
        error = read_full_row(buf);
    }
5168 5169
    trx->stmt_progress.queried++;
    track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5170
cleanup:
5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187
    return error;
}


//
// Reads the next row from the active index (cursor) into buf, and advances cursor
// Parameters:
//      [out]   buf - buffer for the next row, in MySQL format
// Returns:
//      0 on success
//      HA_ERR_END_OF_FILE if not found
//      error otherwise
//
int ha_tokudb::index_next(uchar * buf) {
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next");
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);
    int error = get_next(buf, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5188
    TOKUDB_DBUG_RETURN(error);
5189 5190
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5191 5192 5193 5194 5195 5196

int ha_tokudb::index_read_last(uchar * buf, const uchar * key, uint key_len) {
    return(index_read(buf, key, key_len, HA_READ_PREFIX_LAST));    
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
5197 5198 5199 5200 5201 5202 5203 5204 5205
//
// Reads the previous row from the active index (cursor) into buf, and advances cursor
// Parameters:
//      [out]   buf - buffer for the next row, in MySQL format
// Returns:
//      0 on success
//      HA_ERR_END_OF_FILE if not found
//      error otherwise
//
5206
int ha_tokudb::index_prev(uchar * buf) {
5207
    TOKUDB_DBUG_ENTER("ha_tokudb::index_prev");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5208
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);
5209
    int error = get_next(buf, -1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5210
    TOKUDB_DBUG_RETURN(error);
5211 5212
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5213 5214 5215 5216 5217 5218 5219 5220 5221
//
// Reads the first row from the active index (cursor) into buf, and advances cursor
// Parameters:
//      [out]   buf - buffer for the next row, in MySQL format
// Returns:
//      0 on success
//      HA_ERR_END_OF_FILE if not found
//      error otherwise
//
5222
int ha_tokudb::index_first(uchar * buf) {
5223
    TOKUDB_DBUG_ENTER("ha_tokudb::index_first");
5224
    invalidate_bulk_fetch();
5225
    int error = 0;
5226
    struct smart_dbt_info info;
Yoni Fogel's avatar
Yoni Fogel committed
5227
    uint32_t flags = SET_PRELOCK_FLAG(0);
5228 5229
    THD* thd = ha_thd();
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5230
    HANDLE_INVALID_CURSOR();
5231

5232
    statistic_increment(table->in_use->status_var.ha_read_first_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5233

5234 5235
    info.ha = this;
    info.buf = buf;
5236
    info.keynr = tokudb_active_index;
5237

5238 5239
    error = cursor->c_getf_first(cursor, flags,
            SMART_DBT_CALLBACK, &info);
5240
    error = handle_cursor_error(error,HA_ERR_END_OF_FILE,tokudb_active_index);
5241

5242 5243 5244 5245
    //
    // still need to get entire contents of the row if operation done on
    // secondary DB and it was NOT a covering index
    //
5246
    if (!error && !key_read && (tokudb_active_index != primary_key) && !(table->key_info[tokudb_active_index].flags & HA_CLUSTERING) ) {
5247
        error = read_full_row(buf);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5248
    }
5249 5250 5251
    trx->stmt_progress.queried++;
    track_progress(thd);
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5252
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5253
    TOKUDB_DBUG_RETURN(error);
5254 5255
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5256 5257 5258 5259 5260 5261 5262 5263 5264
//
// Reads the last row from the active index (cursor) into buf, and advances cursor
// Parameters:
//      [out]   buf - buffer for the next row, in MySQL format
// Returns:
//      0 on success
//      HA_ERR_END_OF_FILE if not found
//      error otherwise
//
5265
int ha_tokudb::index_last(uchar * buf) {
5266
    TOKUDB_DBUG_ENTER("ha_tokudb::index_last");
5267
    invalidate_bulk_fetch();
5268
    int error = 0;
5269
    struct smart_dbt_info info;
Yoni Fogel's avatar
Yoni Fogel committed
5270
    uint32_t flags = SET_PRELOCK_FLAG(0);
5271 5272
    THD* thd = ha_thd();
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5273
    HANDLE_INVALID_CURSOR();
5274

5275
    statistic_increment(table->in_use->status_var.ha_read_last_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5276

5277 5278
    info.ha = this;
    info.buf = buf;
5279
    info.keynr = tokudb_active_index;
5280

5281 5282
    error = cursor->c_getf_last(cursor, flags,
            SMART_DBT_CALLBACK, &info);
5283
    error = handle_cursor_error(error,HA_ERR_END_OF_FILE,tokudb_active_index);
5284 5285 5286 5287
    //
    // still need to get entire contents of the row if operation done on
    // secondary DB and it was NOT a covering index
    //
5288
    if (!error && !key_read && (tokudb_active_index != primary_key) && !(table->key_info[tokudb_active_index].flags & HA_CLUSTERING) ) {
5289
        error = read_full_row(buf);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5290
    }
5291 5292 5293 5294 5295

    if (trx) {
        trx->stmt_progress.queried++;
    }
    track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5296
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5297
    TOKUDB_DBUG_RETURN(error);
5298 5299
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5300 5301 5302 5303 5304 5305 5306 5307
//
// Initialize a scan of the table (which is why index_init is called on primary_key)
// Parameters:
//          scan - unused
// Returns:
//      0 on success
//      error otherwise
//
5308
int ha_tokudb::rnd_init(bool scan) {
5309
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_init");
5310
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5311
    range_lock_grabbed = false;
5312
    error = index_init(MAX_KEY, 0);
5313 5314
    if (error) { goto cleanup;}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5315
    if (scan) {
5316
        error = prelock_range(NULL, NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5317
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5318
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5319
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5320
    // only want to set range_lock_grabbed to true after index_init
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5321 5322 5323 5324
    // successfully executed for two reasons:
    // 1) index_init will reset it to false anyway
    // 2) if it fails, we don't want prelocking on,
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5325
    if (scan) { range_lock_grabbed = true; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5326
    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5327
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5328 5329 5330 5331
    if (error) { 
        index_end();
        last_cursor_error = error; 
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5332
    TOKUDB_DBUG_RETURN(error);
5333 5334
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5335 5336 5337
//
// End a scan of the table
//
5338
int ha_tokudb::rnd_end() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5339
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_end");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5340
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5341
    TOKUDB_DBUG_RETURN(index_end());
5342 5343
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5344

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5345 5346 5347 5348 5349 5350 5351 5352 5353
//
// Read the next row in a table scan
// Parameters:
//      [out]   buf - buffer for the next row, in MySQL format
// Returns:
//      0 on success
//      HA_ERR_END_OF_FILE if not found
//      error otherwise
//
5354
int ha_tokudb::rnd_next(uchar * buf) {
5355
    TOKUDB_DBUG_ENTER("ha_tokudb::ha_tokudb::rnd_next");
5356
    statistic_increment(table->in_use->status_var.ha_read_rnd_next_count, &LOCK_status);
5357
    int error = get_next(buf, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5358
    TOKUDB_DBUG_RETURN(error);
5359 5360 5361
}


5362 5363 5364
void ha_tokudb::track_progress(THD* thd) {
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
    if (trx) {
5365
        ulonglong num_written = trx->stmt_progress.inserted + trx->stmt_progress.updated + trx->stmt_progress.deleted;
5366
        bool update_status = 
5367
            (trx->stmt_progress.queried && tokudb_read_status_frequency && (trx->stmt_progress.queried % tokudb_read_status_frequency) == 0) ||
5368
            (num_written && tokudb_write_status_frequency && (num_written % tokudb_write_status_frequency) == 0);
5369
        if (update_status) {
5370 5371 5372 5373
            char *next_status = write_status_msg;
            bool first = true;
            int r;
            if (trx->stmt_progress.queried) {
5374 5375
                r = sprintf(next_status, "Queried about %llu row%s", trx->stmt_progress.queried, trx->stmt_progress.queried == 1 ? "" : "s"); 
                assert(r >= 0);
5376 5377 5378 5379
                next_status += r;
                first = false;
            }
            if (trx->stmt_progress.inserted) {
5380 5381 5382 5383 5384 5385
                if (trx->stmt_progress.using_loader) {
                    r = sprintf(next_status, "%sFetched about %llu row%s, loading data still remains", first ? "" : ", ", trx->stmt_progress.inserted, trx->stmt_progress.inserted == 1 ? "" : "s"); 
                }
                else {
                    r = sprintf(next_status, "%sInserted about %llu row%s", first ? "" : ", ", trx->stmt_progress.inserted, trx->stmt_progress.inserted == 1 ? "" : "s"); 
                }
5386
                assert(r >= 0);
5387 5388 5389 5390
                next_status += r;
                first = false;
            }
            if (trx->stmt_progress.updated) {
5391 5392
                r = sprintf(next_status, "%sUpdated about %llu row%s", first ? "" : ", ", trx->stmt_progress.updated, trx->stmt_progress.updated == 1 ? "" : "s"); 
                assert(r >= 0);
5393 5394 5395 5396
                next_status += r;
                first = false;
            }
            if (trx->stmt_progress.deleted) {
5397 5398
                r = sprintf(next_status, "%sDeleted about %llu row%s", first ? "" : ", ", trx->stmt_progress.deleted, trx->stmt_progress.deleted == 1 ? "" : "s"); 
                assert(r >= 0);
5399 5400 5401
                next_status += r;
                first = false;
            }
5402
            if (!first)
5403
                thd_proc_info(thd, write_status_msg);
5404 5405 5406 5407 5408
        }
    }
}


5409
DBT *ha_tokudb::get_pos(DBT * to, uchar * pos) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5410
    TOKUDB_DBUG_ENTER("ha_tokudb::get_pos");
5411
    /* We don't need to set app_data here */
5412
    memset((void *) to, 0, sizeof(*to));
Yoni Fogel's avatar
Yoni Fogel committed
5413 5414
    to->data = pos + sizeof(uint32_t);
    to->size = *(uint32_t *)pos;
5415
    DBUG_DUMP("key", (const uchar *) to->data, to->size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5416
    DBUG_RETURN(to);
5417 5418
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5419
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5420
// Retrieves a row with based on the primary key saved in pos
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5421 5422 5423 5424 5425
// Returns:
//      0 on success
//      HA_ERR_KEY_NOT_FOUND if not found
//      error otherwise
//
5426
int ha_tokudb::rnd_pos(uchar * buf, uchar * pos) {
5427
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_pos");
5428
    DBT db_pos;
5429
    int error = 0;
5430 5431 5432 5433 5434
    struct smart_dbt_info info;
    bool old_unpack_entire_row = unpack_entire_row;
    DBT* key = get_pos(&db_pos, pos); 

    unpack_entire_row = true;
5435
    statistic_increment(table->in_use->status_var.ha_read_rnd_count, &LOCK_status);
5436
    tokudb_active_index = MAX_KEY;
5437 5438 5439 5440 5441

    info.ha = this;
    info.buf = buf;
    info.keynr = primary_key;

5442 5443 5444
    error = share->file->getf_set(share->file, transaction, 
            get_cursor_isolation_flags(lock.type, ha_thd()), 
            key, smart_dbt_callback_rowread_ptquery, &info);
5445

Barry Perlman's avatar
Barry Perlman committed
5446
    if (error == DB_NOTFOUND) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5447 5448 5449 5450
        error = HA_ERR_KEY_NOT_FOUND;
        goto cleanup;
    }
cleanup:
5451
    unpack_entire_row = old_unpack_entire_row;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5452
    TOKUDB_DBUG_RETURN(error);
5453 5454
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5455
int ha_tokudb::prelock_range( const key_range *start_key, const key_range *end_key) {
5456
    TOKUDB_DBUG_ENTER("ha_tokudb::prelock_range");
5457
    THD* thd = ha_thd(); 
5458 5459

    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5460 5461
    DBT start_dbt_key;
    DBT end_dbt_key;
5462 5463
    uchar* start_key_buff  = prelocked_left_range;
    uchar* end_key_buff = prelocked_right_range;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5464

5465 5466
    memset((void *) &start_dbt_key, 0, sizeof(start_dbt_key));
    memset((void *) &end_dbt_key, 0, sizeof(end_dbt_key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5467

5468
    HANDLE_INVALID_CURSOR();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5469 5470 5471
    if (start_key) {
        switch (start_key->flag) {
        case HA_READ_AFTER_KEY:
5472
            pack_key(&start_dbt_key, tokudb_active_index, start_key_buff, start_key->key, start_key->length, COL_POS_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5473 5474
            break;
        default:
5475
            pack_key(&start_dbt_key, tokudb_active_index, start_key_buff, start_key->key, start_key->length, COL_NEG_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5476 5477
            break;
        }
5478 5479 5480 5481
        prelocked_left_range_size = start_dbt_key.size;
    }
    else {
        prelocked_left_range_size = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5482
    }
5483

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5484 5485 5486
    if (end_key) {
        switch (end_key->flag) {
        case HA_READ_BEFORE_KEY:
5487
            pack_key(&end_dbt_key, tokudb_active_index, end_key_buff, end_key->key, end_key->length, COL_NEG_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5488 5489
            break;
        default:
5490
            pack_key(&end_dbt_key, tokudb_active_index, end_key_buff, end_key->key, end_key->length, COL_POS_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5491
            break;
5492
        }        
5493 5494 5495 5496
        prelocked_right_range_size = end_dbt_key.size;
    }
    else {
        prelocked_right_range_size = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5497 5498
    }

5499 5500
    error = cursor->c_pre_acquire_range_lock(
        cursor, 
5501 5502
        start_key ? &start_dbt_key : share->key_file[tokudb_active_index]->dbt_neg_infty(), 
        end_key ? &end_dbt_key : share->key_file[tokudb_active_index]->dbt_pos_infty()
5503
        );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5504 5505 5506 5507 5508 5509
    if (error){ 
        last_cursor_error = error;
        //
        // cursor should be initialized here, but in case it is not, we still check
        //
        if (cursor) {
5510 5511
            int r = cursor->c_close(cursor);
            assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5512 5513 5514 5515
            cursor = NULL;
        }
        goto cleanup; 
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5516

5517 5518 5519 5520 5521
    //
    // at this point, determine if we will be doing bulk fetch
    // as of now, only do it if we are doing a select
    //
    doing_bulk_fetch = (thd_sql_command(thd) == SQLCOM_SELECT);
5522 5523
    bulk_fetch_iteration = 0;
    rows_fetched_using_bulk_fetch = 0;
5524

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5525 5526
cleanup:
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5527
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5528

5529 5530 5531 5532 5533 5534
//
// Prelock range if possible, start_key is leftmost, end_key is rightmost
// whether scanning forward or backward.  This function is called by MySQL
// for backward range queries (in QUICK_SELECT_DESC::get_next). 
// Forward scans use read_range_first()/read_range_next().
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5535
int ha_tokudb::prepare_range_scan( const key_range *start_key, const key_range *end_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5536 5537 5538 5539 5540
    int error = prelock_range(start_key, end_key);
    if (!error) {
        range_lock_grabbed = true;
    }
    return error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5541 5542 5543 5544 5545 5546 5547 5548
}

int ha_tokudb::read_range_first(
    const key_range *start_key,
    const key_range *end_key,
    bool eq_range, 
    bool sorted) 
{
5549
    int error = prelock_range(start_key, end_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5550 5551 5552 5553 5554 5555 5556
    if (error) { goto cleanup; }
    range_lock_grabbed = true;
    
    error = handler::read_range_first(start_key, end_key, eq_range, sorted);
cleanup:
    return error;
}
5557

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5558 5559
int ha_tokudb::read_range_next()
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5560 5561 5562 5563
    TOKUDB_DBUG_ENTER("ha_tokudb::read_range_next");
    int error;
    error = handler::read_range_next();
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5564
        range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5565 5566
    }
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5567 5568 5569 5570
}



5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591
/*
  Set a reference to the current record in (ref,ref_length).

  SYNOPSIS
  ha_tokudb::position()
  record                      The current record buffer

  DESCRIPTION
  The BDB handler stores the primary key in (ref,ref_length).
  There is either an explicit primary key, or an implicit (hidden)
  primary key.
  During open(), 'ref_length' is calculated as the maximum primary
  key length. When an actual key is shorter than that, the rest of
  the buffer must be cleared out. The row cannot be identified, if
  garbage follows behind the end of the key. There is no length
  field for the current key, so that the whole ref_length is used
  for comparison.

  RETURN
  nothing
*/
5592
void ha_tokudb::position(const uchar * record) {
5593
    TOKUDB_DBUG_ENTER("ha_tokudb::position");
5594 5595
    DBT key;
    if (hidden_primary_key) {
Yoni Fogel's avatar
Yoni Fogel committed
5596 5597 5598
        DBUG_ASSERT(ref_length == (TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH + sizeof(uint32_t)));
        memcpy_fixed(ref + sizeof(uint32_t), current_ident, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
        *(uint32_t *)ref = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5599 5600
    } 
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5601 5602 5603 5604
        bool has_null;
        //
        // save the data
        //
Yoni Fogel's avatar
Yoni Fogel committed
5605
        create_dbt_key_from_table(&key, primary_key, ref + sizeof(uint32_t), record, &has_null);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5606 5607 5608
        //
        // save the size of data in the first four bytes of ref
        //
Yoni Fogel's avatar
Yoni Fogel committed
5609
        memcpy(ref, &key.size, sizeof(uint32_t));
5610 5611 5612 5613
    }
    DBUG_VOID_RETURN;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5614 5615 5616 5617 5618 5619
//
// Per InnoDB: Returns statistics information of the table to the MySQL interpreter,
// in various fields of the handle object. 
// Return:
//      0, always success
//
5620
int ha_tokudb::info(uint flag) {
5621
    TOKUDB_DBUG_ENTER("ha_tokudb::info %p %d %lld", this, flag, (long long) share->rows);
5622 5623 5624 5625
    int error;
    DB_TXN* txn = NULL;
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
    DB_BTREE_STAT64 dict_stats;
5626 5627
    if (flag & HA_STATUS_VARIABLE) {
        // Just to get optimizations right
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5628
        stats.records = share->rows + share->rows_from_locked_table;
5629 5630 5631
        if (stats.records == 0) {
            stats.records++;
        }
5632
        stats.deleted = 0;
5633
        if (!(flag & HA_STATUS_NO_LOCK)) {
Yoni Fogel's avatar
Yoni Fogel committed
5634
            uint64_t num_rows = 0;
5635 5636
            TOKU_DB_FRAGMENTATION_S frag_info;
            memset(&frag_info, 0, sizeof frag_info);
5637

5638 5639 5640
            error = db_env->txn_begin(db_env, NULL, &txn, DB_READ_UNCOMMITTED);
            if (error) { goto cleanup; }

5641 5642 5643
            // we should always have a primary key
            assert(share->file != NULL);

5644 5645 5646 5647
            error = estimate_num_rows(share->file,&num_rows, txn);
            if (error == 0) {
                share->rows = num_rows;
                stats.records = num_rows;
5648 5649 5650
                if (stats.records == 0) {
                    stats.records++;
                }
5651 5652 5653 5654
            }
            else {
                goto cleanup;
            }
5655 5656 5657 5658 5659 5660
            error = share->file->get_fragmentation(
                share->file,
                &frag_info
                );
            if (error) { goto cleanup; }
            stats.delete_length = frag_info.unused_bytes;
5661

5662 5663 5664 5665 5666 5667
            error = share->file->stat64(
                share->file, 
                txn, 
                &dict_stats
                );
            if (error) { goto cleanup; }
5668
            
5669 5670
            stats.create_time = dict_stats.bt_create_time_sec;
            stats.update_time = dict_stats.bt_modify_time_sec;
5671
            stats.check_time = dict_stats.bt_verify_time_sec;
5672
            stats.data_file_length = dict_stats.bt_dsize;
5673 5674 5675 5676 5677
            if (hidden_primary_key) {
                //
                // in this case, we have a hidden primary key, do not
                // want to report space taken up by the hidden primary key to the user
                //
Yoni Fogel's avatar
Yoni Fogel committed
5678
                uint64_t hpk_space = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH*dict_stats.bt_ndata;
5679 5680 5681 5682 5683 5684
                stats.data_file_length = (hpk_space > stats.data_file_length) ? 0 : stats.data_file_length - hpk_space;
            }
            else {
                //
                // one infinity byte per key needs to be subtracted
                //
Yoni Fogel's avatar
Yoni Fogel committed
5685
                uint64_t inf_byte_space = dict_stats.bt_ndata;
5686 5687 5688
                stats.data_file_length = (inf_byte_space > stats.data_file_length) ? 0 : stats.data_file_length - inf_byte_space;
            }

5689
            stats.mean_rec_length = stats.records ? (ulong)(stats.data_file_length/stats.records) : 0;
5690
            stats.index_file_length = 0;
5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701
            // curr_num_DBs is the number of keys we have, according
            // to the mysql layer. if drop index is running concurrently
            // with info() (it can, because info does not take table locks),
            // then it could be the case that one of the dbs was dropped
            // and set to NULL before mysql was able to set table->s->keys
            // accordingly. 
            //
            // we should just ignore any DB * that is NULL. 
            //
            // this solution is much simpler than trying to maintain an 
            // accurate number of valid keys at the handlerton layer.
5702
            for (uint i = 0; i < curr_num_DBs; i++) {
5703 5704
                // skip the primary key, skip dropped indexes
                if (i == primary_key || share->key_file[i] == NULL) {
5705 5706 5707 5708 5709 5710 5711 5712 5713
                    continue;
                }
                error = share->key_file[i]->stat64(
                    share->key_file[i], 
                    txn, 
                    &dict_stats
                    );
                if (error) { goto cleanup; }
                stats.index_file_length += dict_stats.bt_dsize;
5714 5715 5716 5717 5718 5719 5720

                error = share->file->get_fragmentation(
                    share->file,
                    &frag_info
                    );
                if (error) { goto cleanup; }
                stats.delete_length += frag_info.unused_bytes;
5721 5722
            }
        }
5723
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5724
    if ((flag & HA_STATUS_CONST)) {
5725
        stats.max_data_file_length=  9223372036854775807ULL;
5726
        for (uint i = 0; i < table_share->keys; i++) {
5727 5728 5729
            bool is_unique_key = (i == primary_key) || (table->key_info[i].flags & HA_NOSAME);
            ulong val = (is_unique_key) ? 1 : 0;
            table->key_info[i].rec_per_key[table->key_info[i].key_parts - 1] = val;
5730 5731 5732
        }
    }
    /* Don't return key if we got an error for the internal primary key */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5733
    if (flag & HA_STATUS_ERRKEY && last_dup_key < table_share->keys) {
5734
        errkey = last_dup_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5735
    }    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5736
    if (flag & HA_STATUS_AUTO && table->found_next_number_field) {        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5737 5738
        THD *thd= table->in_use;
        struct system_variables *variables= &thd->variables;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5739 5740
        stats.auto_increment_value = share->last_auto_increment + variables->auto_increment_increment;
    }
5741 5742 5743
    error = 0;
cleanup:
    if (txn != NULL) {
5744
        commit_txn(txn, DB_TXN_NOSYNC);
5745 5746 5747
        txn = NULL;
    }
    TOKUDB_DBUG_RETURN(error);
5748 5749
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5750 5751 5752
//
//  Per InnoDB: Tells something additional to the handler about how to do things.
//
5753
int ha_tokudb::extra(enum ha_extra_function operation) {
5754
    TOKUDB_DBUG_ENTER("extra %p %d", this, operation);
5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773
    switch (operation) {
    case HA_EXTRA_RESET_STATE:
        reset();
        break;
    case HA_EXTRA_KEYREAD:
        key_read = 1;           // Query satisfied with key
        break;
    case HA_EXTRA_NO_KEYREAD:
        key_read = 0;
        break;
    case HA_EXTRA_IGNORE_DUP_KEY:
        using_ignore = 1;
        break;
    case HA_EXTRA_NO_IGNORE_DUP_KEY:
        using_ignore = 0;
        break;
    default:
        break;
    }
5774
    TOKUDB_DBUG_RETURN(0);
5775 5776 5777
}

int ha_tokudb::reset(void) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5778
    TOKUDB_DBUG_ENTER("ha_tokudb::reset");
5779 5780
    key_read = 0;
    using_ignore = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5781
    TOKUDB_DBUG_RETURN(0);
5782 5783
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796

//
// helper function that iterates through all DB's 
// and grabs a lock (either read or write, but not both)
// Parameters:
//      [in]    trans - transaction to be used to pre acquire the lock
//              lt - type of lock to get, either lock_read or lock_write
//  Returns:
//      0 on success
//      error otherwise
//
int ha_tokudb::acquire_table_lock (DB_TXN* trans, TABLE_LOCK_TYPE lt) {
    int error = ENOSYS;
5797 5798
    rw_rdlock(&share->num_DBs_lock);
    uint curr_num_DBs = share->num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5799
    if (lt == lock_read) {
5800 5801
        error = 0;
        goto cleanup;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5802 5803
    }
    else if (lt == lock_write) {
5804 5805
        if (tokudb_debug & TOKUDB_DEBUG_LOCK)
            TOKUDB_TRACE("%s\n", __FUNCTION__);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5806 5807 5808
        for (uint i = 0; i < curr_num_DBs; i++) {
            DB* db = share->key_file[i];
            error = db->pre_acquire_table_lock(db, trans);
5809 5810 5811
            if (error == EINVAL) 
                TOKUDB_TRACE("%s %d db=%p trans=%p\n", __FUNCTION__, i, db, trans);
            if (error) break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5812
        }
5813 5814 5815
        if (tokudb_debug & TOKUDB_DEBUG_LOCK)
            TOKUDB_TRACE("%s error=%d\n", __FUNCTION__, error);
        if (error) goto cleanup;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5816 5817 5818 5819 5820 5821 5822 5823
    }
    else {
        error = ENOSYS;
        goto cleanup;
    }

    error = 0;
cleanup:
5824
    rw_unlock(&share->num_DBs_lock);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5825 5826 5827 5828
    return error;
}


5829 5830 5831 5832
int ha_tokudb::create_txn(THD* thd, tokudb_trx_data* trx) {
    int error;
    ulong tx_isolation = thd_tx_isolation(thd);
    HA_TOKU_ISO_LEVEL toku_iso_level = tx_to_toku_iso(tx_isolation);
5833 5834
    bool is_autocommit = !thd_test_options(
            thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN);
5835 5836

    /* First table lock, start transaction */
5837
    if (thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN) && 
5838 5839 5840
         !trx->all &&
         (thd_sql_command(thd) != SQLCOM_CREATE_TABLE) &&
         (thd_sql_command(thd) != SQLCOM_DROP_TABLE) &&
5841 5842
         (thd_sql_command(thd) != SQLCOM_DROP_INDEX) &&
         (thd_sql_command(thd) != SQLCOM_CREATE_INDEX) &&
5843 5844
         (thd_sql_command(thd) != SQLCOM_ALTER_TABLE)) {
        /* QQQ We have to start a master transaction */
5845
        // DBUG_PRINT("trans", ("starting transaction all "));
5846 5847 5848 5849 5850
        if ((error = db_env->txn_begin(db_env, NULL, &trx->all, toku_iso_to_txn_flag(toku_iso_level)))) {
            trx->tokudb_lock_count--;      // We didn't get the lock
            goto cleanup;
        }
        if (tokudb_debug & TOKUDB_DEBUG_TXN) {
5851
            TOKUDB_TRACE("just created master:%p\n", trx->all);
5852 5853
        }
        trx->sp_level = trx->all;
Yoni Fogel's avatar
Yoni Fogel committed
5854
        trans_register_ha(thd, true, tokudb_hton);
5855 5856 5857 5858 5859 5860 5861
    }
    DBUG_PRINT("trans", ("starting transaction stmt"));
    if (trx->stmt) { 
        if (tokudb_debug & TOKUDB_DEBUG_TXN) {
            TOKUDB_TRACE("warning:stmt=%p\n", trx->stmt);
        }
    }
Yoni Fogel's avatar
Yoni Fogel committed
5862
    uint32_t txn_begin_flags;
5863
    if (trx->all == NULL) {
5864
        txn_begin_flags = toku_iso_to_txn_flag(toku_iso_level);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5865 5866 5867 5868 5869 5870 5871
        //
        // if the isolation level that the user has set is serializable,
        // but autocommit is on and this is just a select,
        // then we can go ahead and set the isolation level to
        // be a snapshot read, because we can serialize
        // the transaction to be the point in time at which the snapshot began.
        // 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5872
        if (txn_begin_flags == 0 && is_autocommit && thd_sql_command(thd) == SQLCOM_SELECT) {
5873 5874
            txn_begin_flags = DB_TXN_SNAPSHOT;
        }
5875 5876
    }
    else {
5877
        txn_begin_flags = DB_INHERIT_ISOLATION;
5878 5879 5880 5881 5882 5883
    }
    if ((error = db_env->txn_begin(db_env, trx->sp_level, &trx->stmt, txn_begin_flags))) {
        /* We leave the possible master transaction open */
        trx->tokudb_lock_count--;  // We didn't get the lock
        goto cleanup;
    }
5884
    trx->sub_sp_level = trx->stmt;
5885
    if (tokudb_debug & TOKUDB_DEBUG_TXN) {
5886
        TOKUDB_TRACE("just created stmt:%p:%p\n", trx->sp_level, trx->stmt);
5887
    }
5888
    reset_stmt_progress(&trx->stmt_progress);
Yoni Fogel's avatar
Yoni Fogel committed
5889
    trans_register_ha(thd, false, tokudb_hton);
5890 5891 5892 5893
cleanup:
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5894

5895 5896 5897 5898 5899 5900 5901 5902
/*
  As MySQL will execute an external lock for every new table it uses
  we can use this to start the transactions.
  If we are in auto_commit mode we just need to start a transaction
  for the statement to be able to rollback the statement.
  If not, we have to start a master transaction if there doesn't exist
  one from before.
*/
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5903 5904 5905 5906 5907 5908 5909 5910
//
// Parameters:
//      [in]    thd - handle to the user thread
//              lock_type - the type of lock
// Returns:
//      0 on success
//      error otherwise
//
5911
int ha_tokudb::external_lock(THD * thd, int lock_type) {
5912 5913 5914
    TOKUDB_DBUG_ENTER("ha_tokudb::external_lock cmd=%d %d", thd_sql_command(thd), lock_type);
    if (tokudb_debug & TOKUDB_DEBUG_LOCK)
        TOKUDB_TRACE("%s cmd=%d %d\n", __FUNCTION__, thd_sql_command(thd), lock_type);
5915

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5916 5917 5918
    int error = 0;
    tokudb_trx_data *trx = NULL;
    trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
5919
    if (!trx) {
5920 5921
        error = create_tokudb_trx_data_instance(&trx);
        if (error) { goto cleanup; }
5922
        thd_data_set(thd, tokudb_hton->slot, trx);
5923
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5924 5925
    if (trx->all == NULL) {
        trx->sp_level = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5926
    }
5927
    if (lock_type != F_UNLCK) {
5928
        is_fast_alter_running = false;
5929
        use_write_locks = false;
5930
        if (lock_type == F_WRLCK)
5931
            use_write_locks = true;
5932 5933
        if (!trx->tokudb_lock_count++) {
            DBUG_ASSERT(trx->stmt == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5934
            transaction = NULL;    // Safety
5935 5936
            error = create_txn(thd, trx);
            if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5937
                goto cleanup;
5938 5939
            }
        }
5940 5941
        assert(thd->in_sub_stmt == 0);
        transaction = trx->sub_sp_level;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5942 5943
    }
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954
        pthread_mutex_lock(&share->mutex);
        // hate dealing with comparison of signed vs unsigned, so doing this
        if (deleted_rows > added_rows && share->rows < (deleted_rows - added_rows)) {
            share->rows = 0;
        }
        else {
            share->rows += (added_rows - deleted_rows);
        }
        pthread_mutex_unlock(&share->mutex);
        added_rows = 0;
        deleted_rows = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5955
        share->rows_from_locked_table = 0;
5956
        if (trx->tokudb_lock_count > 0 && !--trx->tokudb_lock_count) {
5957 5958 5959 5960 5961 5962 5963
            if (trx->stmt) {
                /*
                   F_UNLCK is done without a transaction commit / rollback.
                   This happens if the thread didn't update any rows
                   We must in this case commit the work to keep the row locks
                 */
                DBUG_PRINT("trans", ("commiting non-updating transaction"));
5964
                reset_stmt_progress(&trx->stmt_progress);
5965
                if (!is_fast_alter_running) {
5966 5967 5968
                    commit_txn(trx->stmt, 0);
                    if (tokudb_debug & TOKUDB_DEBUG_TXN) {
                        TOKUDB_TRACE("commit:%p:%d\n", trx->stmt, error);
5969 5970 5971 5972
                    }
                    trx->stmt = NULL;
                    trx->sub_sp_level = NULL;
                }
5973 5974
            }
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5975
        transaction = NULL;
5976
        is_fast_alter_running = false;
5977
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5978
cleanup:
5979 5980
    if (tokudb_debug & TOKUDB_DEBUG_LOCK)
        TOKUDB_TRACE("%s error=%d\n", __FUNCTION__, error);
5981
    TOKUDB_DBUG_RETURN(error);
5982 5983 5984 5985 5986 5987 5988 5989 5990
}


/*
  When using LOCK TABLE's external_lock is only called when the actual
  TABLE LOCK is done.
  Under LOCK TABLES, each used tables will force a call to start_stmt.
*/

5991
int ha_tokudb::start_stmt(THD * thd, thr_lock_type lock_type) {
5992
    TOKUDB_DBUG_ENTER("ha_tokudb::start_stmt cmd=%d %d", thd_sql_command(thd), lock_type);
5993
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5994 5995


5996
    tokudb_trx_data *trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
5997 5998 5999 6000 6001 6002 6003
    DBUG_ASSERT(trx);
    /*
       note that trx->stmt may have been already initialized as start_stmt()
       is called for *each table* not for each storage engine,
       and there could be many bdb tables referenced in the query
     */
    if (!trx->stmt) {
6004 6005 6006 6007
        error = create_txn(thd, trx);
        if (error) {
            goto cleanup;
        }
6008
    }
6009 6010 6011 6012 6013
    else {
        if (tokudb_debug & TOKUDB_DEBUG_TXN) {
            TOKUDB_TRACE("trx->stmt already existed\n");
        }
    }
6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024
    //
    // we know we are in lock tables
    // attempt to grab a table lock
    // if fail, continue, do not return error
    // This is because a failure ok, it simply means
    // another active transaction has some locks.
    // That other transaction modify this table
    // until it is unlocked, therefore having acquire_table_lock
    // potentially grab some locks but not all is ok.
    //
    if (lock.type <= TL_READ_NO_INSERT) {
6025
        acquire_table_lock(trx->sub_sp_level,lock_read);
6026 6027
    }
    else {
6028 6029 6030 6031 6032 6033
        if (!(thd_sql_command(thd) == SQLCOM_CREATE_INDEX ||
            thd_sql_command(thd) == SQLCOM_ALTER_TABLE ||
            thd_sql_command(thd) == SQLCOM_DROP_INDEX ||
            thd_sql_command(thd) == SQLCOM_TRUNCATE)) {
            acquire_table_lock(trx->sub_sp_level,lock_write);
        }
6034
    }    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6035 6036 6037
    if (added_rows > deleted_rows) {
        share->rows_from_locked_table = added_rows - deleted_rows;
    }
6038
    transaction = trx->sub_sp_level;
Yoni Fogel's avatar
Yoni Fogel committed
6039
    trans_register_ha(thd, false, tokudb_hton);
6040
cleanup:
6041
    TOKUDB_DBUG_RETURN(error);
6042 6043
}

6044

Yoni Fogel's avatar
Yoni Fogel committed
6045
uint32_t ha_tokudb::get_cursor_isolation_flags(enum thr_lock_type lock_type, THD* thd) {
6046 6047 6048
    uint sql_command = thd_sql_command(thd);
    bool in_lock_tables = thd_in_lock_tables(thd);

6049 6050 6051 6052 6053 6054 6055
    //
    // following InnoDB's lead and having checksum command use a snapshot read if told
    //
    if (sql_command == SQLCOM_CHECKSUM) {
        return 0;
    }
    else if ((lock_type == TL_READ && in_lock_tables) || 
6056 6057 6058
             (lock_type == TL_READ_HIGH_PRIORITY && in_lock_tables) || 
             sql_command != SQLCOM_SELECT ||
             (sql_command == SQLCOM_SELECT && lock_type >= TL_WRITE_ALLOW_WRITE)) { // select for update 
6059 6060 6061 6062 6063 6064 6065
        return DB_SERIALIZABLE;
    }
    else {
        return 0;
    }
}

6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093
/*
  The idea with handler::store_lock() is the following:

  The statement decided which locks we should need for the table
  for updates/deletes/inserts we get WRITE locks, for SELECT... we get
  read locks.

  Before adding the lock into the table lock handler (see thr_lock.c)
  mysqld calls store lock with the requested locks.  Store lock can now
  modify a write lock to a read lock (or some other lock), ignore the
  lock (if we don't want to use MySQL table locks at all) or add locks
  for many tables (like we do when we are using a MERGE handler).

  Tokudb DB changes all WRITE locks to TL_WRITE_ALLOW_WRITE (which
  signals that we are doing WRITES, but we are still allowing other
  reader's and writer's.

  When releasing locks, store_lock() are also called. In this case one
  usually doesn't have to do anything.

  In some exceptional cases MySQL may send a request for a TL_IGNORE;
  This means that we are requesting the same lock as last time and this
  should also be ignored. (This may happen when someone does a flush
  table when we have opened a part of the tables, in which case mysqld
  closes and reopens the tables and tries to get the same locks at last
  time).  In the future we will probably try to remove this.
*/

6094
THR_LOCK_DATA **ha_tokudb::store_lock(THD * thd, THR_LOCK_DATA ** to, enum thr_lock_type lock_type) {
6095
    TOKUDB_DBUG_ENTER("ha_tokudb::store_lock, lock_type=%d cmd=%d", lock_type, thd_sql_command(thd));
6096
    if (tokudb_debug & TOKUDB_DEBUG_LOCK) {
6097
        TOKUDB_TRACE("%s lock_type=%d cmd=%d\n", __FUNCTION__, lock_type, thd_sql_command(thd));
6098
    }
6099 6100 6101

    if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK) {
        // if creating a hot index
6102
        if (thd_sql_command(thd)== SQLCOM_CREATE_INDEX && get_create_index_online(thd)) {
6103 6104 6105 6106 6107 6108
            rw_rdlock(&share->num_DBs_lock);
            if (share->num_DBs == (table->s->keys + test(hidden_primary_key))) {
                lock_type = TL_WRITE_ALLOW_WRITE;
            }
            lock.type = lock_type;
            rw_unlock(&share->num_DBs_lock);
6109
        } 
6110

6111
        // 5.5 supports reads concurrent with alter table.  just use the default lock type.
6112
#if MYSQL_VERSION_ID < 50500
6113 6114 6115
        else if (thd_sql_command(thd)== SQLCOM_CREATE_INDEX || 
                 thd_sql_command(thd)== SQLCOM_ALTER_TABLE ||
                 thd_sql_command(thd)== SQLCOM_DROP_INDEX) {
6116
            // force alter table to lock out other readers
6117
            lock_type = TL_WRITE;
6118
            lock.type = lock_type;
6119
        }
6120
#endif
6121
        else {
6122 6123 6124 6125 6126 6127
            // If we are not doing a LOCK TABLE, then allow multiple writers
            if ((lock_type >= TL_WRITE_CONCURRENT_INSERT && lock_type <= TL_WRITE) && 
                !thd->in_lock_tables && thd_sql_command(thd) != SQLCOM_TRUNCATE && !thd_tablespace_op(thd)) {
                lock_type = TL_WRITE_ALLOW_WRITE;
            }
            lock.type = lock_type;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6128
        }
6129
    }
6130
    *to++ = &lock;
6131 6132
    if (tokudb_debug & TOKUDB_DEBUG_LOCK)
        TOKUDB_TRACE("%s lock_type=%d\n", __FUNCTION__, lock_type);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6133
    DBUG_RETURN(to);
6134 6135
}

6136 6137 6138 6139
static inline enum row_type
compression_method_to_row_type(enum toku_compression_method method)
{
    switch (method) {
6140
#if TOKU_INCLUDE_ROW_TYPE_COMPRESSION
6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152
    case TOKU_NO_COMPRESSION:
        return ROW_TYPE_TOKU_UNCOMPRESSED;
    case TOKU_ZLIB_METHOD:
        return ROW_TYPE_TOKU_ZLIB;
    case TOKU_QUICKLZ_METHOD:
        return ROW_TYPE_TOKU_QUICKLZ;
    case TOKU_LZMA_METHOD:
        return ROW_TYPE_TOKU_LZMA;
    case TOKU_FAST_COMPRESSION_METHOD:
        return ROW_TYPE_TOKU_FAST;
    case TOKU_SMALL_COMPRESSION_METHOD:
        return ROW_TYPE_TOKU_SMALL;
6153
#endif
6154 6155
    case TOKU_DEFAULT_COMPRESSION_METHOD:
        return ROW_TYPE_DEFAULT;
6156 6157 6158 6159 6160
    default:
        assert(false);
    }
}

6161 6162
static enum row_type
get_row_type_for_key(DB *file)
6163 6164
{
    enum toku_compression_method method;
6165
    int r = file->get_compression_method(file, &method);
6166 6167 6168 6169
    assert(r == 0);
    return compression_method_to_row_type(method);
}

6170
enum row_type
6171 6172 6173
#if MYSQL_VERSION_ID >= 50521
ha_tokudb::get_row_type(void) const
#else
6174
ha_tokudb::get_row_type(void)
6175
#endif
6176 6177 6178 6179
{
    return get_row_type_for_key(share->file);
}

6180 6181 6182 6183
static inline enum toku_compression_method
row_type_to_compression_method(enum row_type type)
{
    switch (type) {
6184
#if TOKU_INCLUDE_ROW_TYPE_COMPRESSION
6185 6186 6187 6188 6189 6190 6191 6192 6193 6194
    case ROW_TYPE_TOKU_UNCOMPRESSED:
        return TOKU_NO_COMPRESSION;
    case ROW_TYPE_TOKU_ZLIB:
        return TOKU_ZLIB_METHOD;
    case ROW_TYPE_TOKU_QUICKLZ:
        return TOKU_QUICKLZ_METHOD;
    case ROW_TYPE_TOKU_LZMA:
        return TOKU_LZMA_METHOD;
    case ROW_TYPE_TOKU_SMALL:
        return TOKU_SMALL_COMPRESSION_METHOD;
6195 6196
    case ROW_TYPE_TOKU_FAST:
        return TOKU_FAST_COMPRESSION_METHOD;
6197
#endif
6198 6199 6200 6201
    default:
        DBUG_PRINT("info", ("Ignoring ROW_FORMAT not used by TokuDB, using TOKUDB_FAST by default instead"));
    case ROW_TYPE_DEFAULT:
        return TOKU_DEFAULT_COMPRESSION_METHOD;
6202 6203 6204
    }
}

6205 6206 6207 6208 6209
static int create_sub_table(
    const char *table_name, 
    DBT* row_descriptor, 
    DB_TXN* txn, 
    uint32_t block_size, 
6210
    uint32_t read_block_size,
6211
    enum toku_compression_method compression_method,
6212 6213 6214
    bool is_hot_index
    ) 
{
6215
    TOKUDB_DBUG_ENTER("create_sub_table");
6216
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6217
    DB *file = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
6218
    uint32_t create_flags;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6219
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6220 6221 6222 6223 6224 6225 6226 6227
    
    error = db_create(&file, db_env, 0);
    if (error) {
        DBUG_PRINT("error", ("Got error: %d when creating table", error));
        my_errno = error;
        goto exit;
    }
        
6228

6229 6230
    if (block_size != 0) {
        error = file->set_pagesize(file, block_size);
6231
        if (error != 0) {
6232
            DBUG_PRINT("error", ("Got error: %d when setting block size %u for table '%s'", error, block_size, table_name));
6233 6234 6235
            goto exit;
        }
    }
6236 6237 6238 6239 6240 6241 6242
    if (read_block_size != 0) {
        error = file->set_readpagesize(file, read_block_size);
        if (error != 0) {
            DBUG_PRINT("error", ("Got error: %d when setting read block size %u for table '%s'", error, read_block_size, table_name));
            goto exit;
        }
    }
6243 6244 6245 6246 6247
    error = file->set_compression_method(file, compression_method);
    if (error != 0) {
        DBUG_PRINT("error", ("Got error: %d when setting compression type %u for table '%s'", error, compression_method, table_name));
        goto exit;
    }
6248 6249 6250

    create_flags = DB_THREAD | DB_CREATE | DB_EXCL | (is_hot_index ? DB_IS_HOT_INDEX : 0);    
    error = file->open(file, txn, table_name, NULL, DB_BTREE, create_flags, my_umask);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6251 6252 6253 6254
    if (error) {
        DBUG_PRINT("error", ("Got error: %d when opening table '%s'", error, table_name));
        goto exit;
    } 
6255

6256
    error = file->change_descriptor(file, txn, row_descriptor, (is_hot_index ? DB_IS_HOT_INDEX | DB_UPDATE_CMP_DESCRIPTOR : DB_UPDATE_CMP_DESCRIPTOR));
6257 6258 6259 6260 6261
    if (error) {
        DBUG_PRINT("error", ("Got error: %d when setting row descriptor for table '%s'", error, table_name));
        goto exit;
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6262 6263
    error = 0;
exit:
6264
    if (file) {
6265 6266
        int r = file->close(file, 0);
        assert(r==0);
6267
    }
6268
    TOKUDB_DBUG_RETURN(error);
6269 6270
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6271 6272 6273
void ha_tokudb::update_create_info(HA_CREATE_INFO* create_info) {
    if (share->has_auto_inc) {
        info(HA_STATUS_AUTO);
6274 6275 6276 6277 6278 6279 6280 6281 6282
        if (!(create_info->used_fields & HA_CREATE_USED_AUTO) ||
            create_info->auto_increment_value < stats.auto_increment_value) {
            create_info->auto_increment_value = stats.auto_increment_value;
        }
    }
    if (!(create_info->used_fields & HA_CREATE_USED_ROW_FORMAT)) {
        // show create table asks us to update this create_info, this makes it
        // so we'll always show what compression type we're using
        create_info->row_type = get_row_type();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6283 6284 6285
    }
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6286
//
6287 6288 6289 6290
// removes key name from status.tokudb.
// needed for when we are dropping indexes, so that 
// during drop table, we do not attempt to remove already dropped
// indexes because we did not keep status.tokudb in sync with list of indexes.
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6291
//
6292
int ha_tokudb::remove_key_name_from_status(DB* status_block, char* key_name, DB_TXN* txn) {
6293
    int error;
6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312
    uchar status_key_info[FN_REFLEN + sizeof(HA_METADATA_KEY)];
    HA_METADATA_KEY md_key = hatoku_key_name;
    memcpy(status_key_info, &md_key, sizeof(HA_METADATA_KEY));
    //
    // put index name in status.tokudb
    // 
    memcpy(
        status_key_info + sizeof(HA_METADATA_KEY), 
        key_name, 
        strlen(key_name) + 1
        );
    error = remove_metadata(
        status_block,
        status_key_info,
        sizeof(HA_METADATA_KEY) + strlen(key_name) + 1,
        txn
        );
    return error;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6313

6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340
//
// writes the key name in status.tokudb, so that we may later delete or rename
// the dictionary associated with key_name
//
int ha_tokudb::write_key_name_to_status(DB* status_block, char* key_name, DB_TXN* txn) {
    int error;
    uchar status_key_info[FN_REFLEN + sizeof(HA_METADATA_KEY)];
    HA_METADATA_KEY md_key = hatoku_key_name;
    memcpy(status_key_info, &md_key, sizeof(HA_METADATA_KEY));
    //
    // put index name in status.tokudb
    // 
    memcpy(
        status_key_info + sizeof(HA_METADATA_KEY), 
        key_name, 
        strlen(key_name) + 1
        );
    error = write_metadata(
        status_block,
        status_key_info,
        sizeof(HA_METADATA_KEY) + strlen(key_name) + 1,
        NULL,
        0,
        txn
        );
    return error;
}
6341

6342 6343 6344 6345 6346
//
// some tracing moved out of ha_tokudb::create, because ::create was getting cluttered
//
void ha_tokudb::trace_create_table_info(const char *name, TABLE * form) {
    uint i;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6347 6348 6349
    //
    // tracing information about what type of table we are creating
    //
6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
        for (i = 0; i < form->s->fields; i++) {
            Field *field = form->s->field[i];
            TOKUDB_TRACE("field:%d:%s:type=%d:flags=%x\n", i, field->field_name, field->type(), field->flags);
        }
        for (i = 0; i < form->s->keys; i++) {
            KEY *key = &form->s->key_info[i];
            TOKUDB_TRACE("key:%d:%s:%d\n", i, key->name, key->key_parts);
            uint p;
            for (p = 0; p < key->key_parts; p++) {
                KEY_PART_INFO *key_part = &key->key_part[p];
                Field *field = key_part->field;
                TOKUDB_TRACE("key:%d:%d:length=%d:%s:type=%d:flags=%x\n",
                             i, p, key_part->length, field->field_name, field->type(), field->flags);
            }
        }
    }
6367 6368
}

Yoni Fogel's avatar
Yoni Fogel committed
6369 6370
static uint32_t get_max_desc_size(KEY_AND_COL_INFO* kc_info, TABLE* form) {
    uint32_t max_row_desc_buff_size;
6371 6372 6373 6374 6375 6376
    max_row_desc_buff_size = 2*(form->s->fields * 6)+10; // upper bound of key comparison descriptor
    max_row_desc_buff_size += get_max_secondary_key_pack_desc_size(kc_info); // upper bound for sec. key part
    max_row_desc_buff_size += get_max_clustering_val_pack_desc_size(form->s); // upper bound for clustering val part
    return max_row_desc_buff_size;
}

Yoni Fogel's avatar
Yoni Fogel committed
6377
static uint32_t create_secondary_key_descriptor(
6378 6379 6380 6381 6382 6383
    uchar* buf,
    KEY* key_info,
    KEY* prim_key,
    uint hpk,
    TABLE* form,
    uint primary_key,
Yoni Fogel's avatar
Yoni Fogel committed
6384
    uint32_t keynr,
6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421
    KEY_AND_COL_INFO* kc_info    
    ) 
{
    uchar* ptr = NULL;

    ptr = buf;
    ptr += create_toku_key_descriptor(
        ptr,
        false,
        key_info,
        hpk,
        prim_key
        );

    ptr += create_toku_secondary_key_pack_descriptor(
        ptr,
        hpk,
        primary_key,
        form->s,
        form,
        kc_info,
        key_info,
        prim_key
        );

    ptr += create_toku_clustering_val_pack_descriptor(
        ptr,
        primary_key,
        form->s,
        kc_info,
        keynr,
        key_info->flags & HA_CLUSTERING
        );
    return ptr - buf;
}


6422 6423 6424
//
// creates dictionary for secondary index, with key description key_info, all using txn
//
6425 6426 6427 6428 6429
int ha_tokudb::create_secondary_dictionary(
    const char* name, TABLE* form, 
    KEY* key_info, 
    DB_TXN* txn, 
    KEY_AND_COL_INFO* kc_info, 
Yoni Fogel's avatar
Yoni Fogel committed
6430
    uint32_t keynr,
6431 6432
    bool is_hot_index,
    enum row_type row_type
6433 6434
    ) 
{
6435 6436 6437 6438 6439 6440
    int error;
    DBT row_descriptor;
    uchar* row_desc_buff = NULL;
    char* newname = NULL;
    KEY* prim_key = NULL;
    char dict_name[MAX_DICT_NAME_LEN];
Yoni Fogel's avatar
Yoni Fogel committed
6441
    uint32_t max_row_desc_buff_size;
6442
    uint hpk= (form->s->primary_key >= MAX_KEY) ? TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH : 0;
6443
    uint32_t block_size;
6444 6445
    uint32_t read_block_size;
    THD* thd = ha_thd();
6446

6447
    memset(&row_descriptor, 0, sizeof(row_descriptor));
6448
    
6449
    max_row_desc_buff_size = get_max_desc_size(kc_info,form);
6450 6451

    row_desc_buff = (uchar *)my_malloc(max_row_desc_buff_size, MYF(MY_WME));
6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465
    if (row_desc_buff == NULL){ error = ENOMEM; goto cleanup;}

    newname = (char *)my_malloc(get_max_dict_name_path_length(name),MYF(MY_WME));
    if (newname == NULL){ error = ENOMEM; goto cleanup;}

    sprintf(dict_name, "key-%s", key_info->name);
    make_name(newname, name, dict_name);

    prim_key = (hpk) ? NULL : &form->s->key_info[primary_key];

    //
    // setup the row descriptor
    //
    row_descriptor.data = row_desc_buff;
6466 6467 6468
    //
    // save data necessary for key comparisons
    //
6469
    row_descriptor.size = create_secondary_key_descriptor(
6470 6471
        row_desc_buff,
        key_info,
6472
        prim_key,
6473
        hpk,
6474 6475 6476
        form,
        primary_key,
        keynr,
6477
        kc_info    
6478 6479 6480
        );
    assert(row_descriptor.size <= max_row_desc_buff_size);

6481 6482 6483 6484
    block_size = key_info->block_size << 10;
    if (block_size == 0) {
        block_size = get_tokudb_block_size(thd);
    }
6485
    read_block_size = get_tokudb_read_block_size(thd);
6486

6487
    error = create_sub_table(newname, &row_descriptor, txn, block_size, read_block_size, row_type_to_compression_method(row_type), is_hot_index);
6488 6489 6490 6491 6492 6493
cleanup:    
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    my_free(row_desc_buff, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}

6494

Yoni Fogel's avatar
Yoni Fogel committed
6495
static uint32_t create_main_key_descriptor(
6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527
    uchar* buf,
    KEY* prim_key,
    uint hpk,
    uint primary_key,
    TABLE* form,
    KEY_AND_COL_INFO* kc_info
    ) 
{
    uchar* ptr = buf;
    ptr += create_toku_key_descriptor(
        ptr, 
        hpk,
        prim_key,
        false,
        NULL
        );
    
    ptr += create_toku_main_key_pack_descriptor(
        ptr
        );

    ptr += create_toku_clustering_val_pack_descriptor(
        ptr,
        primary_key,
        form->s,
        kc_info,
        primary_key,
        false
        );
    return ptr - buf;
}

6528 6529 6530 6531
//
// create and close the main dictionarr with name of "name" using table form, all within
// transaction txn.
//
6532
int ha_tokudb::create_main_dictionary(const char* name, TABLE* form, DB_TXN* txn, KEY_AND_COL_INFO* kc_info, enum row_type row_type) {
6533 6534 6535 6536 6537
    int error;
    DBT row_descriptor;
    uchar* row_desc_buff = NULL;
    char* newname = NULL;
    KEY* prim_key = NULL;
Yoni Fogel's avatar
Yoni Fogel committed
6538
    uint32_t max_row_desc_buff_size;
6539
    uint hpk= (form->s->primary_key >= MAX_KEY) ? TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH : 0;
6540
    uint32_t block_size;
6541 6542
    uint32_t read_block_size;
    THD* thd = ha_thd();
6543

6544
    memset(&row_descriptor, 0, sizeof(row_descriptor));
6545
    max_row_desc_buff_size = get_max_desc_size(kc_info, form);
6546 6547

    row_desc_buff = (uchar *)my_malloc(max_row_desc_buff_size, MYF(MY_WME));
6548 6549 6550 6551
    if (row_desc_buff == NULL){ error = ENOMEM; goto cleanup;}

    newname = (char *)my_malloc(get_max_dict_name_path_length(name),MYF(MY_WME));
    if (newname == NULL){ error = ENOMEM; goto cleanup;}
6552

6553
    make_name(newname, name, "main");
6554 6555

    prim_key = (hpk) ? NULL : &form->s->key_info[primary_key];
6556

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6557 6558 6559 6560
    //
    // setup the row descriptor
    //
    row_descriptor.data = row_desc_buff;
6561 6562 6563
    //
    // save data necessary for key comparisons
    //
6564 6565
    row_descriptor.size = create_main_key_descriptor(
        row_desc_buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6566
        prim_key,
6567
        hpk,
6568
        primary_key,
6569 6570
        form,
        kc_info
6571 6572
        );
    assert(row_descriptor.size <= max_row_desc_buff_size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6573

6574 6575 6576 6577 6578 6579
    block_size = 0;
    if (prim_key)
        block_size = prim_key->block_size << 10;
    if (block_size == 0) {
        block_size = get_tokudb_block_size(thd);
    }
6580
    read_block_size = get_tokudb_read_block_size(thd);
6581

6582
    /* Create the main table that will hold the real rows */
6583
    error = create_sub_table(newname, &row_descriptor, txn, block_size, read_block_size, row_type_to_compression_method(row_type), false);
6584 6585 6586 6587 6588 6589
cleanup:    
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    my_free(row_desc_buff, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}

6590 6591 6592
static inline enum row_type
row_format_to_row_type(srv_row_format_t row_format)
{
6593
#if TOKU_INCLUDE_ROW_TYPE_COMPRESSION
6594
    switch (row_format) {
6595 6596 6597 6598 6599 6600 6601 6602 6603 6604
    case SRV_ROW_FORMAT_UNCOMPRESSED:
        return ROW_TYPE_TOKU_UNCOMPRESSED;
    case SRV_ROW_FORMAT_ZLIB:
        return ROW_TYPE_TOKU_ZLIB;
    case SRV_ROW_FORMAT_QUICKLZ:
        return ROW_TYPE_TOKU_QUICKLZ;
    case SRV_ROW_FORMAT_LZMA:
        return ROW_TYPE_TOKU_LZMA;
    case SRV_ROW_FORMAT_SMALL:
        return ROW_TYPE_TOKU_SMALL;
6605 6606
    case SRV_ROW_FORMAT_FAST:
        return ROW_TYPE_TOKU_FAST;
6607
    case SRV_ROW_FORMAT_DEFAULT:
6608
        return ROW_TYPE_DEFAULT;
6609
    }
6610 6611 6612
    assert(0);
#endif
    return ROW_TYPE_DEFAULT;;
6613 6614
}

6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627 6628 6629 6630 6631
//
// Creates a new table
// Parameters:
//      [in]    name - table name
//      [in]    form - info on table, columns and indexes
//      [in]    create_info - more info on table, CURRENTLY UNUSED
// Returns:
//      0 on success
//      error otherwise
//
int ha_tokudb::create(const char *name, TABLE * form, HA_CREATE_INFO * create_info) {
    TOKUDB_DBUG_ENTER("ha_tokudb::create");
    int error;
    DB *status_block = NULL;
    uint version;
    uint capabilities;
    DB_TXN* txn = NULL;
6632
    bool do_commit = false;
6633
    char* newname = NULL;
6634
    KEY_AND_COL_INFO kc_info;
6635
    tokudb_trx_data *trx = NULL;
6636
    THD* thd = ha_thd();
6637
    bool create_from_engine= (create_info->table_options & HA_OPTION_CREATE_FROM_ENGINE);
6638
    memset(&kc_info, 0, sizeof(kc_info));
6639

6640
    pthread_mutex_lock(&tokudb_meta_mutex);
6641

6642 6643
    trx = (tokudb_trx_data *) thd_data_get(ha_thd(), tokudb_hton->slot);

Leif Walsh's avatar
Leif Walsh committed
6644 6645 6646 6647
    const enum row_type row_type = ((create_info->used_fields & HA_CREATE_USED_ROW_FORMAT)
                                    ? create_info->row_type
                                    : row_format_to_row_type(get_row_format(thd)));

6648 6649 6650 6651 6652
    if (create_from_engine) {
        // table already exists, nothing to do
        error = 0;
        goto cleanup;
    }
6653

6654 6655
    newname = (char *)my_malloc(get_max_dict_name_path_length(name),MYF(MY_WME));
    if (newname == NULL){ error = ENOMEM; goto cleanup;}
6656

6657
    if (thd_sql_command(thd) == SQLCOM_CREATE_TABLE && trx && trx->sub_sp_level) {
6658 6659 6660 6661 6662 6663 6664 6665
        txn = trx->sub_sp_level;
    }
    else {
        do_commit = true;
        error = db_env->txn_begin(db_env, 0, &txn, 0);
        if (error) { goto cleanup; }        
    }
    
6666 6667
    primary_key = form->s->primary_key;
    hidden_primary_key = (primary_key  >= MAX_KEY) ? TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH : 0;
6668 6669 6670
    if (hidden_primary_key) {
        primary_key = form->s->keys;
    }
6671 6672 6673 6674

    /* do some tracing */
    trace_create_table_info(name,form);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6675
    /* Create status.tokudb and save relevant metadata */
6676
    make_name(newname, name, "status");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6677

6678 6679 6680
    error = db_create(&status_block, db_env, 0);
    if (error) { goto cleanup; }

6681
    error = status_block->set_pagesize(status_block, status_dict_pagesize);
6682 6683
    if (error) { goto cleanup; }

6684
    error = status_block->open(status_block, txn, newname, NULL, DB_BTREE, DB_CREATE | DB_EXCL, 0);
6685
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6686

6687 6688 6689
    version = HA_TOKU_VERSION;
    capabilities = HA_TOKU_CAP;
    
6690
    error = write_to_status(status_block, hatoku_new_version,&version,sizeof(version), txn);
6691
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6692

6693
    error = write_to_status(status_block, hatoku_capabilities,&capabilities,sizeof(capabilities), txn);
6694 6695
    if (error) { goto cleanup; }

6696
    error = write_auto_inc_create(status_block, create_info->auto_increment_value, txn);
6697
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6698

6699
    // only for tables that are not partitioned
6700
    if (TOKU_PARTITION_WRITE_FRM_DATA || form->part_info == NULL) {
6701 6702 6703
        error = write_frm_data(status_block, txn, form->s->path.str);
        if (error) { goto cleanup; }
    }
6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714
    error = allocate_key_and_col_info(form->s, &kc_info);
    if (error) { goto cleanup; }

    error = initialize_key_and_col_info(
        form->s, 
        form,
        &kc_info,
        hidden_primary_key,
        primary_key
        );
    if (error) { goto cleanup; }
6715

6716
    error = create_main_dictionary(name, form, txn, &kc_info, row_type);
6717 6718 6719 6720
    if (error) {
        goto cleanup;
    }

6721 6722 6723

    for (uint i = 0; i < form->s->keys; i++) {
        if (i != primary_key) {
6724
            error = create_secondary_dictionary(name, form, &form->key_info[i], txn, &kc_info, i, false, row_type);
6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736
            if (error) {
                goto cleanup;
            }

            error = write_key_name_to_status(status_block, form->s->key_info[i].name, txn);
            if (error) { goto cleanup; }
        }
    }

    error = add_table_to_metadata(name, form, txn);
    if (error) { goto cleanup; }

6737
    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6738 6739
cleanup:
    if (status_block != NULL) {
6740 6741
        int r = status_block->close(status_block, 0);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6742
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6743
    free_key_and_col_info(&kc_info);
6744
    if (do_commit && txn) {
6745 6746 6747 6748 6749 6750
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn,0);
        }
6751
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6752
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
6753
    pthread_mutex_unlock(&tokudb_meta_mutex);
6754
    TOKUDB_DBUG_RETURN(error);
6755 6756
}

6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769
int ha_tokudb::discard_or_import_tablespace(my_bool discard) {
    /*
    if (discard) {
        my_errno=HA_ERR_WRONG_COMMAND;
        return my_errno;
    }
    return add_table_to_metadata(share->table_name);
    */
    my_errno=HA_ERR_WRONG_COMMAND;
    return my_errno;
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
6770
//
6771 6772 6773 6774
// deletes from_name or renames from_name to to_name, all using transaction txn.
// is_delete specifies which we are doing
// is_key specifies if it is a secondary index (and hence a "key-" needs to be prepended) or
// if it is not a secondary index
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6775
//
6776
int ha_tokudb::delete_or_rename_dictionary( const char* from_name, const char* to_name, const char* secondary_name, bool is_key, DB_TXN* txn, bool is_delete) {
6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826
    int error;
    char dict_name[MAX_DICT_NAME_LEN];
    char* new_from_name = NULL;
    char* new_to_name = NULL;
    assert(txn);
    
    new_from_name = (char *)my_malloc(
        get_max_dict_name_path_length(from_name), 
        MYF(MY_WME)
        );
    if (new_from_name == NULL) {
        error = ENOMEM;
        goto cleanup;
    }
    if (!is_delete) {
        assert(to_name);
        new_to_name = (char *)my_malloc(
            get_max_dict_name_path_length(to_name), 
            MYF(MY_WME)
            );
        if (new_to_name == NULL) {
            error = ENOMEM;
            goto cleanup;
        }
    }
    
    if (is_key) {
        sprintf(dict_name, "key-%s", secondary_name);
        make_name(new_from_name, from_name, dict_name);
    }
    else {
        make_name(new_from_name, from_name, secondary_name);
    }
    if (!is_delete) {
        if (is_key) {
            sprintf(dict_name, "key-%s", secondary_name);
            make_name(new_to_name, to_name, dict_name);
        }
        else {
            make_name(new_to_name, to_name, secondary_name);
        }
    }

    if (is_delete) {    
        error = db_env->dbremove(db_env, txn, new_from_name, NULL, 0);
    }
    else {
        error = db_env->dbrename(db_env, txn, new_from_name, NULL, new_to_name, 0);
    }
    if (error) { goto cleanup; }
6827

6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839
cleanup:
    my_free(new_from_name, MYF(MY_ALLOW_ZERO_PTR));
    my_free(new_to_name, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}


//
// deletes or renames a table. if is_delete is true, then we delete, and to_name can be NULL
// if is_delete is false, then to_name must be non-NULL, as we are renaming the table.
//
int ha_tokudb::delete_or_rename_table (const char* from_name, const char* to_name, bool is_delete) {
6840
    int error;
6841
    DB* status_db = NULL;
6842 6843 6844 6845
    DBC* status_cursor = NULL;
    DB_TXN* txn = NULL;
    DBT curr_key;
    DBT curr_val;
6846 6847
    memset(&curr_key, 0, sizeof(curr_key));
    memset(&curr_val, 0, sizeof(curr_val));
6848
    pthread_mutex_lock(&tokudb_meta_mutex);
6849

6850 6851 6852
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }

6853
    //
6854
    // modify metadata db
6855
    //
6856 6857
    if (is_delete) {
        error = drop_table_from_metadata(from_name, txn);
6858
    }
6859 6860
    else {
        error = rename_table_in_metadata(from_name, to_name, txn);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6861
    }
6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882 6883 6884 6885 6886 6887 6888 6889 6890 6891 6892 6893 6894 6895 6896 6897 6898 6899
    if (error) { goto cleanup; }

    //
    // open status db,
    // create cursor,
    // for each name read out of there, create a db and delete or rename it
    //
    error = open_status_dictionary(&status_db, from_name, txn);
    if (error) { goto cleanup; }

    error = status_db->cursor(status_db, txn, &status_cursor, 0);
    if (error) { goto cleanup; }

    while (error != DB_NOTFOUND) {
        error = status_cursor->c_get(
            status_cursor,
            &curr_key,
            &curr_val,
            DB_NEXT
            );
        if (error && error != DB_NOTFOUND) { goto cleanup; }
        if (error == DB_NOTFOUND) { break; }

        HA_METADATA_KEY mk = *(HA_METADATA_KEY *)curr_key.data;
        if (mk != hatoku_key_name) {
            continue;
        }
        error = delete_or_rename_dictionary(from_name, to_name, (char *)((char *)curr_key.data + sizeof(HA_METADATA_KEY)), true, txn, is_delete);
        if (error) { goto cleanup; }
    }

    //
    // delete or rename main.tokudb
    //
    error = delete_or_rename_dictionary(from_name, to_name, "main", false, txn, is_delete);
    if (error) { goto cleanup; }

    error = status_cursor->c_close(status_cursor);
6900
    assert(error==0);
6901
    status_cursor = NULL;
6902
    if (error) { goto cleanup; }
6903 6904

    error = status_db->close(status_db, 0);
6905
    assert(error == 0);
6906 6907 6908 6909 6910 6911 6912 6913
    status_db = NULL;
    
    //
    // delete or rename status.tokudb
    //
    error = delete_or_rename_dictionary(from_name, to_name, "status", false, txn, is_delete);
    if (error) { goto cleanup; }

6914
    my_errno = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6915
cleanup:
6916
    if (status_cursor) {
6917 6918
        int r = status_cursor->c_close(status_cursor);
        assert(r==0);
6919 6920
    }
    if (status_db) {
6921 6922
        int r = status_db->close(status_db, 0);
        assert(r==0);
6923 6924 6925 6926 6927 6928 6929 6930
    }
    if (txn) {
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn, 0);
        }
6931
    }
6932
    pthread_mutex_unlock(&tokudb_meta_mutex);
6933 6934 6935 6936 6937 6938 6939 6940 6941 6942 6943 6944 6945 6946
    return error;
}


//
// Drops table
// Parameters:
//      [in]    name - name of table to be deleted
// Returns:
//      0 on success
//      error otherwise
//
int ha_tokudb::delete_table(const char *name) {
    TOKUDB_DBUG_ENTER("ha_tokudb::delete_table");
6947 6948
    int error;
    error = delete_or_rename_table(name, NULL, true);
6949
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6950 6951 6952 6953 6954
        sql_print_error("Could not delete table %s because \
another transaction has accessed the table. \
To drop the table, make sure no transactions touch the table.", name);
    }
    TOKUDB_DBUG_RETURN(error);
6955 6956 6957
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
6958 6959 6960 6961 6962 6963 6964 6965 6966
//
// renames table from "from" to "to"
// Parameters:
//      [in]    name - old name of table
//      [in]    to - new name of table
// Returns:
//      0 on success
//      error otherwise
//
6967
int ha_tokudb::rename_table(const char *from, const char *to) {
6968
    TOKUDB_DBUG_ENTER("%s %s %s", __FUNCTION__, from, to);
6969
    int error;
6970
    error = delete_or_rename_table(from, to, false);
6971
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6972 6973 6974 6975
        sql_print_error("Could not rename table from %s to %s because \
another transaction has accessed the table. \
To rename the table, make sure no transactions touch the table.", from, to);
    }
6976
    TOKUDB_DBUG_RETURN(error);
6977 6978 6979 6980
}


/*
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6981
  Returns estimate on number of seeks it will take to read through the table
6982 6983 6984
  This is to be comparable to the number returned by records_in_range so
  that we can decide if we should scan the table or use keys.
*/
6985
/// QQQ why divide by 3
6986
double ha_tokudb::scan_time() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6987
    TOKUDB_DBUG_ENTER("ha_tokudb::scan_time");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6988
    double ret_val = (double)stats.records / 3;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6989
    DBUG_RETURN(ret_val);
6990 6991
}

6992 6993
double ha_tokudb::keyread_time(uint index, uint ranges, ha_rows rows)
{
6994
  if ((table->key_info[index].flags & HA_CLUSTERING) || (index == primary_key)) {
6995 6996 6997 6998 6999 7000 7001 7002 7003 7004 7005 7006 7007 7008 7009 7010 7011
    return read_time(index, ranges, rows);
  }
  /*
    It is assumed that we will read trough the whole key range and that all
    key blocks are half full (normally things are much better). It is also
    assumed that each time we read the next key from the index, the handler
    performs a random seek, thus the cost is proportional to the number of
    blocks read. This model does not take into account clustered indexes -
    engines that support that (e.g. InnoDB) may want to overwrite this method.
  */
  double keys_per_block= (stats.block_size/2.0/
                          (table->key_info[index].key_length +
                           ref_length) + 1);
  return (rows + keys_per_block - 1)/ keys_per_block;
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023
//
// Calculate the time it takes to read a set of ranges through an index
// This enables us to optimize reads for clustered indexes.
// Implementation pulled from InnoDB
// Parameters:
//          index - index to use
//          ranges - number of ranges
//          rows - estimated number of rows in the range
// Returns:
//      estimated time measured in disk seeks
//
double ha_tokudb::read_time(
7024 7025 7026 7027
    uint    index,
    uint    ranges,
    ha_rows rows
    )
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7028 7029 7030
{
    double total_scan;
    double ret_val; 
7031
    bool is_primary = (index == primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7032
    bool is_clustering;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7033

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7034 7035 7036 7037 7038 7039 7040
    //
    // in case for hidden primary key, this is called
    //
    if (index >= table_share->keys) {
        ret_val = handler::read_time(index, ranges, rows);
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7041 7042
    
    is_clustering = (table->key_info[index].flags & HA_CLUSTERING);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7043 7044


Zardosht Kasheff's avatar
Zardosht Kasheff committed
7045 7046 7047
    //
    // if it is not the primary key, and it is not a clustering key, then return handler::read_time
    //
7048
    if (!(is_primary || is_clustering)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7049 7050 7051 7052
        ret_val = handler::read_time(index, ranges, rows);
        goto cleanup;
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7053 7054 7055
    //
    // for primary key and for clustered keys, return a fraction of scan_time()
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7056 7057 7058
    total_scan = scan_time();

    if (stats.records < rows) {
7059
        ret_val = is_clustering ? total_scan + 0.00001 : total_scan;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7060 7061 7062 7063 7064 7065 7066
        goto cleanup;
    }

    //
    // one disk seek per range plus the proportional scan time of the rows
    //
    ret_val = (ranges + (double) rows / (double) stats.records * total_scan);
7067
    ret_val = is_clustering ? ret_val + 0.00001 : ret_val;
7068
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7069 7070 7071 7072 7073
cleanup:
    return ret_val;
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
7074
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7075 7076 7077 7078 7079 7080 7081 7082 7083 7084 7085
// Estimates the number of index records in a range. In case of errors, return
//   HA_TOKUDB_RANGE_COUNT instead of HA_POS_ERROR. This was behavior
//   when we got the handlerton from MySQL.
// Parameters:
//              keynr -index to use 
//      [in]    start_key - low end of the range
//      [in]    end_key - high end of the range
// Returns:
//      0 - There are no matching keys in the given range
//      number > 0 - There are approximately number matching rows in the range
//      HA_POS_ERROR - Something is wrong with the index tree
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7086
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7087
ha_rows ha_tokudb::records_in_range(uint keynr, key_range* start_key, key_range* end_key) {
7088
    TOKUDB_DBUG_ENTER("ha_tokudb::records_in_range");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7089
    DBT key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7090
    ha_rows ret_val = HA_TOKUDB_RANGE_COUNT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7091
    DB *kfile = share->key_file[keynr];
Yoni Fogel's avatar
Yoni Fogel committed
7092 7093 7094
    uint64_t less, equal, greater;
    uint64_t total_rows_estimate = HA_TOKUDB_RANGE_COUNT;
    uint64_t start_rows, end_rows, rows;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7095
    int is_exact;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7096
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7097
    uchar inf_byte;
7098

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7099
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7100
    // get start_rows and end_rows values so that we can estimate range
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7101 7102 7103 7104
    // when calling key_range64, the only value we can trust is the value for less
    // The reason is that the key being passed in may be a prefix of keys in the DB
    // As a result, equal may be 0 and greater may actually be equal+greater
    // So, we call key_range64 on the key, and the key that is after it.
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7105
    //
7106 7107 7108 7109 7110 7111
    if (!start_key && !end_key) {
        error = estimate_num_rows(kfile, &end_rows, transaction);
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
7112 7113
        ret_val = (end_rows <= 1) ? 1 : end_rows;
        goto cleanup;
7114
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7115
    if (start_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7116 7117 7118 7119 7120 7121 7122 7123 7124 7125
        inf_byte = (start_key->flag == HA_READ_KEY_EXACT) ? 
            COL_NEG_INF : COL_POS_INF;
        pack_key(
            &key, 
            keynr, 
            key_buff, 
            start_key->key, 
            start_key->length, 
            inf_byte
            ); 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7126
        error = kfile->key_range64(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7127 7128
            kfile, 
            transaction, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7129
            &key,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7130 7131 7132 7133
            &less,
            &equal,
            &greater,
            &is_exact
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7134 7135 7136 7137 7138
            );
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7139
        start_rows= less;
7140
        total_rows_estimate = less + equal + greater;
7141
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7142
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7143
        start_rows= 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7144 7145 7146
    }

    if (end_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7147 7148 7149 7150 7151 7152 7153 7154 7155 7156
        inf_byte = (end_key->flag == HA_READ_BEFORE_KEY) ?
            COL_NEG_INF : COL_POS_INF;
        pack_key(
            &key, 
            keynr, 
            key_buff, 
            end_key->key, 
            end_key->length, 
            inf_byte
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7157
        error = kfile->key_range64(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7158 7159
            kfile, 
            transaction, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7160
            &key,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7161 7162 7163 7164
            &less,
            &equal,
            &greater,
            &is_exact
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7165 7166 7167 7168 7169
            );
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7170
        end_rows= less;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7171 7172
    }
    else {
7173 7174 7175 7176 7177
        //
        // first if-clause ensures that start_key is non-NULL
        //
        assert(start_key);
        end_rows = total_rows_estimate;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7178 7179
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7180
    rows = (end_rows > start_rows) ? end_rows - start_rows : 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7181

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7182
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7183 7184
    // MySQL thinks a return value of 0 means there are exactly 0 rows
    // Therefore, always return non-zero so this assumption is not made
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7185
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7186
    ret_val = (ha_rows) (rows <= 1 ? 1 : rows);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7187
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7188
    DBUG_RETURN(ret_val);
7189 7190
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7191 7192

//
7193 7194 7195
// Initializes the auto-increment data in the local "share" object to the
// greater of two values: what's stored in the metadata or the last inserted
// auto-increment field (if auto-increment field is the first field of a key).
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7196 7197 7198 7199 7200 7201
//
void ha_tokudb::init_auto_increment() {
    DBT key;
    DBT value;
    int error;
    HA_METADATA_KEY key_val = hatoku_max_ai;
7202 7203
    memset(&key, 0, sizeof(key));
    memset(&value, 0, sizeof(value));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7204 7205
    key.data = &key_val;
    key.size = sizeof(key_val);
7206
    value.flags = DB_DBT_USERMEM;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218
    DB_TXN* txn = NULL;

    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) {
        share->last_auto_increment = 0;    
    }
    else {
        //
        // First retrieve hatoku_max_ai, which is max value used by auto increment
        // column so far, the max value could have been auto generated (e.g. insert (NULL))
        // or it could have been manually inserted by user (e.g. insert (345))
        //
7219 7220
        value.ulen = sizeof(share->last_auto_increment);
        value.data = &share->last_auto_increment;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7221 7222 7223 7224 7225 7226 7227 7228
        error = share->status_block->get(
            share->status_block, 
            txn, 
            &key, 
            &value, 
            0
            );
        
7229
        if (error || value.size != sizeof(share->last_auto_increment)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7230 7231
            share->last_auto_increment = 0;
        }
7232

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7233 7234 7235 7236 7237 7238
        //
        // Now retrieve the initial auto increment value, as specified by create table
        // so if a user does "create table t1 (a int auto_increment, primary key (a)) auto_increment=100",
        // then the value 100 should be stored here
        //
        key_val = hatoku_ai_create_value;
7239 7240
        value.ulen = sizeof(share->auto_inc_create_value);
        value.data = &share->auto_inc_create_value;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7241 7242 7243 7244 7245 7246 7247 7248
        error = share->status_block->get(
            share->status_block, 
            txn, 
            &key, 
            &value, 
            0
            );
        
7249
        if (error || value.size != sizeof(share->auto_inc_create_value)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7250 7251 7252
            share->auto_inc_create_value = 0;
        }

7253
        commit_txn(txn, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7254 7255 7256 7257 7258 7259
    }
    if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT) {
        TOKUDB_TRACE("init auto increment:%lld\n", share->last_auto_increment);
    }
}

7260
void ha_tokudb::get_auto_increment(ulonglong offset, ulonglong increment, ulonglong nb_desired_values, ulonglong * first_value, ulonglong * nb_reserved_values) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7261
    TOKUDB_DBUG_ENTER("ha_tokudb::get_auto_increment");
7262 7263
    ulonglong nr;

7264
    pthread_mutex_lock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7265 7266 7267 7268

    if (share->auto_inc_create_value > share->last_auto_increment) {
        nr = share->auto_inc_create_value;
        share->last_auto_increment = share->auto_inc_create_value;
7269
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7270 7271 7272 7273
    else {
        nr = share->last_auto_increment + increment;
    }
    share->last_auto_increment = nr + (nb_desired_values - 1)*increment;
7274 7275
    if (delay_updating_ai_metadata) {
        ai_metadata_update_required = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7276 7277 7278 7279
    }
    else {
        update_max_auto_inc(share->status_block, share->last_auto_increment);
    }
7280

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7281
    if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT) {
7282 7283
        TOKUDB_TRACE("get_auto_increment(%lld,%lld,%lld):got:%lld:%lld\n",
                     offset, increment, nb_desired_values, nr, nb_desired_values);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7284
    }
7285
    *first_value = nr;
7286
    *nb_reserved_values = nb_desired_values;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7287
    pthread_mutex_unlock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7288
    DBUG_VOID_RETURN;
7289 7290
}

7291 7292 7293 7294
bool ha_tokudb::is_optimize_blocking() {
    return false;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7295 7296 7297 7298
bool ha_tokudb::is_auto_inc_singleton(){
    return false;
}

7299 7300
volatile int ha_tokudb_tokudb_add_index_wait = 0; // debug

7301 7302 7303 7304
//
// Internal function called by ha_tokudb::add_index and ha_tokudb::alter_table_phase2
// With a transaction, drops dictionaries associated with indexes in key_num
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7305 7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316
//
// Adds indexes to the table. Takes the array of KEY passed in key_info, and creates
// DB's that will go at the end of share->key_file. THE IMPLICIT ASSUMPTION HERE is
// that the table will be modified and that these added keys will be appended to the end
// of the array table->key_info
// Parameters:
//      [in]    table_arg - table that is being modified, seems to be identical to this->table
//      [in]    key_info - array of KEY's to be added
//              num_of_keys - number of keys to be added, number of elements in key_info
//  Returns:
//      0 on success, error otherwise
//
7317 7318 7319 7320 7321 7322 7323 7324 7325 7326
int ha_tokudb::tokudb_add_index(
    TABLE *table_arg, 
    KEY *key_info, 
    uint num_of_keys, 
    DB_TXN* txn, 
    bool* inc_num_DBs,
    bool* modified_DBs
    ) 
{
    TOKUDB_DBUG_ENTER("ha_tokudb::tokudb_add_index");
7327 7328
    assert(txn);

7329 7330
    while (ha_tokudb_tokudb_add_index_wait) sleep(1); // debug

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7331 7332 7333 7334
    int error;
    uint curr_index = 0;
    DBC* tmp_cursor = NULL;
    int cursor_ret_val = 0;
7335
    DBT curr_pk_key, curr_pk_val;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7336
    THD* thd = ha_thd(); 
7337
    DB_LOADER* loader = NULL;
7338
    DB_INDEXER* indexer = NULL;
7339
    bool loader_use_puts = get_load_save_space(thd);
7340
    bool use_hot_index = (lock.type == TL_WRITE_ALLOW_WRITE);
Yoni Fogel's avatar
Yoni Fogel committed
7341 7342 7343 7344 7345
    uint32_t loader_flags = loader_use_puts ? LOADER_USE_PUTS : 0;
    uint32_t indexer_flags = 0;
    uint32_t mult_db_flags[MAX_KEY + 1] = {0};
    uint32_t mult_put_flags[MAX_KEY + 1];
    uint32_t mult_dbt_flags[MAX_KEY + 1];
7346
    bool creating_hot_index = false;
7347 7348
    struct loader_context lc;
    memset(&lc, 0, sizeof lc);
7349 7350 7351
    lc.thd = thd;
    lc.ha = this;
    loader_error = 0;
7352
    bool rw_lock_taken = false;
7353 7354
    *inc_num_DBs = false;
    *modified_DBs = false;
7355 7356
    invalidate_bulk_fetch();
    unpack_entire_row = true; // for bulk fetching rows
Yoni Fogel's avatar
Yoni Fogel committed
7357
    for (uint32_t i = 0; i < MAX_KEY+1; i++) {
7358
        mult_put_flags[i] = 0;
7359 7360
        mult_dbt_flags[i] = DB_DBT_REALLOC;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7361 7362 7363 7364
    //
    // number of DB files we have open currently, before add_index is executed
    //
    uint curr_num_DBs = table_arg->s->keys + test(hidden_primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7365

7366 7367 7368
    //
    // get the row type to use for the indexes we're adding
    //
Leif Walsh's avatar
Leif Walsh committed
7369
    const enum row_type row_type = get_row_type_for_key(share->file);
7370

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7371 7372 7373 7374 7375 7376
    //
    // status message to be shown in "show process list"
    //
    char status_msg[MAX_ALIAS_NAME + 200]; //buffer of 200 should be a good upper bound.
    ulonglong num_processed = 0; //variable that stores number of elements inserted thus far
    thd_proc_info(thd, "Adding indexes");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7377 7378 7379 7380 7381
    
    //
    // in unpack_row, MySQL passes a buffer that is this long,
    // so this length should be good enough for us as well
    //
7382 7383
    memset((void *) &curr_pk_key, 0, sizeof(curr_pk_key));
    memset((void *) &curr_pk_val, 0, sizeof(curr_pk_val));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7384 7385 7386 7387 7388 7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399

    //
    // The files for secondary tables are derived from the name of keys
    // If we try to add a key with the same name as an already existing key,
    // We can crash. So here we check if any of the keys added has the same
    // name of an existing key, and if so, we fail gracefully
    //
    for (uint i = 0; i < num_of_keys; i++) {
        for (uint j = 0; j < table_arg->s->keys; j++) {
            if (strcmp(key_info[i].name, table_arg->s->key_info[j].name) == 0) {
                error = HA_ERR_WRONG_COMMAND;
                goto cleanup;
            }
        }
    }
    
7400 7401
    rw_wrlock(&share->num_DBs_lock);
    rw_lock_taken = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7402 7403 7404 7405
    //
    // open all the DB files and set the appropriate variables in share
    // they go to the end of share->key_file
    //
7406
    creating_hot_index = use_hot_index && num_of_keys == 1 && (key_info[0].flags & HA_NOSAME) == 0;
7407 7408 7409 7410 7411 7412 7413
    if (use_hot_index && (share->num_DBs > curr_num_DBs)) {
        //
        // already have hot index in progress, get out
        //
        error = HA_ERR_INTERNAL_ERROR;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7414
    curr_index = curr_num_DBs;
7415
    *modified_DBs = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7416
    for (uint i = 0; i < num_of_keys; i++, curr_index++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7417 7418
        if (key_info[i].flags & HA_CLUSTERING) {
            set_key_filter(
7419
                &share->kc_info.key_filters[curr_index],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7420 7421 7422 7423 7424 7425
                &key_info[i],
                table_arg,
                false
                );                
            if (!hidden_primary_key) {
                set_key_filter(
7426
                    &share->kc_info.key_filters[curr_index],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7427 7428 7429 7430 7431 7432
                    &table_arg->key_info[primary_key],
                    table_arg,
                    false
                    );
            }

7433
            error = initialize_col_pack_info(&share->kc_info,table_arg->s,curr_index);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7434 7435 7436 7437 7438 7439
            if (error) {
                goto cleanup;
            }
        }


7440
        error = create_secondary_dictionary(share->table_name, table_arg, &key_info[i], txn, &share->kc_info, curr_index, creating_hot_index, row_type);
7441 7442
        if (error) { goto cleanup; }

7443
        error = open_secondary_dictionary(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7444 7445 7446
            &share->key_file[curr_index], 
            &key_info[i],
            share->table_name,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7447
            false,
7448
            txn
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7449 7450 7451 7452
            );
        if (error) { goto cleanup; }
    }
    
7453
    if (creating_hot_index) {
7454
        share->num_DBs++;
7455
        *inc_num_DBs = true;
7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466
        error = db_env->create_indexer(
            db_env,
            txn,
            &indexer,
            share->file,
            num_of_keys,
            &share->key_file[curr_num_DBs],
            mult_db_flags,
            indexer_flags
            );
        if (error) { goto cleanup; }
7467

7468 7469
        error = indexer->set_poll_function(indexer, ai_poll_fun, &lc);
        if (error) { goto cleanup; }
7470

7471 7472
        error = indexer->set_error_callback(indexer, loader_ai_err_fun, &lc);
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7473

7474 7475 7476
        rw_unlock(&share->num_DBs_lock);
        rw_lock_taken = false;
        
7477 7478 7479 7480 7481 7482
#ifdef HA_TOKUDB_HAS_THD_PROGRESS
        // initialize a one phase progress report.
        // incremental reports are done in the indexer's callback function.
        thd_progress_init(thd, 1);
#endif

7483
        error = indexer->build(indexer);
7484

7485
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7486

7487
        rw_wrlock(&share->num_DBs_lock);
7488
        error = indexer->close(indexer);
7489
        rw_unlock(&share->num_DBs_lock);
7490 7491
        if (error) { goto cleanup; }
        indexer = NULL;
7492
    }
7493 7494 7495
    else {
        rw_unlock(&share->num_DBs_lock);
        rw_lock_taken = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7496 7497
        prelocked_right_range_size = 0;
        prelocked_left_range_size = 0;
7498 7499 7500 7501 7502
        struct smart_dbt_bf_info bf_info;
        bf_info.ha = this;
        // you need the val if you have a clustering index and key_read is not 0;
        bf_info.direction = 1;
        bf_info.thd = ha_thd();
Yoni Fogel's avatar
Yoni Fogel committed
7503
        bf_info.need_val = true;
7504

7505 7506 7507 7508 7509 7510 7511 7512 7513 7514 7515 7516
        error = db_env->create_loader(
            db_env, 
            txn, 
            &loader, 
            NULL, // no src_db needed
            num_of_keys, 
            &share->key_file[curr_num_DBs], 
            mult_put_flags,
            mult_dbt_flags,
            loader_flags
            );
        if (error) { goto cleanup; }
7517

7518
        error = loader->set_poll_function(loader, loader_poll_fun, &lc);
7519
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7520

7521 7522 7523 7524 7525
        error = loader->set_error_callback(loader, loader_ai_err_fun, &lc);
        if (error) { goto cleanup; }
        //
        // scan primary table, create each secondary key, add to each DB
        //    
7526
        if ((error = share->file->cursor(share->file, txn, &tmp_cursor, DB_SERIALIZABLE))) {
7527
            tmp_cursor = NULL;             // Safety
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7528 7529
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7530

7531 7532 7533 7534 7535
        //
        // grab some locks to make this go faster
        // first a global read lock on the main DB, because
        // we intend to scan the entire thing
        //
7536 7537 7538 7539 7540
        error = tmp_cursor->c_pre_acquire_range_lock(
            tmp_cursor,
            share->file->dbt_neg_infty(),
            share->file->dbt_pos_infty()
            );
7541
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7542

7543 7544 7545 7546 7547
        // set the bulk fetch iteration to its max so that adding an
        // index fills the bulk fetch buffer every time. we do not
        // want it to grow exponentially fast.
        rows_fetched_using_bulk_fetch = 0;
        bulk_fetch_iteration = HA_TOKU_BULK_FETCH_ITERATION_MAX;
7548
        cursor_ret_val = tmp_cursor->c_getf_next(tmp_cursor, DB_PRELOCKED,smart_dbt_bf_callback, &bf_info);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7549

7550 7551 7552 7553 7554 7555
#ifdef HA_TOKUDB_HAS_THD_PROGRESS
        // initialize a two phase progress report.
        // first phase: putting rows into the loader
        thd_progress_init(thd, 2);
#endif

7556 7557
        while (cursor_ret_val != DB_NOTFOUND || ((bytes_used_in_range_query_buff - curr_range_query_buff_offset) > 0)) {
            if ((bytes_used_in_range_query_buff - curr_range_query_buff_offset) == 0) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7558
                invalidate_bulk_fetch(); // reset the buffers
7559 7560 7561 7562 7563
                cursor_ret_val = tmp_cursor->c_getf_next(tmp_cursor, DB_PRELOCKED, smart_dbt_bf_callback, &bf_info);
                if (cursor_ret_val != DB_NOTFOUND && cursor_ret_val != 0) {
                    error = cursor_ret_val;
                    goto cleanup;
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7564
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7565 7566
            // do this check in case the the c_getf_next did not put anything into the buffer because
            // there was no more data
7567 7568 7569
            if ((bytes_used_in_range_query_buff - curr_range_query_buff_offset) == 0) {
                break;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7570
            // at this point, we know the range query buffer has at least one key/val pair
7571 7572
            uchar* curr_pos = range_query_buff+curr_range_query_buff_offset;
            
Yoni Fogel's avatar
Yoni Fogel committed
7573
            uint32_t key_size = *(uint32_t *)curr_pos;    
7574 7575 7576 7577 7578 7579
            curr_pos += sizeof(key_size);    
            uchar* curr_key_buff = curr_pos;    
            curr_pos += key_size;        
            curr_pk_key.data = curr_key_buff;    
            curr_pk_key.size = key_size;
            
Yoni Fogel's avatar
Yoni Fogel committed
7580
            uint32_t val_size = *(uint32_t *)curr_pos;    
7581 7582 7583 7584 7585 7586 7587
            curr_pos += sizeof(val_size);    
            uchar* curr_val_buff = curr_pos;    
            curr_pos += val_size;        
            curr_pk_val.data = curr_val_buff;    
            curr_pk_val.size = val_size;
            
            curr_range_query_buff_offset = curr_pos - range_query_buff;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7588

7589 7590
            error = loader->put(loader, &curr_pk_key, &curr_pk_val);
            if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7591

7592 7593 7594 7595 7596 7597 7598 7599 7600 7601
            num_processed++; 

            if ((num_processed % 1000) == 0) {
                if (loader_use_puts) {
                    sprintf(status_msg, "Adding indexes: Processed %llu of about %llu rows.", num_processed, (long long unsigned) share->rows);
                }
                else {
                    sprintf(status_msg, "Adding indexes: Fetched %llu of about %llu rows, loading of data still remains.", num_processed, (long long unsigned) share->rows);
                }
                thd_proc_info(thd, status_msg);
7602 7603 7604 7605 7606

#ifdef HA_TOKUDB_HAS_THD_PROGRESS
                thd_progress_report(thd, num_processed, (long long unsigned) share->rows);
#endif

7607 7608 7609 7610 7611 7612 7613 7614 7615 7616
                if (thd->killed) {
                    error = ER_ABORTING_CONNECTION;
                    goto cleanup;
                }
            }
        }
        error = tmp_cursor->c_close(tmp_cursor);
        assert(error==0);
        tmp_cursor = NULL;

7617 7618 7619 7620 7621 7622
#ifdef HA_TOKUDB_HAS_THD_PROGRESS
        // next progress report phase: closing the loader. 
        // incremental reports are done in the loader's callback function.
        thd_progress_next_stage(thd);
#endif

7623 7624
        error = loader->close(loader);
        loader = NULL;
7625

7626 7627
        if (error) goto cleanup;
    }
7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638 7639 7640 7641
    curr_index = curr_num_DBs;
    for (uint i = 0; i < num_of_keys; i++, curr_index++) {
        if (key_info[i].flags & HA_NOSAME) {
            bool is_unique;
            error = is_index_unique(
                &is_unique, 
                txn, 
                share->key_file[curr_index], 
                &key_info[i]
                );
            if (error) goto cleanup;
            if (!is_unique) {
                error = HA_ERR_FOUND_DUPP_KEY;
                last_dup_key = i;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7642 7643 7644 7645 7646
                goto cleanup;
            }
        }
    }

7647 7648 7649
    //
    // We have an accurate row count, might as well update share->rows
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7650 7651 7652 7653 7654
    if(!creating_hot_index) {
        pthread_mutex_lock(&share->mutex);
        share->rows = num_processed;
        pthread_mutex_unlock(&share->mutex);
    }
7655 7656 7657 7658 7659 7660 7661 7662
    //
    // now write stuff to status.tokudb
    //
    pthread_mutex_lock(&share->mutex);
    for (uint i = 0; i < num_of_keys; i++) {
        write_key_name_to_status(share->status_block, key_info[i].name, txn);
    }
    pthread_mutex_unlock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7663 7664 7665
    
    error = 0;
cleanup:
7666 7667 7668
#ifdef HA_TOKUDB_HAS_THD_PROGRESS
    thd_progress_end(thd);
#endif
7669 7670 7671 7672
    if (rw_lock_taken) {
        rw_unlock(&share->num_DBs_lock);
        rw_lock_taken = false;
    }
7673
    if (tmp_cursor) {            
7674 7675
        int r = tmp_cursor->c_close(tmp_cursor);
        assert(r==0);
7676 7677
        tmp_cursor = NULL;
    }
7678
    if (loader != NULL) {
7679 7680
        sprintf(status_msg, "aborting creation of indexes.");
        thd_proc_info(thd, status_msg);
7681 7682
        loader->abort(loader);
    }
7683 7684 7685
    if (indexer != NULL) {
        sprintf(status_msg, "aborting creation of indexes.");
        thd_proc_info(thd, status_msg);
7686
        rw_wrlock(&share->num_DBs_lock);
7687
        indexer->abort(indexer);
7688
        rw_unlock(&share->num_DBs_lock);
7689
    }
7690 7691
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
        sql_print_error("Could not add indexes to table %s because \
7692 7693
another transaction has accessed the table. \
To add indexes, make sure no transactions touch the table.", share->table_name);
7694
    }
7695
    TOKUDB_DBUG_RETURN(error ? error : loader_error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7696 7697
}

7698 7699 7700 7701
//
// Internal function called by ha_tokudb::add_index and ha_tokudb::alter_table_phase2
// Closes added indexes in case of error in error path of add_index and alter_table_phase2
//
7702 7703 7704 7705 7706 7707 7708 7709 7710 7711 7712 7713 7714 7715 7716 7717 7718 7719 7720 7721 7722 7723 7724 7725 7726 7727 7728 7729 7730 7731 7732 7733 7734
void ha_tokudb::restore_add_index(TABLE* table_arg, uint num_of_keys, bool incremented_numDBs, bool modified_DBs) {
    uint curr_num_DBs = table_arg->s->keys + test(hidden_primary_key);
    uint curr_index = 0;

    //
    // need to restore num_DBs, and we have to do it before we close the dictionaries
    // so that there is not a window 
    //
    if (incremented_numDBs) {
        rw_wrlock(&share->num_DBs_lock);
        share->num_DBs--;
    }
    if (modified_DBs) {
        curr_index = curr_num_DBs;
        for (uint i = 0; i < num_of_keys; i++, curr_index++) {
            reset_key_and_col_info(&share->kc_info, curr_index);
        }
        curr_index = curr_num_DBs;
        for (uint i = 0; i < num_of_keys; i++, curr_index++) {
            if (share->key_file[curr_index]) {
                int r = share->key_file[curr_index]->close(
                    share->key_file[curr_index],
                    0
                    );
                assert(r==0);
                share->key_file[curr_index] = NULL;
            }
        }
    }
    if (incremented_numDBs) {
        rw_unlock(&share->num_DBs_lock);
    }
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7735

7736 7737
volatile int ha_tokudb_drop_indexes_wait = 0; // debug

7738 7739 7740 7741
//
// Internal function called by ha_tokudb::prepare_drop_index and ha_tokudb::alter_table_phase2
// With a transaction, drops dictionaries associated with indexes in key_num
//
7742
int ha_tokudb::drop_indexes(TABLE *table_arg, uint *key_num, uint num_of_keys, KEY *key_info, DB_TXN* txn) {
7743
    TOKUDB_DBUG_ENTER("ha_tokudb::drop_indexes");
7744 7745
    assert(txn);

7746
    while (ha_tokudb_drop_indexes_wait) sleep(1); // debug
7747

7748
    int error = 0;
7749 7750 7751 7752 7753 7754 7755
    for (uint i = 0; i < num_of_keys; i++) {
        uint curr_index = key_num[i];
        error = share->key_file[curr_index]->pre_acquire_fileops_lock(share->key_file[curr_index],txn);
        if (error != 0) {
            goto cleanup;
        }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7756 7757
    for (uint i = 0; i < num_of_keys; i++) {
        uint curr_index = key_num[i];
7758 7759
        int r = share->key_file[curr_index]->close(share->key_file[curr_index],0);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7760 7761
        share->key_file[curr_index] = NULL;

7762
        error = remove_key_name_from_status(share->status_block, key_info[curr_index].name, txn);
7763 7764
        if (error) { goto cleanup; }
        
7765
        error = delete_or_rename_dictionary(share->table_name, NULL, key_info[curr_index].name, true, txn, true);
7766
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7767
    }
7768

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7769
cleanup:
7770 7771
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
        sql_print_error("Could not drop indexes from table %s because \
7772 7773
another transaction has accessed the table. \
To drop indexes, make sure no transactions touch the table.", share->table_name);
7774
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7775 7776 7777
    TOKUDB_DBUG_RETURN(error);
}

7778 7779 7780 7781
//
// Internal function called by ha_tokudb::prepare_drop_index and ha_tokudb::alter_table_phase2
// Restores dropped indexes in case of error in error path of prepare_drop_index and alter_table_phase2
//
7782
void ha_tokudb::restore_drop_indexes(TABLE *table_arg, uint *key_num, uint num_of_keys) {
7783

7784 7785 7786 7787 7788 7789 7790 7791 7792 7793 7794 7795 7796 7797 7798 7799 7800 7801
    //
    // reopen closed dictionaries
    //
    for (uint i = 0; i < num_of_keys; i++) {
        int r;
        uint curr_index = key_num[i];
        if (share->key_file[curr_index] == NULL) {
            r = open_secondary_dictionary(
                &share->key_file[curr_index], 
                &table_share->key_info[curr_index],
                share->table_name,
                false, // 
                NULL
                );
            assert(!r);
        }
    }            
}
7802

7803
void ha_tokudb::print_error(int error, myf errflag) {
7804
    if (error == DB_LOCK_DEADLOCK)
7805
        error = HA_ERR_LOCK_DEADLOCK;
7806
    if (error == DB_LOCK_NOTGRANTED)
7807
        error = HA_ERR_LOCK_WAIT_TIMEOUT;
7808
#if defined(HA_ERR_DISK_FULL)
7809 7810 7811
    if (error == ENOSPC) {
        error = HA_ERR_DISK_FULL;
    }
7812
#endif
7813 7814 7815
    if (error == DB_KEYEXIST) {
        error = HA_ERR_FOUND_DUPP_KEY;
    }
7816
#if defined(HA_ALTER_ERROR)
7817 7818
    if (error == HA_ALTER_ERROR) {
        error = HA_ERR_UNSUPPORTED;
7819 7820
    }
#endif
7821 7822
    // TODO: should rename debug code to something better
    // just reusing this so that tests don' start complaining
7823
#if MYSQL_VERSION_ID < 50500
7824
    if ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0) {
7825
        THD* thd = ha_thd();
7826 7827 7828
        if (get_log_client_errors(thd)) {
            sql_print_error("query \"%s\" returned handler error %d", thd->query_string.str, error);
        }
7829
    }
7830
#endif
7831 7832 7833
    handler::print_error(error, errflag);
}

7834

7835 7836 7837 7838 7839 7840 7841 7842
//
// truncate's dictionary associated with keynr index using transaction txn
// does so by deleting and then recreating the dictionary in the context
// of a transaction
//
int ha_tokudb::truncate_dictionary( uint keynr, DB_TXN* txn ) {
    int error;
    bool is_pk = (keynr == primary_key);
7843

7844
    const enum row_type row_type = get_row_type_for_key(share->key_file[keynr]);
7845
    error = share->key_file[keynr]->close(share->key_file[keynr], 0);
7846
    assert(error == 0);
7847 7848 7849 7850 7851 7852 7853 7854 7855

    share->key_file[keynr] = NULL;
    if (is_pk) { share->file = NULL; }

    if (is_pk) {
        error = delete_or_rename_dictionary(
            share->table_name, 
            NULL,
            "main", 
7856
            false, //is_key
7857 7858 7859 7860 7861 7862 7863 7864 7865 7866
            txn,
            true // is a delete
            );
        if (error) { goto cleanup; }
    }
    else {
        error = delete_or_rename_dictionary(
            share->table_name, 
            NULL,
            table_share->key_info[keynr].name, 
7867
            true, //is_key
7868 7869 7870 7871 7872 7873 7874
            txn,
            true // is a delete
            );
        if (error) { goto cleanup; }
    }

    if (is_pk) {
7875
        error = create_main_dictionary(share->table_name, table, txn, &share->kc_info, row_type);
7876 7877 7878 7879 7880 7881
    }
    else {
        error = create_secondary_dictionary(
            share->table_name, 
            table, 
            &table_share->key_info[keynr], 
7882 7883
            txn,
            &share->kc_info,
7884
            keynr,
7885 7886
            false,
            row_type
7887 7888 7889 7890 7891 7892 7893 7894
            );
    }
    if (error) { goto cleanup; }

cleanup:
    return error;
}

7895
volatile int ha_tokudb_truncate_wait = 0; // debug
7896 7897 7898

// for 5.5
int ha_tokudb::truncate() {
7899 7900
    TOKUDB_DBUG_ENTER("truncate");
    while (ha_tokudb_truncate_wait) sleep(1); // debug
7901
    int error = delete_all_rows_internal();
7902
    TOKUDB_DBUG_RETURN(error);
7903 7904
}

7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915
// delete all rows from a table
//
// effects: delete all of the rows in the main dictionary and all of the
// indices.  this must be atomic, so we use the statement transaction
// for all of the truncate operations.
// locks:  if we have an exclusive table write lock, all of the concurrency
// issues go away.
// returns: 0 if success
int ha_tokudb::delete_all_rows() {
    TOKUDB_DBUG_ENTER("delete_all_rows");
    int error = 0;
7916 7917 7918 7919 7920 7921 7922 7923 7924 7925 7926 7927
    if (thd_sql_command(ha_thd()) != SQLCOM_TRUNCATE) {
        share->try_table_lock = true;
        error = HA_ERR_WRONG_COMMAND;
    }
    if (error == 0)
        error = delete_all_rows_internal();
    TOKUDB_DBUG_RETURN(error);
}

int ha_tokudb::delete_all_rows_internal() {
    TOKUDB_DBUG_ENTER("delete_all_rows_internal");
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7928
    uint curr_num_DBs = 0;
7929 7930 7931 7932
    DB_TXN* txn = NULL;

    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }
7933

Zardosht Kasheff's avatar
Zardosht Kasheff committed
7934
    curr_num_DBs = table->s->keys + test(hidden_primary_key);
7935 7936 7937 7938 7939 7940
    for (uint i = 0; i < curr_num_DBs; i++) {
        error = share->key_file[i]->pre_acquire_fileops_lock(
            share->key_file[i], 
            txn
            );
        if (error) { goto cleanup; }
7941 7942 7943 7944 7945
        error = share->key_file[i]->pre_acquire_table_lock(
            share->key_file[i], 
            txn
            );
        if (error) { goto cleanup; }
7946
    }
7947
    for (uint i = 0; i < curr_num_DBs; i++) {
7948 7949
        error = truncate_dictionary(i, txn);
        if (error) { goto cleanup; }
7950 7951 7952
    }

    // zap the row count
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7953
    if (error == 0) {
7954
        share->rows = 0;
7955 7956 7957 7958 7959 7960 7961 7962 7963 7964
	// update auto increment
	share->last_auto_increment = 0;
	// calling write_to_status directly because we need to use txn
	write_to_status(
	    share->status_block,
            hatoku_max_ai,
	    &share->last_auto_increment,
	    sizeof(share->last_auto_increment), 
	    txn
	    );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7965
    }
7966

7967
    share->try_table_lock = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7968
cleanup:
7969 7970 7971 7972 7973 7974 7975 7976 7977
    if (txn) {
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn,0);
        }
    }

7978 7979 7980 7981 7982
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
        sql_print_error("Could not truncate table %s because another transaction has accessed the \
        table. To truncate the table, make sure no transactions touch the table.", 
        share->table_name);
    }
7983 7984 7985 7986 7987 7988 7989 7990 7991 7992 7993
    //
    // regardless of errors, need to reopen the DB's
    //    
    for (uint i = 0; i < curr_num_DBs; i++) {
        int r = 0;
        if (share->key_file[i] == NULL) {
            if (i != primary_key) {
                r = open_secondary_dictionary(
                    &share->key_file[i], 
                    &table_share->key_info[i],
                    share->table_name,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
7994
                    false, // 
7995 7996 7997 7998 7999 8000 8001
                    NULL
                    );
                assert(!r);
            }
            else {
                r = open_main_dictionary(
                    share->table_name, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
8002
                    false, 
8003 8004 8005 8006 8007 8008
                    NULL
                    );
                assert(!r);
            }
        }
    }
8009 8010 8011
    TOKUDB_DBUG_RETURN(error);
}

8012 8013 8014
void ha_tokudb::set_loader_error(int err) {
    loader_error = err;
}
8015

8016 8017 8018 8019 8020
void ha_tokudb::set_dup_value_for_pk(DBT* key) {
    assert(!hidden_primary_key);
    unpack_key(table->record[0],key,primary_key);
    last_dup_key = primary_key;
}
8021

8022 8023
// table admin 
#include "ha_tokudb_admin.cc"
8024

8025
// alter table code for various mysql distros
8026
#include "ha_tokudb_update_fun.cc"
8027 8028 8029
#include "ha_tokudb_alter_51.cc"
#include "ha_tokudb_alter_55.cc"
#include "ha_tokudb_alter_56.cc"
8030 8031 8032 8033 8034 8035

// key comparisons
#include "hatoku_cmp.cc"

// handlerton
#include "hatoku_hton.cc"