ha_tokudb.cc 208 KB
Newer Older
1
#ifdef USE_PRAGMA_IMPLEMENTATION
2
#pragma implementation          // gcc: Class implementation
3 4 5 6
#endif

#define MYSQL_SERVER 1
#include "mysql_priv.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
#undef PACKAGE
#undef VERSION
#undef HAVE_DTRACE
#undef _DTRACE_VERSION

28
//#include "tokudb_config.h"
29 30 31 32 33 34 35

/* 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

Zardosht Kasheff's avatar
Zardosht Kasheff committed
36
#include "hatoku_defines.h"
37
#include "ha_tokudb.h"
Zardosht Kasheff's avatar
Zardosht Kasheff committed
38
#include "hatoku_hton.h"
39 40
#include <mysql/plugin.h>

Zardosht Kasheff's avatar
Zardosht Kasheff committed
41 42 43 44
static const char *ha_tokudb_exts[] = {
    ha_tokudb_ext,
    NullS
};
45 46

#define lockretryN(N) \
47
  for (ulonglong lockretrycount=0; lockretrycount<(N/(1<<3) + 1); lockretrycount++)
48 49

#define lockretry_wait \
50 51 52 53
        if (error != DB_LOCK_NOTGRANTED) { \
            break;  \
        } \
        if (tokudb_debug & TOKUDB_DEBUG_LOCKRETRY) { \
54
	  TOKUDB_TRACE("%s count=%d\n", __FUNCTION__, (int) lockretrycount); \
55
        } \
56 57 58 59 60 61 62
        if (lockretrycount%200 == 0) { \
            if (ha_thd()->killed) { \
                error = DB_LOCK_NOTGRANTED; \
                break; \
            } \
        } \
        usleep((lockretrycount<4 ? (1<<lockretrycount) : (1<<3)) * 1024); \
63

Zardosht Kasheff's avatar
Zardosht Kasheff committed
64 65 66
//
// This offset is calculated starting from AFTER the NULL bytes
//
67
inline u_int32_t get_var_len_offset(KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
68 69
    uint offset = 0;
    for (uint i = 0; i < table_share->fields; i++) {
70 71
        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
72 73 74 75 76 77
        }
    }
    return offset;
}


78
inline u_int32_t get_len_of_offsets(KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
79 80
    uint len = 0;
    for (uint i = 0; i < table_share->fields; i++) {
81 82
        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
83 84 85 86 87 88
        }
    }
    return len;
}


89
static int allocate_key_and_col_info ( TABLE_SHARE* table_share, KEY_AND_COL_INFO* kc_info) {
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108
    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
    //
109
    kc_info->field_lengths = (u_int16_t *)my_malloc(table_share->fields*sizeof(u_int16_t), MYF(MY_WME | MY_ZEROFILL));
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130
    kc_info->length_bytes= (uchar *)my_malloc(table_share->fields, MYF(MY_WME | MY_ZEROFILL));
    kc_info->blob_fields= (u_int32_t *)my_malloc(table_share->fields*sizeof(u_int32_t), MYF(MY_WME | MY_ZEROFILL));
    
    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;
}

131 132 133 134 135
/** @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.
*/
136
static TOKUDB_SHARE *get_share(const char *table_name, TABLE_SHARE* table_share) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
137 138
    TOKUDB_SHARE *share = NULL;
    int error = 0;
139
    uint length;
140

141 142 143
    pthread_mutex_lock(&tokudb_mutex);
    length = (uint) strlen(table_name);

144
    if (!(share = (TOKUDB_SHARE *) my_hash_search(&tokudb_open_tables, (uchar *) table_name, length))) {
145 146
        char *tmp_name;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
147 148 149 150
        //
        // create share and fill it with all zeroes
        // hence, all pointers are initialized to NULL
        //
151 152 153 154 155
        if (!(share = (TOKUDB_SHARE *) 
            my_multi_malloc(MYF(MY_WME | MY_ZEROFILL), 
                            &share, sizeof(*share),
                            &tmp_name, length + 1, 
                            NullS))) {
156 157 158 159 160 161 162
            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);
163

164 165
        error = allocate_key_and_col_info(table_share, &share->kc_info);
        if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
166 167 168
            goto exit;
        }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
169
        bzero((void *) share->key_file, sizeof(share->key_file));
170

Zardosht Kasheff's avatar
Zardosht Kasheff committed
171 172 173 174
        error = my_hash_insert(&tokudb_open_tables, (uchar *) share);
        if (error) {
            goto exit;
        }
175 176
        thr_lock_init(&share->lock);
        pthread_mutex_init(&share->mutex, MY_MUTEX_INIT_FAST);
177
    }
178
    pthread_mutex_unlock(&tokudb_mutex);
179

Zardosht Kasheff's avatar
Zardosht Kasheff committed
180 181 182 183 184 185
exit:
    if (error) {
        pthread_mutex_destroy(&share->mutex);
        my_free((uchar *) share, MYF(0));
        share = NULL;
    }
186
    return share;
187 188
}

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203

void free_key_and_col_info (KEY_AND_COL_INFO* kc_info) {
    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));
    }
    
    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));
}

204
static int free_share(TOKUDB_SHARE * share, bool mutex_is_locked) {
205 206 207 208 209 210 211 212 213
    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
214 215 216 217 218 219 220 221 222 223 224
        //
        // 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);
225
                assert(error == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
226 227 228 229 230
                if (error) {
                    result = error;
                }
                share->key_file[i] = NULL;
            }
231
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
232

233
        free_key_and_col_info(&share->kc_info);
234

Zardosht Kasheff's avatar
Zardosht Kasheff committed
235
        if (share->status_block && (error = share->status_block->close(share->status_block, 0))) {
236
            assert(error == 0);
237
            result = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
238 239
        }
        
240

241
        my_hash_delete(&tokudb_open_tables, (uchar *) share);
242 243 244 245 246 247 248
        thr_lock_delete(&share->lock);
        pthread_mutex_destroy(&share->mutex);
        my_free((uchar *) share, MYF(0));
    }
    pthread_mutex_unlock(&tokudb_mutex);

    return result;
249 250
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
251

Zardosht Kasheff's avatar
Zardosht Kasheff committed
252
#define HANDLE_INVALID_CURSOR() \
Zardosht Kasheff's avatar
Zardosht Kasheff committed
253 254 255 256 257
    if (cursor == NULL) { \
        error = last_cursor_error; \
        goto cleanup; \
    }

258 259


Zardosht Kasheff's avatar
Zardosht Kasheff committed
260 261 262
/* 
 *  returns NULL terminated file extension string
 */
263
const char **ha_tokudb::bas_ext() const {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
264 265
    TOKUDB_DBUG_ENTER("ha_tokudb::bas_ext");
    DBUG_RETURN(ha_tokudb_exts);
266 267
}

268 269 270 271 272 273 274 275
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) {
276
    return thd->lex->duplicates == DUP_REPLACE;
277 278
}

279
static inline bool do_ignore_flag_optimization(THD* thd, TABLE* table, bool opt_eligible) {
280 281
    uint pk_insert_mode = get_pk_insert_mode(thd);
    return ( 
282
        opt_eligible && 
283
        (is_replace_into(thd) || is_insert_ignore(thd)) && 
284 285 286 287 288
        ((!table->triggers && pk_insert_mode < 2) || pk_insert_mode == 0)
        );
}

ulonglong ha_tokudb::table_flags() const {
289
    return (table && do_ignore_flag_optimization(ha_thd(), table, share->replace_into_fast) ? 
290 291 292 293
        int_table_flags | HA_BINLOG_STMT_CAPABLE : 
        int_table_flags | HA_BINLOG_ROW_CAPABLE | HA_BINLOG_STMT_CAPABLE);
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
294 295 296 297
//
// Returns a bit mask of capabilities of the key or its part specified by 
// the arguments. The capabilities are defined in sql/handler.h.
//
298
ulong ha_tokudb::index_flags(uint idx, uint part, bool all_parts) const {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
299
    TOKUDB_DBUG_ENTER("ha_tokudb::index_flags");
300
    assert(table_share);
301
    ulong flags = (HA_READ_NEXT | HA_READ_PREV | HA_READ_ORDER | HA_KEYREAD_ONLY | HA_READ_RANGE);
302 303 304
    if (table_share->key_info[idx].flags & HA_CLUSTERING) {
        flags |= HA_CLUSTERED_INDEX;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
305
    DBUG_RETURN(flags);
306 307
}

Yoni Fogel's avatar
Yoni Fogel committed
308

Zardosht Kasheff's avatar
Zardosht Kasheff committed
309 310 311 312 313 314 315 316 317 318
//
// 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;

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;

325 326 327 328 329
typedef struct row_buffers {
    uchar** key_buff;
    uchar** rec_buff;
} *ROW_BUFFERS;

330 331 332 333 334 335 336

int poll_fun(void *extra, float progress) {
    LOADER_CONTEXT context = (LOADER_CONTEXT)extra;
    if (context->thd->killed) {
        sprintf(context->write_status_msg, "The process has been killed, aborting bulk load.");
        return 1;
    }
337
    sprintf(context->write_status_msg, "Loading of data about %.1f%% done", progress*100);
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355
    thd_proc_info(context->thd, context->write_status_msg);
    return 0;
}


void loader_ai_err_fun(DB *db, int i, int err, DBT *key, DBT *val, void *error_extra) {
    LOADER_CONTEXT context = (LOADER_CONTEXT)error_extra;
    assert(context->ha);
    context->ha->set_loader_error(err);
}

void loader_dup_fun(DB *db, int i, int err, DBT *key, DBT *val, void *error_extra) {
    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
356 357
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
358 359 360 361 362 363 364
//
// 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
//
365
static int smart_dbt_do_nothing (DBT const *key, DBT  const *row, void *context) {
Barry Perlman's avatar
Barry Perlman committed
366
  return 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
367 368
}

369 370 371
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));
372 373 374 375
    if (val->data == NULL) return ENOMEM;
    memcpy(val->data, row->data, row->size);
    val->size = row->size;
    return 0;
376 377
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
378

379 380 381
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
382
    info->ha->extract_hidden_primary_key(info->keynr, key);
383
    return info->ha->read_row_callback(info->buf,info->keynr,row,key);
384 385
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
386 387 388
//
// Smart DBT callback function in case where we have a covering index
//
Barry Perlman's avatar
Barry Perlman committed
389 390
static int
smart_dbt_callback_keyread(DBT const *key, DBT  const *row, void *context) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
391
    SMART_DBT_INFO info = (SMART_DBT_INFO)context;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
392 393
    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
394
    return 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
395 396 397 398 399
}

//
// Smart DBT callback function in case where we do NOT have a covering index
//
Barry Perlman's avatar
Barry Perlman committed
400 401
static int
smart_dbt_callback_rowread(DBT const *key, DBT  const *row, void *context) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
402
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
403
    SMART_DBT_INFO info = (SMART_DBT_INFO)context;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
404
    info->ha->extract_hidden_primary_key(info->keynr, key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
405 406
    error = info->ha->read_primary_key(info->buf,info->keynr,row,key);
    return error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
407 408
}

409 410 411 412 413 414 415 416 417 418 419 420 421
//
// 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
422 423 424 425 426 427 428 429
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;
}


430 431 432 433 434 435 436 437 438 439 440 441 442
//
// 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
443 444 445 446 447
//
// 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 ) 
448
#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
449 450


Zardosht Kasheff's avatar
Zardosht Kasheff committed
451 452 453 454
//
// macro that modifies read flag for cursor operations depending on whether
// we have preacquired lock or not
//
455
#define SET_READ_FLAG(flg) ((range_lock_grabbed) ? ((flg) | DB_PRELOCKED) : (flg))
Zardosht Kasheff's avatar
Zardosht Kasheff committed
456

Zardosht Kasheff's avatar
Zardosht Kasheff committed
457 458 459 460 461 462 463 464 465 466 467 468 469

//
// 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
470
    const uchar *key;     /* Key */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
471 472
    ulonglong   unsigned_autoinc = 0;  /* Unsigned auto-increment */
    longlong      signed_autoinc = 0;  /* Signed auto-increment */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516
    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
517
    break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552

    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
553 554 555
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
556 557 558

inline bool
is_null_field( TABLE* table, Field* field, const uchar* record) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
559
    uint null_offset;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
560
    bool ret_val;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
561 562
    if (!field->null_ptr) {
        ret_val = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
563
        goto exitpt;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
564 565
    }
    null_offset = get_null_offset(table,field);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
566 567 568
    ret_val = (record[null_offset] & field->null_bit) ? true: false;

exitpt:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
569 570 571 572 573
    return ret_val;
}

inline ulong field_offset(Field* field, TABLE* table) {
    return((ulong) (field->ptr - table->record[0]));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
574 575
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
576 577 578 579
inline HA_TOKU_ISO_LEVEL tx_to_toku_iso(ulong tx_isolation) {
    if (tx_isolation == ISO_READ_UNCOMMITTED) {
        return hatoku_iso_read_uncommitted;
    }
580 581 582
    else if (tx_isolation == ISO_READ_COMMITTED) {
        return hatoku_iso_read_committed;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
583 584 585 586 587 588 589
    else {
        return hatoku_iso_serializable;
    }
}

inline u_int32_t toku_iso_to_txn_flag (HA_TOKU_ISO_LEVEL lvl) {
    if (lvl == hatoku_iso_read_uncommitted) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
590
        return DB_READ_UNCOMMITTED;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
591
    }
592 593 594
    else if (lvl == hatoku_iso_read_committed) {
        return DB_READ_COMMITTED;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
595 596 597 598 599
    else {
        return 0;
    }
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
600 601


Zardosht Kasheff's avatar
Zardosht Kasheff committed
602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801
int filter_key_part_compare (const void* left, const void* right) {
    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;
                }
            }
        }
    }
}


inline uchar* pack_fixed_field(
    uchar* to_tokudb,
    const uchar* from_mysql,
    u_int32_t num_bytes
    )
{
    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;
}

inline const uchar* unpack_fixed_field(
    uchar* to_mysql,
    const uchar* from_tokudb,
    u_int32_t num_bytes
    )
{
    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;
}


inline uchar* pack_var_field(
    uchar* to_tokudb_offset_ptr, //location where offset data is going to be written
    uchar* to_tokudb_data,
    uchar* to_tokudb_offset_start, //location where offset starts
    const uchar * from_mysql,
    u_int32_t mysql_length_bytes,
    u_int32_t offset_bytes
    )
{
    uint data_length = 0;
    u_int32_t offset = 0;
    switch(mysql_length_bytes) {
    case(1):
        data_length = from_mysql[0];
        break;
    case(2):
        data_length = uint2korr(from_mysql);
        break;
    default:
        assert(false);
        break;
    }
    memcpy(to_tokudb_data, from_mysql + mysql_length_bytes, data_length);
    //
    // for offset, we pack the offset where the data ENDS!
    //
    offset = to_tokudb_data + data_length - to_tokudb_offset_start;
    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;
}

inline void unpack_var_field(
    uchar* to_mysql,
    const uchar* from_tokudb_data,
    u_int32_t from_tokudb_data_len,
    u_int32_t mysql_length_bytes
    )
{
    //
    // 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);
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836
uchar* pack_toku_field_blob(
    uchar* to_tokudb,
    const uchar* from_mysql,
    Field* field
    )
{
    u_int32_t len_bytes = field->row_pack_length();
    u_int32_t length = 0;
    uchar* data_ptr = NULL;
    memcpy(to_tokudb, from_mysql, len_bytes);

    switch (len_bytes) {
    case (1):
        length = (u_int32_t)(*from_mysql);
        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);
}

837

838
static int add_table_to_metadata(const char *name, TABLE* table, DB_TXN* txn) {
839 840 841 842
    int error = 0;
    DBT key;
    DBT val;
    uchar hidden_primary_key = (table->s->primary_key >= MAX_KEY);
843
    assert(txn);
844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860
    
    bzero((void *)&key, sizeof(key));
    bzero((void *)&val, sizeof(val));
    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,
        DB_YESOVERWRITE
        );
    return error;
}

861
static int drop_table_from_metadata(const char *name, DB_TXN* txn) {
862 863 864
    int error = 0;
    DBT key;
    DBT data;
865
    assert(txn);
866 867 868 869 870 871 872 873 874 875 876 877 878
    bzero((void *)&key, sizeof(key));
    bzero((void *)&data, sizeof(data));
    key.data = (void *)name;
    key.size = strlen(name) + 1;
    error = metadata_db->del(
        metadata_db, 
        txn, 
        &key , 
        DB_DELETE_ANY
        );
    return error;
}

879
static int rename_table_in_metadata(const char *from, const char *to, DB_TXN* txn) {
880 881
    int error = 0;
    DBT from_key;
882 883
    DBT to_key;
    DBT val;
884
    assert(txn);
885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903
    
    bzero((void *)&from_key, sizeof(from_key));
    bzero((void *)&to_key, sizeof(to_key));
    bzero((void *)&val, sizeof(val));
    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) {
904
        goto cleanup;
905 906 907 908 909 910 911 912 913
    }

    error = metadata_db->put(
        metadata_db,
        txn,
        &to_key,
        &val,
        DB_YESOVERWRITE
        );
914 915 916
    if (error) {
        goto cleanup;
    }
917

918
    error = metadata_db->del(
919 920 921 922 923
        metadata_db, 
        txn, 
        &from_key, 
        DB_DELETE_ANY
        );
924 925 926
    if (error) {
        goto cleanup;
    }
927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969

    error = 0;

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

    return error;
}


static int check_table_in_metadata(const char *name, bool* table_found) {
    int error = 0;
    DBT key;
    DB_TXN* txn = NULL;
    pthread_mutex_lock(&tokudb_meta_mutex);
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) {
        goto cleanup;
    }
    
    bzero((void *)&key, sizeof(key));
    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;
    }

cleanup:
    if (txn) {
970
        commit_txn(txn, 0);
971 972 973 974 975
    }
    pthread_mutex_unlock(&tokudb_meta_mutex);
    return error;
}

976 977 978 979 980 981 982 983 984 985
int create_tokudb_trx_data_instance(tokudb_trx_data** out_trx) {
    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;
986
    error = 0;
987 988 989 990
cleanup:
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
991 992 993 994 995 996 997 998 999 1000
int generate_row_for_put(
    DB *dest_db, 
    DB *src_db,
    DBT *dest_key, 
    DBT *dest_val,
    const DBT *src_key, 
    const DBT *src_val,
    void *extra
    ) 
{
1001 1002
    int error;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1003 1004 1005 1006 1007
    DB* curr_db = dest_db;
    uchar* row_desc = NULL;
    u_int32_t desc_size;
    uchar* buff = NULL;
    u_int32_t max_key_len = 0;
1008
    
1009
    row_desc = (uchar *)curr_db->descriptor->dbt.data;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
    row_desc += (*(u_int32_t *)row_desc);
    desc_size = (*(u_int32_t *)row_desc) - 4;
    row_desc += 4;
    
    if (is_key_pk(row_desc, desc_size)) {
        assert(dest_key->flags != DB_DBT_USERMEM);
        assert(dest_val->flags != DB_DBT_USERMEM);
        if (dest_key->flags == DB_DBT_REALLOC && dest_key->data != NULL) {
            free(dest_key->data);
        }
        if (dest_val->flags == DB_DBT_REALLOC && dest_val->data != NULL) {
            free(dest_val->data);
        }
        dest_key->data = src_key->data;
        dest_key->size = src_key->size;
1025
        dest_key->flags = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1026 1027
        dest_val->data = src_val->data;
        dest_val->size = src_val->size;
1028
        dest_val->flags = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1029 1030 1031 1032 1033 1034 1035 1036 1037
        error = 0;
        goto cleanup;
    }
    if (dest_key->flags == DB_DBT_USERMEM) {
        buff = (uchar *)dest_key->data;
    }
    else if (dest_key->flags == DB_DBT_REALLOC) {
        max_key_len = max_key_size_from_desc(row_desc, desc_size);
        max_key_len += src_key->size;
1038
        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1039 1040 1041 1042 1043 1044 1045
        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;
1046
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089

        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;
    desc_size = (*(u_int32_t *)row_desc) - 4;
    row_desc += 4;
    if (!is_key_clustering(row_desc, desc_size)) {
        dest_val->size = 0;
    }
    else {
        uchar* buff = NULL;
        if (dest_val->flags == DB_DBT_USERMEM) {
            buff = (uchar *)dest_val->data;
        }
        else if (dest_val->flags == DB_DBT_REALLOC){
            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;
1090
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1091 1092
            buff = (uchar *)dest_val->data;
            assert(buff != NULL);
1093 1094
        }
        else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1095
            assert(false);
1096
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1097 1098 1099 1100 1101 1102 1103
        dest_val->size = pack_clustering_val_from_desc(
            buff,
            row_desc,
            desc_size,
            src_val
            );
        assert(dest_val->ulen >= dest_val->size);
1104 1105 1106
    }

    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1107
cleanup:
1108 1109 1110
    return error;
}

1111

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1112 1113 1114
ha_tokudb::ha_tokudb(handlerton * hton, TABLE_SHARE * table_arg):handler(hton, table_arg) 
    // flags defined in sql\handler.h
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1115
    int_table_flags = HA_REC_NOT_IN_SEQ  | HA_NULL_IN_KEY | HA_CAN_INDEX_BLOBS | HA_PRIMARY_KEY_IN_READ_INDEX | 
1116
                    HA_FILE_BASED | HA_AUTO_PART_KEY | HA_TABLE_SCAN_ON_INDEX;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133
    alloc_ptr = NULL;
    rec_buff = NULL;
    transaction = NULL;
    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
1134 1135
    blob_buff = NULL;
    num_blob_bytes = 0;
1136 1137
    delay_updating_ai_metadata = false;
    ai_metadata_update_required = false;
1138
    read_lock_wait_time = 4000;
1139 1140
    bzero(mult_key_buff, sizeof(mult_key_buff));
    bzero(mult_rec_buff, sizeof(mult_rec_buff));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1141 1142
    bzero(mult_key_dbt, sizeof(mult_key_dbt));
    bzero(mult_rec_dbt, sizeof(mult_rec_dbt));
1143 1144 1145
    loader = NULL;
    abort_loader = false;
    bzero(&lc, sizeof(lc));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171
}

//
// 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;
}

1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197
int ha_tokudb::open_status_dictionary(DB** ptr, const char* name, DB_TXN* txn) {
    int error;
    char* newname = NULL;
    uint open_mode = DB_THREAD;
    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; 
    }
cleanup:
    if (error) {
        if (*ptr) {
1198 1199
            int r = (*ptr)->close(*ptr, 0);
            assert(r==0);
1200 1201 1202 1203 1204 1205 1206
            *ptr = NULL;
        }
    }
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1207
int ha_tokudb::open_main_dictionary(const char* name, bool is_read_only, DB_TXN* txn) {
1208 1209
    int error;    
    char* newname = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1210
    uint open_flags = (is_read_only ? DB_RDONLY : 0) | DB_THREAD;
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229

    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
1230

1231 1232 1233 1234 1235 1236 1237 1238
    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);
    }
1239 1240

    error = 0;
1241 1242 1243
exit:
    if (error) {
        if (share->file) {
1244
            int r = share->file->close(
1245 1246 1247
                share->file,
                0
                );
1248
            assert(r==0);
1249 1250 1251 1252 1253 1254 1255
            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
1256

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1257 1258 1259
//
// 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
1260
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
1261
    int error = ENOSYS;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1262
    char dict_name[MAX_DICT_NAME_LEN];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1263
    uint open_flags = (is_read_only ? DB_RDONLY : 0) | DB_THREAD;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1264
    char* newname = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1265 1266 1267 1268 1269 1270
    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
1271 1272 1273 1274
    if (newname == NULL) {
        error = ENOMEM;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1275
    make_name(newname, name, dict_name);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1276 1277 1278 1279 1280 1281


    if ((error = db_create(ptr, db_env, 0))) {
        my_errno = error;
        goto cleanup;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1282

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1283

1284
    if ((error = (*ptr)->open(*ptr, txn, newname, NULL, DB_BTREE, open_flags, 0))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1285 1286 1287 1288 1289 1290 1291
        my_errno = error;
        goto cleanup;
    }
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
        TOKUDB_TRACE("open:%s:file=%p\n", newname, *ptr);
    }
cleanup:
1292 1293
    if (error) {
        if (*ptr) {
1294 1295
            int r = (*ptr)->close(*ptr, 0);
            assert(r==0);
1296 1297 1298
            *ptr = NULL;
        }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1299
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1300 1301 1302
    return error;
}

1303
int initialize_col_pack_info(KEY_AND_COL_INFO* kc_info, TABLE_SHARE* table_share, uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1304 1305 1306 1307
    int error = ENOSYS;
    //
    // set up the cp_info
    //
1308 1309
    assert(kc_info->cp_info[keynr] == NULL);
    kc_info->cp_info[keynr] = (COL_PACK_INFO *)my_malloc(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1310 1311 1312
        table_share->fields*sizeof(COL_PACK_INFO), 
        MYF(MY_WME | MY_ZEROFILL)
        );
1313
    if (kc_info->cp_info[keynr] == NULL) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1314 1315 1316 1317 1318 1319 1320
        error = ENOMEM;
        goto exit;
    }
    {
    u_int32_t curr_fixed_offset = 0;
    u_int32_t curr_var_index = 0;
    for (uint j = 0; j < table_share->fields; j++) {
1321
        COL_PACK_INFO* curr = &kc_info->cp_info[keynr][j];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1322 1323 1324 1325
        //
        // need to set the offsets / indexes
        // offsets are calculated AFTER the NULL bytes
        //
1326 1327
        if (!bitmap_is_set(&kc_info->key_filters[keynr],j)) {
            if (kc_info->field_lengths[j]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1328
                curr->col_pack_val = curr_fixed_offset;
1329
                curr_fixed_offset += kc_info->field_lengths[j];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1330
            }
1331
            else if (kc_info->length_bytes[j]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1332 1333 1334 1335 1336 1337 1338 1339 1340
                curr->col_pack_val = curr_var_index;
                curr_var_index++;
            }
        }
    }
    
    //
    // set up the mcp_info
    //
1341 1342
    kc_info->mcp_info[keynr].var_len_offset = get_var_len_offset(
        kc_info,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1343 1344 1345
        table_share,
        keynr
        );
1346 1347
    kc_info->mcp_info[keynr].len_of_offsets = get_len_of_offsets(
        kc_info,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1348 1349 1350 1351 1352 1353 1354 1355 1356 1357
        table_share,
        keynr
        );

    error = 0;
    }
exit:
    return error;
}

1358
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) {
1359
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1360
    u_int32_t curr_blob_field_index = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1361
    u_int32_t max_var_bytes = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376
    //
    // 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();
1377 1378
            assert(pack_length < 1<<16);
            kc_info->field_lengths[i] = (u_int16_t)pack_length;
1379
            kc_info->length_bytes[i] = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1380 1381
            break;
        case toku_type_blob:
1382 1383 1384
            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
1385 1386 1387 1388 1389 1390 1391
            curr_blob_field_index++;
            break;
        case toku_type_varstring:
        case toku_type_varbinary:
            //
            // meaning it is variable sized
            //
1392 1393
            kc_info->field_lengths[i] = 0;
            kc_info->length_bytes[i] = (uchar)((Field_varstring *)field)->length_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1394
            max_var_bytes += field->field_length;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1395 1396 1397 1398 1399
            break;
        default:
            assert(false);
        }
    }
1400
    kc_info->num_blobs = curr_blob_field_index;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1401

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1402 1403 1404 1405 1406 1407
    //
    // 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) {
1408
        kc_info->num_offset_bytes = 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1409 1410
    }
    else {
1411
        kc_info->num_offset_bytes = 2;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1412 1413 1414
    }


Zardosht Kasheff's avatar
Zardosht Kasheff committed
1415 1416 1417 1418 1419 1420 1421
    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(
1422
                    &kc_info->key_filters[primary_key],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1423 1424 1425 1426 1427
                    &table_share->key_info[primary_key],
                    table,
                    true
                    );
            }
1428
            else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1429
                set_key_filter(
1430
                    &kc_info->key_filters[i],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1431 1432 1433 1434 1435 1436
                    &table_share->key_info[i],
                    table,
                    true
                    );
                if (!hidden_primary_key) {
                    set_key_filter(
1437
                        &kc_info->key_filters[i],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1438 1439 1440 1441 1442 1443 1444 1445
                        &table_share->key_info[primary_key],
                        table,
                        true
                        );
                }
            }
        }
        if (i == primary_key || table_share->key_info[i].flags & HA_CLUSTERING) {
1446
            error = initialize_col_pack_info(kc_info,table_share,i);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1447 1448 1449 1450 1451 1452
            if (error) {
                goto exit;
            }
        }

    }
1453 1454 1455
exit:
    return error;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1456

1457 1458
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
1459 1460 1461 1462 1463 1464
    bool ret_val;
    if (curr_num_DBs == 1) {
        ret_val = true;
        goto exit;
    }
    ret_val = true;
1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480
    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;
            }
            
        }
1481
    }
1482 1483 1484
exit:
    return ret_val;
}
1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514

int ha_tokudb::initialize_share(
    const char* name,
    int mode
    )
{
    int error = 0;
    u_int64_t num_rows = 0;
    bool table_exists;
    DBUG_PRINT("info", ("share->use_count %u", share->use_count));

    table_exists = true;
    error = check_table_in_metadata(name, &table_exists);

    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;
    }
    
    error = initialize_key_and_col_info(
        table_share,
        table, 
        &share->kc_info,
        hidden_primary_key,
        primary_key
        );
1515
    if (error) { goto exit; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1516
    
1517 1518 1519 1520
    error = open_main_dictionary(name, mode == O_RDONLY, NULL);
    if (error) { goto exit; }

    share->has_unique_keys = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1521
    /* Open other keys;  These are part of the share structure */
1522 1523 1524 1525
    for (uint i = 0; i < table_share->keys + test(hidden_primary_key); i++) {
        if (table_share->key_info[i].flags & HA_NOSAME) {
            share->has_unique_keys = true;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1526
        if (i != primary_key) {
1527
            error = open_secondary_dictionary(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1528 1529 1530
                &share->key_file[i],
                &table_share->key_info[i],
                name,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1531
                mode == O_RDONLY,
1532
                NULL
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1533 1534 1535 1536 1537 1538
                );
            if (error) {
                goto exit;
            }
        }
    }
1539 1540 1541 1542 1543 1544
    share->replace_into_fast = can_replace_into_be_fast(
        table_share, 
        &share->kc_info, 
        primary_key
        );
        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568
    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
        //
        ref_length = sizeof(u_int32_t) + sizeof(uchar);
        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;

    error = get_status();
    if (error) {
        goto exit;
    }
    if (share->version < HA_TOKU_VERSION) {
        error = ENOSYS;
        goto exit;
    }

1569
    error = estimate_num_rows(share->file,&num_rows, NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586
    //
    // 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();
    }

1587 1588 1589 1590 1591 1592 1593
    if (may_table_be_empty()) {
        share->try_table_lock = true;
    }
    else {
        share->try_table_lock = false;
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1594 1595 1596 1597 1598
    error = 0;
exit:
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1599 1600


Zardosht Kasheff's avatar
Zardosht Kasheff committed
1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611
//
// 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
//
1612
int ha_tokudb::open(const char *name, int mode, uint test_if_locked) {
1613
    TOKUDB_DBUG_ENTER("ha_tokudb::open %p %s", this, name);
1614

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1615 1616
    int error = 0;
    int ret_val = 0;
1617

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1618 1619 1620
    transaction = NULL;
    cursor = NULL;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1621

1622 1623 1624 1625 1626 1627
    /* 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
1628 1629
        hidden_primary_key = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
        ref_length = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH + sizeof(u_int32_t);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1630
    } 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1631
    else {
1632
        key_used_on_scan = primary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1633
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1634

1635
    /* Need some extra memory in case of packed keys */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1636
    // the "+ 1" is for the first byte that states +/- infinity
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1637 1638
    // 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
1639 1640 1641
    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
1642
        &key_buff3, max_key_length, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1643 1644 1645 1646 1647 1648 1649 1650
        &primary_key_buff, (hidden_primary_key ? 0 : max_key_length),
        &fixed_cols_for_query, table_share->fields*sizeof(u_int32_t),
        &var_cols_for_query, table_share->fields*sizeof(u_int32_t),
        NullS
        );
    if (alloc_ptr == NULL) {
        ret_val = 1;
        goto exit;
1651 1652
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1653
    alloced_rec_buff_length = table_share->rec_buff_length + table_share->fields;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1654
    rec_buff = (uchar *) my_malloc(alloced_rec_buff_length, MYF(MY_WME));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1655 1656 1657 1658

    if (rec_buff == NULL) {
        ret_val = 1;
        goto exit;
1659
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1660

1661 1662 1663 1664 1665 1666
    for (u_int32_t i = 0; i < (table_share->keys); i++) {
        if (i == primary_key) {
            continue;
        }
        mult_key_buff[i] = (uchar *)my_malloc(max_key_length, MYF(MY_WME));
        assert(mult_key_buff[i] != NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1667 1668 1669
        mult_key_dbt[i].ulen = max_key_length;
        mult_key_dbt[i].flags = DB_DBT_USERMEM;
        mult_key_dbt[i].data = mult_key_buff[i];
1670 1671 1672
        if (table_share->key_info[i].flags & HA_CLUSTERING) {
            mult_rec_buff[i] = (uchar *) my_malloc(alloced_rec_buff_length, MYF(MY_WME));
            assert(mult_rec_buff[i]);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1673 1674 1675
            mult_rec_dbt[i].ulen = alloced_rec_buff_length;
            mult_rec_dbt[i].flags = DB_DBT_USERMEM;
            mult_rec_dbt[i].data = mult_rec_buff[i];
1676 1677 1678 1679
        }
    }
    alloced_mult_rec_buff_length = alloced_rec_buff_length;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1680 1681 1682 1683 1684
    /* Init shared structure */
    share = get_share(name, table_share);
    if (share == NULL) {
        ret_val = 1;
        goto exit;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1685 1686
    }

1687 1688 1689 1690
    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
1691
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
1692 1693
        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
1694
    }
1695
    if (!share->use_count++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1696 1697 1698 1699 1700
        ret_val = initialize_share(
            name,
            mode
            );
        if (ret_val) {
1701
            free_share(share, 1);
1702
            goto exit;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1703
        }
1704 1705 1706 1707
    }
    ref_length = share->ref_length;     // If second open
    pthread_mutex_unlock(&share->mutex);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1708
    key_read = false;
1709
    stats.block_size = 1<<20;    // QQQ Tokudb DB block size
1710

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1711 1712
    init_hidden_prim_key_info();

1713 1714
    info(HA_STATUS_NO_LOCK | HA_STATUS_VARIABLE | HA_STATUS_CONST);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1715 1716 1717 1718 1719 1720
exit:
    if (ret_val) {
        my_free(alloc_ptr, MYF(MY_ALLOW_ZERO_PTR));
        alloc_ptr = NULL;
        my_free(rec_buff, MYF(MY_ALLOW_ZERO_PTR));
        rec_buff = NULL;
1721 1722 1723 1724 1725
        for (u_int32_t i = 0; i < (table_share->keys); i++) {
            my_free(mult_key_buff[i], MYF(MY_ALLOW_ZERO_PTR));
            my_free(mult_rec_buff[i], MYF(MY_ALLOW_ZERO_PTR));
        }
        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1726 1727 1728 1729 1730
        if (error) {
            my_errno = error;
        }
    }
    TOKUDB_DBUG_RETURN(ret_val);
1731 1732
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1733 1734 1735 1736 1737 1738 1739 1740 1741
//
// 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
//
1742
int ha_tokudb::estimate_num_rows(DB* db, u_int64_t* num_rows, DB_TXN* txn) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1743 1744 1745 1746 1747 1748 1749
    DBT key;
    DBT data;
    int error = ENOSYS;
    DBC* crsr = NULL;
    u_int64_t less, equal, greater;
    int is_exact;
    bool do_commit = false;
1750
    DB_TXN* txn_to_use = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1751 1752 1753 1754

    bzero((void *)&key, sizeof(key));
    bzero((void *)&data, sizeof(data));

1755 1756
    if (txn == NULL) {
        error = db_env->txn_begin(db_env, 0, &txn_to_use, DB_READ_UNCOMMITTED);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1757 1758 1759
        if (error) goto cleanup;
        do_commit = true;
    }
1760 1761 1762
    else {
        txn_to_use = txn;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1763
    
1764
    error = db->cursor(db, txn_to_use, &crsr, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780
    if (error) { goto cleanup; }

    //
    // get the first element, then estimate number of records
    // by calling key_range64 on the first element
    //
    error = crsr->c_get(crsr, &key, &data, DB_FIRST);
    if (error == DB_NOTFOUND) {
        *num_rows = 0;
        error = 0;
        goto cleanup;
    }
    else if (error) { goto cleanup; }

    error = db->key_range64(
        db, 
1781
        txn_to_use, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795
        &key, 
        &less,
        &equal,
        &greater,
        &is_exact
        );
    if (error) {
        goto cleanup;
    }


    *num_rows = equal + greater;
    error = 0;
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1796
    if (crsr != NULL) {
1797 1798
        int r = crsr->c_close(crsr);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1799 1800
        crsr = NULL;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1801
    if (do_commit) {
1802 1803
        commit_txn(txn_to_use, 0);
        txn_to_use = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1804
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1805 1806 1807
    return error;
}

1808

1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851
int ha_tokudb::write_to_status(DB* db, HA_METADATA_KEY curr_key_data, void* data, uint size, DB_TXN* txn ){
    return write_metadata(db, &curr_key_data, sizeof(curr_key_data), data, size, txn);
}


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;
    }

    bzero(&key, sizeof(key));
    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;
1852 1853
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1854 1855 1856
//
// helper function to write a piece of metadata in to status.tokudb
//
1857
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
1858 1859 1860 1861
    int error;
    DBT key;
    DBT value;
    DB_TXN* txn = NULL;
1862
    bool do_commit = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1863 1864 1865
    //
    // transaction to be used for putting metadata into status.tokudb
    //
1866 1867 1868 1869 1870 1871 1872 1873 1874
    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
1875 1876 1877 1878
    }

    bzero(&key, sizeof(key));
    bzero(&value, sizeof(value));
1879 1880 1881 1882
    key.data = key_data;
    key.size = key_size;
    value.data = val_data;
    value.size = val_size;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1883 1884 1885 1886 1887 1888 1889
    error = db->put(db, txn, &key, &value, 0);
    if (error) { 
        goto cleanup; 
    }
    
    error = 0;
cleanup:
1890
    if (do_commit && txn) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1891
        if (!error) {
1892
            commit_txn(txn, DB_TXN_NOSYNC);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1893 1894
        }
        else {
1895
            abort_txn(txn);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912
        }
    }
    return error;
}



//
// 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){
1913
    return write_to_status(db,hatoku_max_ai,&val,sizeof(val), NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926
}

//
// 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
//
//
1927 1928
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
1929
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1930 1931


Zardosht Kasheff's avatar
Zardosht Kasheff committed
1932 1933 1934
//
// Closes a handle to a table. 
//
1935
int ha_tokudb::close(void) {
1936
    TOKUDB_DBUG_ENTER("ha_tokudb::close %p", this);
1937
    TOKUDB_DBUG_RETURN(__close(0));
1938 1939 1940
}

int ha_tokudb::__close(int mutex_is_locked) {
1941
    TOKUDB_DBUG_ENTER("ha_tokudb::__close %p", this);
1942
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) 
1943
        TOKUDB_TRACE("close:%p\n", this);
1944
    my_free(rec_buff, MYF(MY_ALLOW_ZERO_PTR));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1945
    my_free(blob_buff, MYF(MY_ALLOW_ZERO_PTR));
1946
    my_free(alloc_ptr, MYF(MY_ALLOW_ZERO_PTR));
1947 1948 1949 1950
    for (u_int32_t i = 0; i < (table_share->keys); i++) {
        my_free(mult_key_buff[i], MYF(MY_ALLOW_ZERO_PTR));
        my_free(mult_rec_buff[i], MYF(MY_ALLOW_ZERO_PTR));
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1951 1952
    rec_buff = NULL;
    alloc_ptr = NULL;
1953
    ha_tokudb::reset();
1954
    TOKUDB_DBUG_RETURN(free_share(share, mutex_is_locked));
1955 1956
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1957 1958 1959 1960 1961 1962
//
// Reallocate record buffer (rec_buff) if needed
// If not needed, does nothing
// Parameters:
//          length - size of buffer required for rec_buff
//
1963
bool ha_tokudb::fix_rec_buff_for_blob(ulong length) {
1964
    if (!rec_buff || (length > alloced_rec_buff_length)) {
1965
        uchar *newptr;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1966
        if (!(newptr = (uchar *) my_realloc((void *) rec_buff, length, MYF(MY_ALLOW_ZERO_PTR))))
1967 1968 1969 1970 1971
            return 1;
        rec_buff = newptr;
        alloced_rec_buff_length = length;
    }
    return 0;
1972 1973
}

1974 1975 1976 1977 1978 1979 1980 1981 1982
void ha_tokudb::fix_mult_rec_buff() {
    if (alloced_rec_buff_length > alloced_mult_rec_buff_length) {
        for (uint i = 0; i < table_share->keys; i++) {
            if (table_share->key_info[i].flags & HA_CLUSTERING) {
                uchar *newptr;
                if (!(newptr = (uchar *) my_realloc((void *) mult_rec_buff[i], alloced_rec_buff_length, MYF(MY_ALLOW_ZERO_PTR)))) {
                    assert(false);
                }
                mult_rec_buff[i] = newptr;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1983 1984 1985
                mult_rec_dbt[i].ulen = alloced_rec_buff_length;
                mult_rec_dbt[i].flags = DB_DBT_USERMEM;
                mult_rec_dbt[i].data = mult_rec_buff[i];
1986 1987 1988 1989 1990 1991 1992
            }
        }
        alloced_mult_rec_buff_length = alloced_rec_buff_length;
    }
}


1993
/* Calculate max length needed for row */
1994 1995 1996 1997 1998
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
1999
        length += blob->get_length((uchar *) (buf + field_offset(blob, table))) + 2;
2000 2001
    }
    return length;
2002 2003 2004 2005
}

/*
*/
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
//
// 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
2016
//      [out]   buf - buffer where row is packed
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2017 2018
//      [in]    record - row in MySQL format
//
2019

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2020 2021
int ha_tokudb::pack_row(
    DBT * row, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2022 2023
    const uchar* record,
    uint index
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2024 2025
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2026 2027 2028 2029
    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
2030
    int r = ENOSYS;
2031
    bzero((void *) row, sizeof(*row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2032 2033 2034

    my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2035
    if (table_share->blob_fields) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2036 2037 2038 2039
        if (fix_rec_buff_for_blob(max_row_length(record))) {
            r = HA_ERR_OUT_OF_MEM;
            goto cleanup;
        }
2040 2041
    }

2042

2043
    /* Copy null bits */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2044 2045
    memcpy(rec_buff, record, table_share->null_bytes);
    fixed_field_ptr = rec_buff + table_share->null_bytes;
2046 2047 2048
    var_field_offset_ptr = fixed_field_ptr + share->kc_info.mcp_info[index].var_len_offset;
    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;
2049

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2050 2051 2052
    //
    // assert that when the hidden primary key exists, primary_key_offsets is NULL
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2053 2054 2055
    for (uint i = 0; i < table_share->fields; i++) {
        Field* field = table->field[i];
        uint curr_field_offset = field_offset(field, table);
2056
        if (bitmap_is_set(&share->kc_info.key_filters[index],i)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2057 2058
            continue;
        }
2059
        if (share->kc_info.field_lengths[i]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2060 2061 2062
            fixed_field_ptr = pack_fixed_field(
                fixed_field_ptr,
                record + curr_field_offset, 
2063
                share->kc_info.field_lengths[i]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2064 2065
                );
        }
2066
        else if (share->kc_info.length_bytes[i]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2067 2068 2069 2070 2071
            var_field_data_ptr = pack_var_field(
                var_field_offset_ptr,
                var_field_data_ptr,
                start_field_data_ptr,
                record + curr_field_offset,
2072 2073
                share->kc_info.length_bytes[i],
                share->kc_info.num_offset_bytes
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2074
                );
2075
            var_field_offset_ptr += share->kc_info.num_offset_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2076 2077 2078
        }
    }

2079 2080
    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
2081
        var_field_data_ptr = pack_toku_field_blob(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2082
            var_field_data_ptr,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2083 2084
            record + field_offset(field, table),
            field
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2085
            );
2086
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2087

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2088 2089
    row->data = rec_buff;
    row->size = (size_t) (var_field_data_ptr - rec_buff);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2090 2091 2092 2093 2094
    r = 0;

cleanup:
    dbug_tmp_restore_column_map(table->write_set, old_map);
    return r;
2095 2096
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2097

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2098
int ha_tokudb::unpack_blobs(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2099 2100
    uchar* record,
    const uchar* from_tokudb_blob,
2101 2102
    u_int32_t num_bytes,
    bool check_bitmap
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2103 2104
    )
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2105 2106 2107
    uint error = 0;
    uchar* ptr = NULL;
    const uchar* buff = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2108
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2109
    // assert that num_bytes > 0 iff share->num_blobs > 0
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2110
    //
2111
    assert( !((share->kc_info.num_blobs == 0) && (num_bytes > 0)) );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123
    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;
2124
    for (uint i = 0; i < share->kc_info.num_blobs; i++) {
2125 2126 2127 2128 2129 2130
        u_int32_t curr_field_index = share->kc_info.blob_fields[i]; 
        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];
2131
        u_int32_t len_bytes = field->row_pack_length();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2132
        buff = unpack_toku_field_blob(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2133
            record + field_offset(field, table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2134
            buff,
2135
            len_bytes,
2136
            skip
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2137 2138 2139
            );
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2140 2141 2142
    error = 0;
exit:
    return error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2143 2144
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2145 2146 2147 2148 2149 2150
//
// 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
2151
int ha_tokudb::unpack_row(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2152 2153
    uchar* record, 
    DBT const *row, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2154 2155
    DBT const *key,
    uint index
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2156 2157
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2158 2159 2160 2161
    //
    // two cases, fixed length row, and variable length row
    // fixed length row is first below
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2162
    /* Copy null bits */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2163
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2164 2165 2166 2167 2168 2169
    const uchar* fixed_field_ptr = (const uchar *) row->data;
    const uchar* var_field_offset_ptr = NULL;
    const uchar* var_field_data_ptr = NULL;
    u_int32_t data_end_offset = 0;
    memcpy(record, fixed_field_ptr, table_share->null_bytes);
    fixed_field_ptr += table_share->null_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2170

2171 2172
    var_field_offset_ptr = fixed_field_ptr + share->kc_info.mcp_info[index].var_len_offset;
    var_field_data_ptr = var_field_offset_ptr + share->kc_info.mcp_info[index].len_of_offsets;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2173

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192
    //
    // unpack the key, if necessary
    //
    if (!(hidden_primary_key && index == primary_key)) {
        unpack_key(record,key,index);
    }

    u_int32_t last_offset = 0;
    //
    // 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];
2193
            if (bitmap_is_set(&share->kc_info.key_filters[index],i)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2194 2195
                continue;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2196

2197
            if (share->kc_info.field_lengths[i]) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2198 2199 2200
                fixed_field_ptr = unpack_fixed_field(
                    record + field_offset(field, table),
                    fixed_field_ptr,
2201
                    share->kc_info.field_lengths[i]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2202 2203
                    );
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2204
            //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2205 2206
            // 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
2207
            //
2208 2209
            else if (share->kc_info.length_bytes[i]) {
                switch (share->kc_info.num_offset_bytes) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223
                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,
2224
                    share->kc_info.length_bytes[i]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2225
                    );
2226
                var_field_offset_ptr += share->kc_info.num_offset_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2227 2228
                var_field_data_ptr += data_end_offset - last_offset;
                last_offset = data_end_offset;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2229
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2230
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2231
        error = unpack_blobs(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2232 2233
            record,
            var_field_data_ptr,
2234 2235
            row->size - (u_int32_t)(var_field_data_ptr - (const uchar *)row->data),
            false
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2236
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2237 2238 2239
        if (error) {
            goto exit;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2240
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2241 2242 2243 2244
    //
    // in this case, we unpack only what is specified 
    // in fixed_cols_for_query and var_cols_for_query
    //
2245
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2246 2247 2248 2249 2250 2251 2252 2253
        //
        // first the fixed fields
        //
        for (u_int32_t i = 0; i < num_fixed_cols_for_query; i++) {
            uint field_index = fixed_cols_for_query[i];
            Field* field = table->field[field_index];
            unpack_fixed_field(
                record + field_offset(field, table),
2254 2255
                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
2256
                );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2257 2258 2259
        }

        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2260 2261
        // 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
2262
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2263 2264 2265
        for (u_int32_t i = 0; i < num_var_cols_for_query; i++) {
            uint field_index = var_cols_for_query[i];
            Field* field = table->field[field_index];
2266
            u_int32_t var_field_index = share->kc_info.cp_info[index][field_index].col_pack_val;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2267
            u_int32_t data_start_offset;
2268
            u_int32_t field_len;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2269
            
2270 2271 2272 2273 2274 2275 2276
            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
2277 2278 2279 2280

            unpack_var_field(
                record + field_offset(field, table),
                var_field_data_ptr + data_start_offset,
2281 2282
                field_len,
                share->kc_info.length_bytes[field_index]
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2283 2284 2285
                );
        }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2286 2287 2288 2289
        if (read_blobs) {
            //
            // now the blobs
            //
2290 2291 2292 2293 2294 2295
            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
2296

2297
            var_field_data_ptr += data_end_offset;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2298
            error = unpack_blobs(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2299 2300
                record,
                var_field_data_ptr,
2301 2302
                row->size - (u_int32_t)(var_field_data_ptr - (const uchar *)row->data),
                true
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2303
                );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2304 2305 2306
            if (error) {
                goto exit;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2307
        }
2308
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2309 2310 2311
    error = 0;
exit:
    return error;
2312 2313
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2314 2315 2316 2317 2318 2319
u_int32_t ha_tokudb::place_key_into_mysql_buff(
    KEY* key_info, 
    uchar * record, 
    uchar* data
    ) 
{
2320
    KEY_PART_INFO *key_part = key_info->key_part, *end = key_part + key_info->key_parts;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2321
    uchar *pos = data;
2322 2323

    for (; key_part != end; key_part++) {
2324 2325
        if (key_part->field->null_bit) {
            uint null_offset = get_null_offset(table, key_part->field);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2326
            if (*pos++ == NULL_COL_VAL) { // Null value
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2327 2328 2329 2330
                //
                // We don't need to reset the record data as we will not access it
                // if the null data is set
                //            
2331
                record[null_offset] |= key_part->field->null_bit;
2332 2333
                continue;
            }
2334
            record[null_offset] &= ~key_part->field->null_bit;
2335
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2336 2337 2338 2339
        //
        // HOPEFULLY TEMPORARY
        //
        assert(table->s->db_low_byte_first);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2340
        pos = unpack_toku_key_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2341
            record + field_offset(key_part->field, table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2342
            pos,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2343 2344
            key_part->field,
            key_part->length
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2345
            );
2346
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2347
    return pos-data;
2348 2349
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2350
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2351
// Store the key and the primary key into the row
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2352
// Parameters:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2353 2354 2355 2356
//      [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
2357
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2358 2359 2360
void ha_tokudb::unpack_key(uchar * record, DBT const *key, uint index) {
    u_int32_t bytes_read;
    uchar *pos = (uchar *) key->data + 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2361 2362 2363 2364 2365
    bytes_read = place_key_into_mysql_buff(
        &table->key_info[index], 
        record, 
        pos
        );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2366
    if( (index != primary_key) && !hidden_primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2367 2368 2369
        //
        // also unpack primary key
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2370 2371 2372 2373 2374
        place_key_into_mysql_buff(
            &table->key_info[primary_key], 
            record, 
            pos+bytes_read
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2375 2376
    }
}
2377

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2378 2379 2380 2381 2382 2383 2384 2385
u_int32_t ha_tokudb::place_key_into_dbt_buff(
    KEY* key_info, 
    uchar * buff, 
    const uchar * record, 
    bool* has_null, 
    int key_length
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2386 2387 2388
    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
2389
    *has_null = false;
2390
    for (; key_part != end && key_length > 0; key_part++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2391 2392 2393 2394 2395 2396
        //
        // 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) {
2397
            /* Store 0 if the key part is a NULL part */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2398
            uint null_offset = get_null_offset(table, key_part->field);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2399
            if (record[null_offset] & key_part->field->null_bit) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2400
                *curr_buff++ = NULL_COL_VAL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2401
                *has_null = true;
2402 2403
                continue;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2404
            *curr_buff++ = NONNULL_COL_VAL;        // Store NOT NULL marker
2405
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2406
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2407 2408 2409 2410
        // HOPEFULLY TEMPORARY
        //
        assert(table->s->db_low_byte_first);
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2411 2412 2413 2414
        // 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
2415
        curr_buff = pack_toku_key_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2416
            curr_buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2417
            (uchar *) (record + field_offset(key_part->field, table)),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2418 2419
            key_part->field,
            key_part->length
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2420
            );
2421 2422
        key_length -= key_part->length;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441
    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
2442
DBT* ha_tokudb::create_dbt_key_from_key(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2443
    DBT * key,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2444
    KEY* key_info, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2445
    uchar * buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2446
    const uchar * record, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2447 2448
    bool* has_null,
    bool dont_pack_pk,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2449 2450 2451
    int key_length
    ) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2452 2453 2454 2455 2456 2457 2458 2459
    u_int32_t size = 0;
    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
2460
    // positive infinity or negative infinity or zero. For this, because we are creating key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2461 2462 2463
    // 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
    //
2464
    *tmp_buff++ = COL_ZERO;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2465
    size++;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2466 2467 2468 2469 2470 2471 2472
    size += place_key_into_dbt_buff(
        key_info, 
        tmp_buff, 
        record, 
        has_null, 
        key_length
        );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2473
    if (!dont_pack_pk) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491
        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;
2492 2493
    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
2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506
    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
2507
//      [out]   has_null - says if the key has a NULL value for one of its columns
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2508 2509 2510 2511
//              key_length - currently set to MAX_KEY_LENGTH, is it size of buff?
// Returns:
//      the parameter key
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2512 2513 2514 2515 2516 2517 2518 2519 2520
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
2521 2522 2523
    TOKUDB_DBUG_ENTER("ha_tokudb::create_dbt_key_from_table");
    bzero((void *) key, sizeof(*key));
    if (hidden_primary_key && keynr == primary_key) {
2524 2525
        key->data = buff;
        memcpy(buff, &current_ident, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2526
        key->size = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2527
        *has_null = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2528 2529
        DBUG_RETURN(key);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2530
    DBUG_RETURN(create_dbt_key_from_key(key, &table->key_info[keynr],buff,record, has_null, (keynr == primary_key), key_length));
2531 2532
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544
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));    
}
2545

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557
//
// 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
2558 2559 2560 2561 2562 2563
DBT *ha_tokudb::pack_key(
    DBT * key, 
    uint keynr, 
    uchar * buff, 
    const uchar * key_ptr, 
    uint key_length, 
2564
    int8_t inf_byte
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2565 2566
    ) 
{
2567
    TOKUDB_DBUG_ENTER("ha_tokudb::pack_key");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2568
    KEY *key_info = &table->key_info[keynr];
2569 2570 2571 2572 2573 2574 2575
    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);

    bzero((void *) key, sizeof(*key));
    key->data = buff;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2576 2577 2578 2579
    //
    // first put the "infinity" byte at beginning. States if missing columns are implicitly
    // positive infinity or negative infinity
    //
2580
    *buff++ = (uchar)inf_byte;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2581

2582 2583 2584
    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
2585 2586
            if (!(*key_ptr == 0)) {
                *buff++ = NULL_COL_VAL;
2587 2588 2589 2590
                key_length -= key_part->store_length;
                key_ptr += key_part->store_length;
                continue;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2591
            *buff++ = NONNULL_COL_VAL;
2592 2593
            offset = 1;         // Data is at key_ptr+1
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2594 2595
        assert(table->s->db_low_byte_first);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2596
        buff = pack_key_toku_key_field(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2597
            buff,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2598
            (uchar *) key_ptr + offset,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2599 2600
            key_part->field,
            key_part->length
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2601
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2602
        
2603 2604 2605 2606 2607 2608 2609
        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);
2610 2611
}

2612 2613 2614 2615
//
// Reads the last element of dictionary of index keynr, and places
// the data into table->record[1].
//
2616
int ha_tokudb::read_last(uint keynr) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2617
    TOKUDB_DBUG_ENTER("ha_tokudb::read_last");
2618
    int do_commit = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2619
    if (transaction == NULL) {
2620 2621 2622 2623
        int r = db_env->txn_begin(db_env, 0, &transaction, 0);
        assert(r == 0);
        do_commit = 1;
    }
2624
    int error = index_init(keynr, 0);
2625 2626 2627 2628
    if (error == 0)
        error = index_last(table->record[1]);
    index_end();
    if (do_commit) {
2629
        commit_txn(transaction, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2630
        transaction = NULL;
2631
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2632
    TOKUDB_DBUG_RETURN(error);
2633 2634
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2635 2636 2637 2638 2639
//
// 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
2640
    pthread_mutex_lock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2641 2642
    if (!(share->status & STATUS_PRIMARY_KEY_INIT)) {
        (void) extra(HA_EXTRA_KEYREAD);
2643
        int error = read_last(primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2644 2645
        (void) extra(HA_EXTRA_NO_KEYREAD);
        if (error == 0) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2646
            share->auto_ident = hpk_char_to_num(current_ident);
2647
        }
2648

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2649
        share->status |= STATUS_PRIMARY_KEY_INIT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2650 2651
    }
    pthread_mutex_unlock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2652 2653
    DBUG_VOID_RETURN;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2654

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2655 2656 2657 2658 2659


/** @brief
    Get metadata info stored in status.tokudb
    */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2660
int ha_tokudb::get_status() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2661
    TOKUDB_DBUG_ENTER("ha_tokudb::get_status");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2662 2663 2664 2665
    DB_TXN* txn = NULL;
    DBT key, value;
    HA_METADATA_KEY curr_key;
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2666 2667 2668 2669
    //
    // open status.tokudb
    //
    if (!share->status_block) {
2670 2671 2672 2673
        error = open_status_dictionary(
            &share->status_block, 
            share->table_name, 
            NULL
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2674
            );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2675 2676 2677
        if (error) { 
            goto cleanup; 
        }
2678
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2679
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2680 2681 2682 2683 2684 2685 2686
    //
    // transaction to be used for putting metadata into status.tokudb
    //
    bzero(&key, sizeof(key));
    bzero(&value, sizeof(value));
    key.data = &curr_key;
    key.size = sizeof(curr_key);
2687
    value.flags = DB_DBT_USERMEM;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2688 2689
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2690

2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710
    assert(share->status_block);
    //
    // get version
    //
    value.ulen = sizeof(share->version);
    value.data = &share->version;
    curr_key = hatoku_version;
    error = share->status_block->get(
        share->status_block, 
        txn, 
        &key, 
        &value, 
        0
        );
    if (error == DB_NOTFOUND) {
        share->version = 0;
    }
    else if (error || value.size != sizeof(share->version)) {
        if (error == 0) {
            error = HA_ERR_INTERNAL_ERROR;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2711
        }
2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732
        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
2733
        }
2734
        goto cleanup;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2735
    }
2736
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2737 2738 2739
    error = 0;
cleanup:
    if (txn) {
2740
        commit_txn(txn,0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2741 2742
    }
    TOKUDB_DBUG_RETURN(error);
2743 2744
}

2745 2746 2747
/** @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
2748
    This is used in filesort.cc. 
2749
*/
2750
ha_rows ha_tokudb::estimate_rows_upper_bound() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2751
    TOKUDB_DBUG_ENTER("ha_tokudb::estimate_rows_upper_bound");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2752
    DBUG_RETURN(share->rows + HA_TOKUDB_EXTRA_ROWS);
2753
}
2754

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2755 2756 2757 2758
//
// Function that compares two primary keys that were saved as part of rnd_pos
// and ::position
//
2759
int ha_tokudb::cmp_ref(const uchar * ref1, const uchar * ref2) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2760 2761
    int ret_val = 0;
    ret_val = tokudb_compare_two_keys(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2762 2763 2764
        ref1 + sizeof(u_int32_t),
        *(u_int32_t *)ref1,
        ref2 + sizeof(u_int32_t),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2765
        *(u_int32_t *)ref2,
2766 2767
        (uchar *)share->file->descriptor->dbt.data + 4,
        *(u_int32_t *)share->file->descriptor->dbt.data - 4,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2768
        false
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2769 2770
        );
    return ret_val;
2771 2772
}

2773
bool ha_tokudb::check_if_incompatible_data(HA_CREATE_INFO * info, uint table_changes) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785
    //
    // 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;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2786
    if (table_changes != IS_EQUAL_YES)
2787 2788 2789
        return COMPATIBLE_DATA_NO;
    return COMPATIBLE_DATA_YES;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2790

2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802
//
// 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
//
2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832
//
// 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.
//
bool ha_tokudb::may_table_be_empty() {
    int error;
    bool ret_val = false;
    DBC* tmp_cursor = NULL;
    DB_TXN* txn = NULL;

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

    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) {
2833 2834
        int r = tmp_cursor->c_close(tmp_cursor);
        assert(r==0);
2835 2836 2837
        tmp_cursor = NULL;
    }
    if (txn) {
2838
        commit_txn(txn, 0);
2839 2840 2841 2842 2843
        txn = NULL;
    }
    return ret_val;
}

2844
void ha_tokudb::start_bulk_insert(ha_rows rows) {
2845 2846
    TOKUDB_DBUG_ENTER("ha_tokudb::start_bulk_insert");
    THD* thd = ha_thd();
2847
    delay_updating_ai_metadata = true;
2848
    ai_metadata_update_required = false;
2849
    abort_loader = false;
2850
    if (share->try_table_lock) {
2851
        if (get_prelock_empty(thd) && may_table_be_empty()) {
2852
            if (using_ignore || get_load_save_space(thd)) {
2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873
                acquire_table_lock(transaction, lock_write);
            }
            else {
                u_int32_t mult_put_flags[MAX_KEY + 1] = {DB_YESOVERWRITE};
                u_int32_t mult_dbt_flags[MAX_KEY + 1] = {DB_DBT_REALLOC};
                uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
                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;
                }
                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,
                    0
                    );
2874 2875 2876 2877
                if (error) { 
                    assert(loader == NULL);
                    goto exit_try_table_lock;
                }
2878 2879 2880 2881 2882 2883 2884 2885 2886 2887

                lc.thd = thd;
                lc.ha = this;
                
                error = loader->set_poll_function(loader, poll_fun, &lc);
                assert(!error);

                error = loader->set_error_callback(loader, loader_dup_fun, &lc);
                assert(!error);
            }
2888
        }
2889
    exit_try_table_lock:
2890 2891 2892 2893
        pthread_mutex_lock(&share->mutex);
        share->try_table_lock = false;
        pthread_mutex_unlock(&share->mutex);
    }
2894
    DBUG_VOID_RETURN;
2895 2896 2897 2898 2899 2900 2901 2902
}

//
// 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.
//
int ha_tokudb::end_bulk_insert() {
2903
    TOKUDB_DBUG_ENTER("ha_tokudb::end_bulk_insert");
2904
    int error = 0;
2905
    THD* thd = ha_thd();
2906
    bool using_loader = (loader != NULL);
2907 2908 2909 2910
    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);
2911
        if (error) { goto cleanup; }
2912 2913 2914
    }
    delay_updating_ai_metadata = false;
    ai_metadata_update_required = false;
2915 2916
    loader_error = 0;
    if (loader) {
2917
        if (!abort_loader && !thd->killed) {
2918 2919
            error = loader->close(loader);
            loader = NULL;
2920 2921 2922 2923 2924 2925
            if (error) { 
                if (thd->killed) {
                    my_error(ER_QUERY_INTERRUPTED, MYF(0));
                }
                goto cleanup; 
            }
2926 2927 2928 2929

            for (uint i = 0; i < table_share->keys; i++) {
                if (table_share->key_info[i].flags & HA_NOSAME) {
                    bool is_unique;
2930 2931 2932
                    if (i == primary_key) {
                        continue;
                    }
2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948
                    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 {
2949 2950
            error = sprintf(write_status_msg, "aborting bulk load"); 
            thd_proc_info(thd, write_status_msg);
2951 2952
            loader->abort(loader);
            loader = NULL;
2953
            share->try_table_lock = true;
2954 2955 2956 2957 2958
        }
    }

cleanup:
    if (loader) {
2959 2960
        error = sprintf(write_status_msg, "aborting bulk load"); 
        thd_proc_info(thd, write_status_msg);
2961 2962 2963 2964 2965 2966 2967
        loader->abort(loader);
        loader = NULL;
    }
    abort_loader = false;
    bzero(&lc,sizeof(lc));
    if (error || loader_error) {
        my_errno = error ? error : loader_error;
2968 2969 2970
        if (using_loader) {
            share->try_table_lock = true;
        }
2971
    }
2972
    TOKUDB_DBUG_RETURN(error ? error : loader_error);
2973 2974 2975
}


2976 2977 2978 2979 2980
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;
2981
    u_int64_t cnt = 0;
2982
    char status_msg[MAX_ALIAS_NAME + 200]; //buffer of 200 should be a good upper bound.
2983
    THD* thd = ha_thd();
2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073
    bzero(&key1, sizeof(key1));
    bzero(&key2, sizeof(key2));
    bzero(&val, sizeof(val));
    bzero(&packed_key1, sizeof(packed_key1));
    bzero(&packed_key2, sizeof(packed_key2));
    *is_unique = true;
    
    error = db->cursor(
        db, 
        txn, 
        &tmp_cursor1, 
        0
        );
    if (error) { goto cleanup; }

    error = db->cursor(
        db, 
        txn, 
        &tmp_cursor2,
        0
        );
    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) {
3074 3075 3076 3077 3078 3079
                memcpy(key_buff, key1.data, key1.size);
                place_key_into_mysql_buff(
                    key_info,
                    table->record[0], 
                    (uchar *) key_buff + 1
                    );
3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098
                *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; }
3099 3100

        cnt++;
3101 3102 3103 3104 3105 3106 3107 3108
        if ((cnt % 10000) == 0) {
            sprintf(
                status_msg, 
                "Verifying index uniqueness: Checked %llu of %llu rows in key-%s.", 
                cnt, 
                share->rows, 
                key_info->name);
            thd_proc_info(thd, status_msg);
3109 3110 3111 3112 3113 3114
            if (thd->killed) {
                my_error(ER_QUERY_INTERRUPTED, MYF(0));
                error = ER_QUERY_INTERRUPTED;
                goto cleanup;
            }
        }
3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130
    }

    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;
}

3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208
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;
    bzero((void *)&key, sizeof(key));
    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, 
        0
        );
    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;
3209 3210 3211 3212 3213 3214
            //
            // don't need to do check for primary key
            //
            if (keynr == primary_key) {
                continue;
            }
3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236
            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;
}

3237
void ha_tokudb::test_row_packing(uchar* record, DBT* pk_key, DBT* pk_val) {
3238 3239 3240 3241 3242 3243 3244 3245 3246 3247
    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
3248
    int cmp;
3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280

    bzero(&tmp_pk_key, sizeof(DBT));
    bzero(&tmp_pk_val, sizeof(DBT));

    //
    //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++) {
        u_int32_t tmp_num_bytes = 0;
        uchar* row_desc = NULL;
        u_int32_t desc_size = 0;
        
        if (keynr == primary_key) {
            continue;
        }

        create_dbt_key_from_table(&key, keynr, mult_key_buff[keynr], record, &has_null); 

        //
        // TEST
        //
3281
        row_desc = (uchar *)share->key_file[keynr]->descriptor->dbt.data;
3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299
        row_desc += (*(u_int32_t *)row_desc);
        desc_size = (*(u_int32_t *)row_desc) - 4;
        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);
        cmp = memcmp(key_buff3,mult_key_buff[keynr],tmp_num_bytes);
        assert(cmp == 0);

        //
        // test key packing of clustering keys
        //
        if (table->key_info[keynr].flags & HA_CLUSTERING) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3300
            error = pack_row(&row, (const uchar *) record, keynr);
3301
            assert(error == 0);
3302 3303 3304
            uchar* tmp_buff = NULL;
            tmp_buff = (uchar *)my_malloc(alloced_rec_buff_length,MYF(MY_WME));
            assert(tmp_buff);
3305
            row_desc = (uchar *)share->key_file[keynr]->descriptor->dbt.data;
3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316
            row_desc += (*(u_int32_t *)row_desc);
            row_desc += (*(u_int32_t *)row_desc);
            desc_size = (*(u_int32_t *)row_desc) - 4;
            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
3317
            cmp = memcmp(tmp_buff,rec_buff,tmp_num_bytes);
3318 3319 3320 3321 3322
            assert(cmp == 0);
            my_free(tmp_buff,MYF(MY_ALLOW_ZERO_PTR));
        }
    }

3323 3324 3325 3326
    //
    // copy stuff back out
    //
    error = pack_row(pk_val, (const uchar *) record, primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3327 3328 3329
    assert(pk_val->size == tmp_pk_val.size);
    cmp = memcmp(pk_val->data, tmp_pk_val_data, pk_val->size);    
    assert( cmp == 0);
3330

3331 3332 3333 3334
    my_free(tmp_pk_key_data,MYF(MY_ALLOW_ZERO_PTR));
    my_free(tmp_pk_val_data,MYF(MY_ALLOW_ZERO_PTR));
}

3335 3336 3337 3338
//
// set the put flags for the main dictionary
//
void ha_tokudb::set_main_dict_put_flags(THD* thd, u_int32_t* put_flags) {
3339
    //
3340
    // optimization for "REPLACE INTO..." (and "INSERT IGNORE") command
3341
    // if the command is "REPLACE INTO" and the only table
3342 3343
    // is the main table (or all indexes are a subset of the pk), 
    // then we can simply insert the element
3344 3345 3346
    // 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
3347
    // to do. We cannot do this if otherwise, because then we lose
3348 3349
    // consistency between indexes
    //
3350 3351 3352 3353
    if (hidden_primary_key){
        *put_flags = DB_YESOVERWRITE;
    }
    else if (thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS) && 
3354
        !is_replace_into(thd) && 
3355 3356
        !is_insert_ignore(thd)
        ) 
3357
    {
3358
        *put_flags = DB_YESOVERWRITE;
3359
    }
3360 3361 3362
    else if (do_ignore_flag_optimization(thd,table,share->replace_into_fast) && 
        is_replace_into(thd)
        ) 
3363 3364
    {
        *put_flags = DB_YESOVERWRITE;
3365
    }
3366 3367 3368
    else if (do_ignore_flag_optimization(thd,table,share->replace_into_fast) && 
        is_insert_ignore(thd)
        ) 
3369 3370 3371 3372 3373 3374
    {
        *put_flags = DB_NOOVERWRITE_NO_ERROR;
    }
    else 
    {
        *put_flags = DB_NOOVERWRITE;
3375
    }
3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387
}

int ha_tokudb::insert_row_to_main_dictionary(uchar* record, DBT* pk_key, DBT* pk_val, DB_TXN* txn) {
    int error = 0;
    u_int32_t put_flags = 0;
    THD *thd = ha_thd();
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
    ulonglong wait_lock_time = get_write_lock_wait_time(thd);

    assert(curr_num_DBs == 1);
    
    set_main_dict_put_flags(thd,&put_flags);
3388

3389
    lockretryN(wait_lock_time){
3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405
        error = share->file->put(
            share->file, 
            txn, 
            pk_key,
            pk_val, 
            put_flags
            );
        lockretry_wait;
    }

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

cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3406 3407 3408
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3409
int ha_tokudb::insert_rows_to_dictionaries_mult(DBT* pk_key, DBT* pk_val, DB_TXN* txn, THD* thd) {
3410
    int error = 0;
3411
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
3412
    ulonglong wait_lock_time = get_write_lock_wait_time(thd);
3413
    u_int32_t mult_put_flags[MAX_KEY + 1] = {DB_YESOVERWRITE};
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3414

3415
    set_main_dict_put_flags(thd, &mult_put_flags[primary_key]);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3416 3417 3418 3419 3420 3421 3422
    if (mult_put_flags[primary_key] == DB_NOOVERWRITE_NO_ERROR) {
        //
        //hopefully temporary, right now, put_multiple does not
        // support use of DB_NOOVERWRITE_NO_ERROR as put_flag
        //
        mult_put_flags[primary_key] = DB_NOOVERWRITE;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3423
    
3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434
    lockretryN(wait_lock_time){
        error = db_env->put_multiple(
            db_env, 
            NULL, 
            txn, 
            pk_key, 
            pk_val,
            curr_num_DBs, 
            share->key_file, 
            mult_key_dbt,
            mult_rec_dbt,
3435
            mult_put_flags, 
3436 3437 3438 3439
            NULL
            );
        lockretry_wait;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3440

3441 3442 3443 3444 3445 3446
    //
    // 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
3447 3448 3449
    }
    return error;
}
3450

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3451 3452 3453 3454 3455 3456 3457 3458
//
// 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
//
3459
int ha_tokudb::write_row(uchar * record) {
3460
    TOKUDB_DBUG_ENTER("ha_tokudb::write_row");
3461
    DBT row, prim_key;
3462
    int error;
3463
    THD *thd = ha_thd();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3464
    bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3465 3466
    DB_TXN* sub_trans = NULL;
    DB_TXN* txn = NULL;
3467
    tokudb_trx_data *trx = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3468
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
3469
    bool create_sub_trans = false;
3470

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3471 3472 3473 3474
    //
    // 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
    //
3475
    statistic_increment(table->in_use->status_var.ha_write_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3476
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT) {
3477
        table->timestamp_field->set_time();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3478 3479
    }
    if (table->next_number_field && record == table->record[0]) {
3480
        update_auto_increment();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3481
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492

    //
    // 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
3493
            field_offset(table->field[share->ai_field_index], table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3494 3495 3496
            record
            );
        if (curr_auto_inc > share->last_auto_increment) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3497
            share->last_auto_increment = curr_auto_inc;
3498 3499
            if (delay_updating_ai_metadata) {
                ai_metadata_update_required = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3500
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3501 3502 3503 3504
            else {
                update_max_auto_inc(share->status_block, share->last_auto_increment);
            }
         }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3505 3506 3507
        pthread_mutex_unlock(&share->mutex);
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3508 3509 3510 3511
    
    if (hidden_primary_key) {
        get_auto_primary_key(current_ident);
    }
3512

3513 3514 3515
    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
3516 3517 3518
            goto cleanup;
        }
    }
3519

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3520 3521
    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))){
3522
        goto cleanup;
3523 3524
    }

3525
    create_sub_trans = (using_ignore && !(do_ignore_flag_optimization(thd,table,share->replace_into_fast)));
3526
    if (create_sub_trans) {
3527
        error = db_env->txn_begin(db_env, transaction, &sub_trans, DB_INHERIT_ISOLATION);
3528 3529 3530
        if (error) {
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3531
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3532
    
3533
    txn = create_sub_trans ? sub_trans : transaction;    
3534

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3535
    //
3536
    // make sure the buffers for the rows are big enough
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3537
    //
3538
    fix_mult_rec_buff();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3539

3540
    if (tokudb_debug & TOKUDB_DEBUG_CHECK_KEY) {
3541
        test_row_packing(record,&prim_key,&row);
3542
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3543

3544 3545 3546 3547 3548 3549
    if (loader) {
        error = loader->put(loader, &prim_key, &row);
        if (error) {
            abort_loader = true;
            goto cleanup;
        }
3550 3551
    }
    else {
3552 3553 3554 3555 3556 3557 3558
        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);
            if (error) { goto cleanup; }
3559

3560 3561 3562
            error = insert_rows_to_dictionaries_mult(&prim_key, &row, txn, thd);
            if (error) { goto cleanup; }
        }
3563
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3564

3565
    trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3566 3567
    if (!error) {
        added_rows++;
3568 3569
        trx->stmt_progress.inserted++;
        track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3570 3571
    }
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3572
    if (error == DB_KEYEXIST) {
3573
        error = HA_ERR_FOUND_DUPP_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3574
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3575 3576 3577 3578 3579
    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) {
3580
            abort_txn(sub_trans);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3581 3582
        }
        else {
3583
            commit_txn(sub_trans, DB_TXN_NOSYNC);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3584 3585
        }
    }
3586
    TOKUDB_DBUG_RETURN(error);
3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606
}

/* Compare if a key in a row has changed */
int ha_tokudb::key_cmp(uint keynr, const uchar * old_row, const uchar * new_row) {
    KEY_PART_INFO *key_part = table->key_info[keynr].key_part;
    KEY_PART_INFO *end = key_part + table->key_info[keynr].key_parts;

    for (; key_part != end; key_part++) {
        if (key_part->null_bit) {
            if ((old_row[key_part->null_offset] & key_part->null_bit) != (new_row[key_part->null_offset] & key_part->null_bit))
                return 1;
        }
        if (key_part->key_part_flag & (HA_BLOB_PART | HA_VAR_LENGTH_PART)) {

            if (key_part->field->cmp_binary((uchar *) (old_row + key_part->offset), (uchar *) (new_row + key_part->offset), (ulong) key_part->length))
                return 1;
        } else {
            if (memcmp(old_row + key_part->offset, new_row + key_part->offset, key_part->length))
                return 1;
        }
3607
    }
3608
    return 0;
3609 3610
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3611 3612 3613 3614 3615 3616 3617 3618 3619
//
// 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
//
3620
int ha_tokudb::update_row(const uchar * old_row, uchar * new_row) {
3621
    TOKUDB_DBUG_ENTER("update_row");
3622
    DBT prim_key, key, old_prim_key, row, prim_row;
3623 3624
    int error;
    bool primary_key_changed;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3625 3626
    bool has_null;
    THD* thd = ha_thd();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3627 3628
    DB_TXN* sub_trans = NULL;
    DB_TXN* txn = NULL;
3629 3630 3631 3632 3633 3634
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
    u_int32_t mult_put_flags[MAX_KEY + 1] = {DB_YESOVERWRITE};
    DB* dbs[MAX_KEY + 1];
    DBT key_dbts[MAX_KEY + 1];
    DBT rec_dbts[MAX_KEY + 1];
    u_int32_t curr_db_index;
3635
    ulonglong wait_lock_time = get_write_lock_wait_time(thd);
3636 3637

    LINT_INIT(error);
3638 3639 3640 3641 3642 3643 3644 3645 3646
    bzero((void *) &row, sizeof(row));
    bzero((void *) &prim_key, sizeof(prim_key));
    bzero((void *) &old_prim_key, sizeof(old_prim_key));
    bzero((void *) &prim_row, sizeof(prim_row));
    bzero((void *) &key, sizeof(key));
    bzero((void *) &key_dbts, sizeof(key));
    bzero((void *) &rec_dbts, sizeof(key));


3647
    statistic_increment(table->in_use->status_var.ha_update_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3648
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_UPDATE) {
3649
        table->timestamp_field->set_time();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3650
    }
3651

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3652 3653 3654 3655 3656 3657 3658 3659 3660 3661
    //
    // 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
3662
            field_offset(table->field[share->ai_field_index], table),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673
            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);
    }

3674 3675 3676 3677 3678
    if (using_ignore) {
        error = db_env->txn_begin(db_env, transaction, &sub_trans, DB_INHERIT_ISOLATION);
        if (error) {
            goto cleanup;
        }
3679
    }
3680
    txn = using_ignore ? sub_trans : transaction;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3681 3682


3683 3684 3685 3686 3687 3688
    if (hidden_primary_key) {
        primary_key_changed = 0;
        bzero((void *) &prim_key, sizeof(prim_key));
        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
3689 3690
    } 
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3691
        create_dbt_key_from_table(&prim_key, primary_key, key_buff, new_row, &has_null);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3692
        if ((primary_key_changed = key_cmp(primary_key, old_row, new_row))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3693
            create_dbt_key_from_table(&old_prim_key, primary_key, primary_key_buff, old_row, &has_null);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3694 3695
        }
        else {
3696
            old_prim_key = prim_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3697
        }
3698 3699
    }

3700 3701 3702 3703 3704 3705
    if (primary_key_changed) {
        // Primary key changed or we are updating a key that can have duplicates.
        // Delete the old row and add a new one
        error = remove_key(txn, primary_key, old_row, &old_prim_key);
        if (error) { goto cleanup; }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3706

3707 3708 3709 3710
    error = pack_row(&prim_row, new_row, primary_key);
    if (error) { goto cleanup; }


3711 3712 3713 3714 3715
    dbs[0] = share->key_file[primary_key];
    key_dbts[0] = prim_key;
    rec_dbts[0] = prim_row;
    mult_put_flags[0] = primary_key_changed ? DB_NOOVERWRITE : DB_YESOVERWRITE;

3716
    curr_db_index = 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3717 3718
    // Update all other keys
    for (uint keynr = 0; keynr < table_share->keys; keynr++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3719
        bool secondary_key_changed = key_cmp(keynr, old_row, new_row);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3720 3721 3722
        if (keynr == primary_key) {
            continue;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3723
        if (table->key_info[keynr].flags & HA_CLUSTERING ||
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3724
            secondary_key_changed || 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3725 3726 3727
            primary_key_changed
            ) 
        {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3728
            bool is_unique_key = table->key_info[keynr].flags & HA_NOSAME;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3729 3730 3731 3732 3733 3734 3735 3736 3737 3738
            //
            // only remove the old value if the key has changed 
            // if the key has not changed (in case of clustering keys, 
            // then we overwrite  the old value)
            // 
            if (secondary_key_changed || primary_key_changed) {
                error = remove_key(txn, keynr, old_row, &old_prim_key);
                if (error) {
                    goto cleanup;
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3739
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3740 3741 3742 3743 3744 3745

            //
            // 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
            //
3746
            if (is_unique_key && !thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3747 3748 3749 3750 3751 3752 3753 3754
                bool is_unique = false;
                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
3755
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3756

3757 3758 3759 3760
            dbs[curr_db_index] = share->key_file[keynr];
            key_dbts[curr_db_index] = mult_key_dbt[keynr];
            rec_dbts[curr_db_index] = mult_rec_dbt[keynr];
            curr_db_index++;
3761 3762
        }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3763

3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778
    lockretryN(wait_lock_time){
        error = db_env->put_multiple(
            db_env, 
            NULL, 
            txn, 
            &prim_key, 
            &prim_row,
            curr_db_index, 
            dbs, 
            key_dbts,
            rec_dbts,
            mult_put_flags, 
            NULL
            );
        lockretry_wait;
3779
    }
3780 3781 3782 3783
    if (error == DB_KEYEXIST) {
        last_dup_key = primary_key;
    }    
    else if (!error) {
3784 3785
        trx->stmt_progress.updated++;
        track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3786
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3787 3788


Zardosht Kasheff's avatar
Zardosht Kasheff committed
3789 3790
cleanup:
    if (error == DB_KEYEXIST) {
3791
        error = HA_ERR_FOUND_DUPP_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3792
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3793 3794 3795 3796 3797
    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) {
3798
            abort_txn(sub_trans);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3799 3800
        }
        else {
3801
            commit_txn(sub_trans, DB_TXN_NOSYNC);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3802 3803
        }
    }
3804
    TOKUDB_DBUG_RETURN(error);
3805 3806
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820
//
//
// Delete one key in key_file[keynr]
// This uses key_buff2, when keynr != primary key, so it's important that
// a function that calls this doesn't use this buffer for anything else.
// Parameters:
//      [in]    trans - transaction to be used for the delete
//              keynr - index for which a key needs to be deleted
//      [in]    record - row in MySQL format. Must delete a key for this row
//      [in]    prim_key - key for record in primary table
// Returns:
//      0 on success
//      error otherwise
//
3821
int ha_tokudb::remove_key(DB_TXN * trans, uint keynr, const uchar * record, DBT * prim_key) {
3822
    TOKUDB_DBUG_ENTER("ha_tokudb::remove_key");
3823
    int error = 0;
3824
    DBT key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3825
    bool has_null;
3826
    ulonglong wait_lock_time = get_write_lock_wait_time(ha_thd());
3827
    DBUG_PRINT("enter", ("index: %d", keynr));
3828 3829
    DBUG_PRINT("primary", ("index: %d", primary_key));
    DBUG_DUMP("prim_key", (uchar *) prim_key->data, prim_key->size);
3830

3831
    if (keynr == primary_key) {  // Unique key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3832
        DBUG_PRINT("Primary key", ("index: %d", keynr));
3833 3834 3835 3836
        lockretryN(wait_lock_time){
            error = share->key_file[keynr]->del(share->key_file[keynr], trans, prim_key , DB_DELETE_ANY);
            lockretry_wait;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3837
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3838
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3839
        DBUG_PRINT("Secondary key", ("index: %d", keynr));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3840
        create_dbt_key_from_table(&key, keynr, key_buff2, record, &has_null);
3841 3842 3843 3844
        lockretryN(wait_lock_time){
            error = share->key_file[keynr]->del(share->key_file[keynr], trans, &key , DB_DELETE_ANY);
            lockretry_wait;
        }
3845
    }
3846
    TOKUDB_DBUG_RETURN(error);
3847 3848
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860
//
// Delete all keys for new_record
// Parameters:
//      [in]    trans - transaction to be used for the delete
//      [in]    record - row in MySQL format. Must delete all keys for this row
//      [in]    prim_key - key for record in primary table
//      [in]    keys - array that states if a key is set, and hence needs 
//                  removal
// Returns:
//      0 on success
//      error otherwise
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3861
int ha_tokudb::remove_keys(DB_TXN * trans, const uchar * record, DBT * prim_key) {
3862 3863
    int result = 0;
    for (uint keynr = 0; keynr < table_share->keys + test(hidden_primary_key); keynr++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3864 3865 3866 3867
        int new_error = remove_key(trans, keynr, record, prim_key);
        if (new_error) {
            result = new_error;     // Return last error
            break;          // Let rollback correct things
3868 3869 3870 3871 3872
        }
    }
    return result;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3873
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3874
// Deletes a row in the table, called when handling a DELETE query
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3875 3876 3877 3878 3879 3880
// Parameters:
//      [in]    record - row to be deleted, in MySQL format
// Returns:
//      0 on success
//      error otherwise
//
3881
int ha_tokudb::delete_row(const uchar * record) {
3882
    TOKUDB_DBUG_ENTER("ha_tokudb::delete_row");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3883 3884
    int error = ENOSYS;
    DBT prim_key;
3885
    key_map keys = table_share->keys_in_use;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3886
    bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3887
    THD* thd = ha_thd();
3888
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3889

3890 3891
    statistic_increment(table->in_use->status_var.ha_delete_count, &LOCK_status);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3892
    create_dbt_key_from_table(&prim_key, primary_key, key_buff, record, &has_null);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3893
    if (hidden_primary_key) {
3894
        keys.set_bit(primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3895
    }
3896 3897 3898
    /* Subtransactions may be used in order to retry the delete in
       case we get a DB_LOCK_DEADLOCK error. */
    DB_TXN *sub_trans = transaction;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3899
    error = remove_keys(sub_trans, record, &prim_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3900 3901
    if (error) {
        DBUG_PRINT("error", ("Got error %d", error));
3902
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3903
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3904
        deleted_rows++;
3905 3906
        trx->stmt_progress.deleted++;
        track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3907
    }
3908
    TOKUDB_DBUG_RETURN(error);
3909 3910
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926
//
// 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) {
    u_int32_t curr_fixed_col_index = 0;
    u_int32_t curr_var_col_index = 0;
    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
3927
    if (keynr == primary_key || keynr == MAX_KEY) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3928 3929 3930 3931 3932 3933 3934 3935 3936 3937
        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)
            ) 
        {
3938
            if (bitmap_is_set(&share->kc_info.key_filters[key_index],i)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3939 3940 3941 3942 3943 3944
                read_key = true;
            }
            else {
                //
                // if fixed field length
                //
3945
                if (share->kc_info.field_lengths[i] != 0) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3946 3947 3948 3949 3950 3951 3952 3953 3954
                    //
                    // save the offset into the list
                    //
                    fixed_cols_for_query[curr_fixed_col_index] = i;
                    curr_fixed_col_index++;
                }
                //
                // varchar or varbinary
                //
3955
                else if (share->kc_info.length_bytes[i] != 0) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971
                    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
3972
void ha_tokudb::column_bitmaps_signal() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3973 3974 3975 3976
    //
    // if we have max number of indexes, then MAX_KEY == primary_key
    //
    if (active_index != MAX_KEY || active_index == primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3977 3978 3979
        set_query_columns(active_index);
    }
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3980

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3981 3982 3983 3984 3985 3986 3987 3988
//
// 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() {
3989
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3990
    DB* db = share->key_file[active_index];
3991
    lockretryN(read_lock_wait_time){
3992 3993 3994 3995 3996 3997 3998 3999
        error = db->pre_acquire_read_lock(
            db, 
            transaction, 
            db->dbt_neg_infty(), db->dbt_neg_infty(), 
            db->dbt_pos_infty(), db->dbt_pos_infty()
            );
        lockretry_wait;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4000 4001 4002 4003 4004 4005 4006 4007 4008
    if (error) { last_cursor_error = error; goto cleanup; }

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


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019
//
// 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
//
int ha_tokudb::prepare_index_key_scan( const uchar * key, uint key_len ) {
    int error = 0;
    DBT start_key, end_key;
    pack_key(&start_key, active_index, key_buff, key, key_len, COL_NEG_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4020
    pack_key(&end_key, active_index, key_buff2, key, key_len, COL_POS_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4021

4022
    lockretryN(read_lock_wait_time){
4023 4024 4025 4026 4027 4028 4029 4030 4031 4032
        error = share->key_file[active_index]->pre_acquire_read_lock(
            share->key_file[active_index], 
            transaction, 
            &start_key, 
            share->key_file[active_index]->dbt_neg_infty(), 
            &end_key, 
            share->key_file[active_index]->dbt_pos_infty()
            );
        lockretry_wait;            
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045
    if (error){ 
        goto cleanup; 
    }

    range_lock_grabbed = true;
    error = 0;
cleanup:
    if (error) {
        last_cursor_error = error;
        //
        // cursor should be initialized here, but in case it is not, we still check
        //
        if (cursor) {
4046 4047
            int r = cursor->c_close(cursor);
            assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4048 4049 4050 4051 4052 4053 4054 4055
            cursor = NULL;
        }
    }
    return error;
}



Zardosht Kasheff's avatar
Zardosht Kasheff committed
4056 4057 4058 4059 4060 4061 4062 4063 4064
//
// 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
//
4065
int ha_tokudb::index_init(uint keynr, bool sorted) {
4066
    TOKUDB_DBUG_ENTER("ha_tokudb::index_init %p %d", this, keynr);
4067
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4068
    THD* thd = ha_thd(); 
4069
    DBUG_PRINT("enter", ("table: '%s'  key: %d", table_share->table_name.str, keynr));
4070
    read_lock_wait_time = get_read_lock_wait_time(ha_thd());
4071 4072 4073 4074 4075 4076 4077

    /*
       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"));
4078 4079
        int r = cursor->c_close(cursor);
        assert(r==0);
4080 4081
    }
    active_index = keynr;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4082
    last_cursor_error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4083
    range_lock_grabbed = false;
4084
    DBUG_ASSERT(keynr <= table->s->keys);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4085
    DBUG_ASSERT(share->key_file[keynr]);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4086
    if ((error = share->key_file[keynr]->cursor(share->key_file[keynr], transaction, &cursor, 0))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4087
        last_cursor_error = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4088
        cursor = NULL;             // Safety
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4089
        goto exit;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4090
    }
4091
    bzero((void *) &last_key, sizeof(last_key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4092 4093 4094 4095 4096 4097 4098 4099 4100 4101

    if (thd_sql_command(thd) == SQLCOM_SELECT) {
        set_query_columns(keynr);
        unpack_entire_row = false;
    }
    else {
        unpack_entire_row = true;
    }
    error = 0;
exit:
4102
    TOKUDB_DBUG_RETURN(error);
4103 4104
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4105 4106 4107
//
// closes the local cursor
//
4108
int ha_tokudb::index_end() {
4109
    TOKUDB_DBUG_ENTER("ha_tokudb::index_end %p", this);
4110
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4111
    range_lock_grabbed = false;
4112 4113 4114
    if (cursor) {
        DBUG_PRINT("enter", ("table: '%s'", table_share->table_name.str));
        error = cursor->c_close(cursor);
4115
        assert(error==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4116
        cursor = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4117
        last_cursor_error = 0;
4118
    }
4119
    active_index = MAX_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4120 4121 4122 4123 4124 4125 4126 4127 4128

    //
    // 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;
4129
    TOKUDB_DBUG_RETURN(error);
4130 4131
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4132 4133 4134

int ha_tokudb::handle_cursor_error(int error, int err_to_return, uint keynr) {
    TOKUDB_DBUG_ENTER("ha_tokudb::handle_cursor_error");
4135
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4136
        last_cursor_error = error;
4137
        table->status = STATUS_NOT_FOUND;
4138 4139
        int r = cursor->c_close(cursor);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4140
        cursor = NULL;
Barry Perlman's avatar
Barry Perlman committed
4141
        if (error == DB_NOTFOUND) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4142 4143
            error = err_to_return;
            if ((share->key_file[keynr]->cursor(share->key_file[keynr], transaction, &cursor, 0))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4144 4145 4146
                cursor = NULL;             // Safety
            }
        }
4147
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4148 4149 4150 4151 4152 4153 4154 4155 4156 4157
    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
4158
void ha_tokudb::extract_hidden_primary_key(uint keynr, DBT const *found_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4159
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4160
    // extract hidden primary key to current_ident
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4161 4162
    //
    if (hidden_primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4163 4164 4165
        if (keynr == primary_key) {
            memcpy_fixed(current_ident, (char *) found_key->data, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4166
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4167
        // if secondary key, hidden primary key is at end of found_key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4168
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4169
        else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4170 4171 4172 4173 4174 4175
            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
4176
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4177 4178
}

4179

4180
int ha_tokudb::read_row_callback (uchar * buf, uint keynr, DBT const *row, DBT const *found_key) {
4181
    assert(keynr == primary_key);
4182
    return unpack_row(buf, row,found_key, keynr);
4183 4184
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4185 4186 4187 4188 4189 4190 4191 4192 4193 4194
//
// 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
4195
void ha_tokudb::read_key_only(uchar * buf, uint keynr, DBT const *found_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4196 4197
    TOKUDB_DBUG_ENTER("ha_tokudb::read_key_only");
    table->status = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4198 4199 4200 4201
    //
    // 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
4202 4203 4204
    if (!(hidden_primary_key && keynr == primary_key)) {
        unpack_key(buf, found_key, keynr);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217
    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
4218
int ha_tokudb::read_primary_key(uchar * buf, uint keynr, DBT const *row, DBT const *found_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4219
    TOKUDB_DBUG_ENTER("ha_tokudb::read_primary_key");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4220
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4221
    table->status = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4222 4223 4224 4225
    //
    // 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
4226
        bool has_null;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4227
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4228 4229
        // 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
4230 4231
        //
        bzero((void *) &last_key, sizeof(last_key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4232 4233 4234 4235 4236 4237 4238 4239 4240 4241
        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
4242
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4243
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4244
    // else read from clustered/primary key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4245
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4246
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4247 4248
        error = unpack_row(buf, row, found_key, keynr);
        if (error) { goto exit; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4249 4250
    }
    if (found_key) { DBUG_DUMP("read row key", (uchar *) found_key->data, found_key->size); }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4251 4252 4253
    error = 0;
exit:
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265
}

//
// 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");
4266
    int error = 0;
4267 4268 4269 4270
    struct smart_dbt_info info;
    info.ha = this;
    info.buf = buf;
    info.keynr = primary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4271
    //
4272
    // assumes key is stored in this->last_key
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4273
    //
4274
    lockretryN(read_lock_wait_time){
4275 4276 4277 4278 4279 4280 4281 4282 4283 4284
        error = share->file->getf_set(
            share->file, 
            transaction, 
            0, 
            &last_key, 
            smart_dbt_callback_rowread_ptquery, 
            &info
            );
        lockretry_wait;
    }
4285
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4286 4287 4288 4289
        table->status = STATUS_NOT_FOUND;
        TOKUDB_DBUG_RETURN(error == DB_NOTFOUND ? HA_ERR_CRASHED : error);
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4290
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4291 4292 4293
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306
// 
// 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) { 
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next_same %p", this); 
4307
    int error = 0; 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4308 4309 4310 4311 4312
    struct smart_dbt_info info; 
    DBT curr_key;
    DBT found_key;
    bool has_null;
    int cmp;
Rich Prohaska's avatar
Rich Prohaska committed
4313
    u_int32_t flags;
4314 4315
    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
4316 4317 4318 4319 4320 4321 4322
    HANDLE_INVALID_CURSOR(); 

    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status); 
    info.ha = this; 
    info.buf = buf; 
    info.keynr = active_index; 

4323
    pack_key(&curr_key, active_index, key_buff2, key, keylen, COL_ZERO);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4324

Rich Prohaska's avatar
Rich Prohaska committed
4325
    flags = SET_READ_FLAG(0); 
4326
    lockretryN(read_lock_wait_time){
4327 4328 4329 4330
        error = cursor->c_getf_next(cursor, flags, SMART_DBT_CALLBACK, &info);
        lockretry_wait;
    }
    error = handle_cursor_error(error, HA_ERR_END_OF_FILE,active_index);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347
    if (error) {
        goto cleanup;
    }
    if (!key_read && active_index != primary_key && !(table->key_info[active_index].flags & HA_CLUSTERING)) { 
        error = read_full_row(buf); 
        if (error) {
            goto cleanup;
        }
    } 
    //
    // now do the comparison
    //
    create_dbt_key_from_table(&found_key,active_index,key_buff3,buf,&has_null);
    cmp = tokudb_prefix_cmp_dbt_key(share->key_file[active_index], &curr_key, &found_key);
    if (cmp) {
        error = HA_ERR_END_OF_FILE; 
    }
4348 4349
    trx->stmt_progress.queried++;
    track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4350 4351 4352 4353 4354
 cleanup: 
    TOKUDB_DBUG_RETURN(error); 
 } 


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371
//
// 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
//
4372 4373
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);
4374
    // TOKUDB_DBUG_DUMP("key=", key, key_len);
4375
    DBT row;
4376
    DBT lookup_key;
4377
    int error = 0;    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4378
    u_int32_t flags = 0;
4379 4380
    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
4381
    struct smart_dbt_info info;
4382
    struct index_read_info ir_info;
4383

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4384
    HANDLE_INVALID_CURSOR();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4385

4386 4387 4388
    table->in_use->status_var.ha_read_key_count++;
    bzero((void *) &row, sizeof(row));

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4389 4390 4391 4392
    info.ha = this;
    info.buf = buf;
    info.keynr = active_index;

4393
    ir_info.smart_dbt_info = info;
4394
    ir_info.cmp = 0;
4395

4396
    flags = SET_READ_FLAG(0);
4397
    switch (find_flag) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4398
    case HA_READ_KEY_EXACT: /* Find first record else error */
4399
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_NEG_INF);
4400
        ir_info.orig_key = &lookup_key;
4401 4402 4403 4404
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range(cursor, flags, &lookup_key, SMART_DBT_IR_CALLBACK, &ir_info);
            lockretry_wait;
        }
4405 4406
        if (ir_info.cmp) {
            error = DB_NOTFOUND;
4407
        }
4408
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4409
    case HA_READ_AFTER_KEY: /* Find next rec. after key-record */
4410
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_POS_INF);
4411 4412 4413 4414
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range(cursor, flags, &lookup_key, SMART_DBT_CALLBACK, &info);
            lockretry_wait;
        }
4415
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4416
    case HA_READ_BEFORE_KEY: /* Find next rec. before key-record */
4417
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_NEG_INF);
4418 4419 4420 4421
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range_reverse(cursor, flags, &lookup_key, SMART_DBT_CALLBACK, &info);
            lockretry_wait;
        }
4422
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4423
    case HA_READ_KEY_OR_NEXT: /* Record or next record */
4424
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_NEG_INF);
4425 4426 4427 4428
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range(cursor, flags, &lookup_key, SMART_DBT_CALLBACK, &info);
            lockretry_wait;
        }
4429
        break;
4430 4431 4432
    //
    // This case does not seem to ever be used, it is ok for it to be slow
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4433
    case HA_READ_KEY_OR_PREV: /* Record or previous */
4434
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_NEG_INF);
4435
        ir_info.orig_key = &lookup_key;
4436 4437 4438 4439
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range(cursor, flags, &lookup_key, SMART_DBT_IR_CALLBACK, &ir_info);
            lockretry_wait;
        }
4440
        if (error == DB_NOTFOUND) {
4441 4442
            error = cursor->c_getf_last(cursor, flags, SMART_DBT_CALLBACK, &info);
        }
4443 4444 4445
        else if (ir_info.cmp) {
            error = cursor->c_getf_prev(cursor, flags, SMART_DBT_CALLBACK, &info);
        }
4446
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4447
    case HA_READ_PREFIX_LAST_OR_PREV: /* Last or prev key with the same prefix */
4448
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_POS_INF);
4449 4450 4451 4452
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range_reverse(cursor, flags, &lookup_key, SMART_DBT_CALLBACK, &info);
            lockretry_wait;
        }
4453
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4454
    case HA_READ_PREFIX_LAST:
4455
        pack_key(&lookup_key, active_index, key_buff3, key, key_len, COL_POS_INF);
4456
        ir_info.orig_key = &lookup_key;
4457 4458 4459 4460
        lockretryN(read_lock_wait_time){
            error = cursor->c_getf_set_range_reverse(cursor, flags, &lookup_key, SMART_DBT_IR_CALLBACK, &ir_info);
            lockretry_wait;
        }
4461 4462
        if (ir_info.cmp) {
            error = DB_NOTFOUND;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4463 4464
        }
        break;
4465
    default:
4466
        TOKUDB_TRACE("unsupported:%d\n", find_flag);
4467
        error = HA_ERR_UNSUPPORTED;
4468
        break;
4469
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4470
    error = handle_cursor_error(error,HA_ERR_KEY_NOT_FOUND,active_index);
4471
    if (!error && !key_read && active_index != primary_key && !(table->key_info[active_index].flags & HA_CLUSTERING)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4472
        error = read_full_row(buf);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4473
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4474 4475 4476 4477
    
    if (error && (tokudb_debug & TOKUDB_DEBUG_ERROR)) {
        TOKUDB_TRACE("error:%d:%d\n", error, find_flag);
    }
4478 4479 4480
    trx->stmt_progress.queried++;
    track_progress(thd);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4481
cleanup:
4482
    TOKUDB_DBUG_RETURN(error);
4483 4484
}

4485 4486


Zardosht Kasheff's avatar
Zardosht Kasheff committed
4487 4488 4489 4490 4491 4492 4493 4494 4495
//
// 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
//
4496
int ha_tokudb::index_next(uchar * buf) {
4497
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next");
4498
    int error = 0; 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4499 4500
    struct smart_dbt_info info;
    u_int32_t flags = SET_READ_FLAG(0);
4501 4502
    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
4503
    HANDLE_INVALID_CURSOR();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4504
    
4505
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4506 4507 4508 4509

    info.ha = this;
    info.buf = buf;
    info.keynr = active_index;
4510

4511
    lockretryN(read_lock_wait_time){
4512 4513 4514 4515
        error = cursor->c_getf_next(cursor, flags, SMART_DBT_CALLBACK, &info);
        lockretry_wait;
    }
    error = handle_cursor_error(error, HA_ERR_END_OF_FILE,active_index);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4516 4517 4518 4519
    //
    // still need to get entire contents of the row if operation done on
    // secondary DB and it was NOT a covering index
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4520
    if (!error && !key_read && (active_index != primary_key) && !(table->key_info[active_index].flags & HA_CLUSTERING) ) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4521 4522
        error = read_full_row(buf);
    }
4523 4524
    trx->stmt_progress.queried++;
    track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4525
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4526
    TOKUDB_DBUG_RETURN(error);
4527 4528
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4529 4530 4531 4532 4533 4534

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
4535 4536 4537 4538 4539 4540 4541 4542 4543
//
// 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
//
4544
int ha_tokudb::index_prev(uchar * buf) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4545
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next");
4546
    int error = 0; 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4547 4548
    struct smart_dbt_info info;
    u_int32_t flags = SET_READ_FLAG(0);
4549 4550
    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
4551 4552 4553 4554 4555 4556 4557
    HANDLE_INVALID_CURSOR();
    
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);

    info.ha = this;
    info.buf = buf;
    info.keynr = active_index;
4558

4559
    lockretryN(read_lock_wait_time){
4560 4561 4562 4563 4564
        error = cursor->c_getf_prev(cursor, flags, SMART_DBT_CALLBACK, &info);
        lockretry_wait;
    }
    error = handle_cursor_error(error,HA_ERR_END_OF_FILE,active_index);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4565 4566 4567 4568
    //
    // still need to get entire contents of the row if operation done on
    // secondary DB and it was NOT a covering index
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4569
    if (!error && !key_read && (active_index != primary_key) && !(table->key_info[active_index].flags & HA_CLUSTERING) ) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4570 4571
        error = read_full_row(buf);
    }
4572 4573 4574
    trx->stmt_progress.queried++;
    track_progress(thd);
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4575
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4576
    TOKUDB_DBUG_RETURN(error);
4577 4578
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4579 4580 4581 4582 4583 4584 4585 4586 4587
//
// 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
//
4588
int ha_tokudb::index_first(uchar * buf) {
4589
    TOKUDB_DBUG_ENTER("ha_tokudb::index_first");
4590
    int error = 0;
4591 4592
    struct smart_dbt_info info;
    u_int32_t flags = SET_READ_FLAG(0);
4593 4594
    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
4595
    HANDLE_INVALID_CURSOR();
4596

4597
    statistic_increment(table->in_use->status_var.ha_read_first_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4598

4599 4600 4601 4602
    info.ha = this;
    info.buf = buf;
    info.keynr = active_index;

4603
    lockretryN(read_lock_wait_time){
4604 4605 4606 4607 4608
        error = cursor->c_getf_first(cursor, flags, SMART_DBT_CALLBACK, &info);
        lockretry_wait;
    }
    error = handle_cursor_error(error,HA_ERR_END_OF_FILE,active_index);

4609 4610 4611 4612 4613 4614
    //
    // still need to get entire contents of the row if operation done on
    // secondary DB and it was NOT a covering index
    //
    if (!error && !key_read && (active_index != primary_key) && !(table->key_info[active_index].flags & HA_CLUSTERING) ) {
        error = read_full_row(buf);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4615
    }
4616 4617 4618
    trx->stmt_progress.queried++;
    track_progress(thd);
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4619
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4620
    TOKUDB_DBUG_RETURN(error);
4621 4622
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4623 4624 4625 4626 4627 4628 4629 4630 4631
//
// 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
//
4632
int ha_tokudb::index_last(uchar * buf) {
4633
    TOKUDB_DBUG_ENTER("ha_tokudb::index_last");
4634
    int error = 0;
4635 4636
    struct smart_dbt_info info;
    u_int32_t flags = SET_READ_FLAG(0);
4637 4638
    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
4639
    HANDLE_INVALID_CURSOR();
4640

4641
    statistic_increment(table->in_use->status_var.ha_read_last_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4642

4643 4644 4645 4646
    info.ha = this;
    info.buf = buf;
    info.keynr = active_index;

4647
    lockretryN(read_lock_wait_time){
4648 4649 4650 4651
        error = cursor->c_getf_last(cursor, flags, SMART_DBT_CALLBACK, &info);
        lockretry_wait;
    }
    error = handle_cursor_error(error,HA_ERR_END_OF_FILE,active_index);
4652 4653 4654 4655 4656 4657
    //
    // still need to get entire contents of the row if operation done on
    // secondary DB and it was NOT a covering index
    //
    if (!error && !key_read && (active_index != primary_key) && !(table->key_info[active_index].flags & HA_CLUSTERING) ) {
        error = read_full_row(buf);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4658
    }
4659 4660 4661 4662 4663

    if (trx) {
        trx->stmt_progress.queried++;
    }
    track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4664
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4665
    TOKUDB_DBUG_RETURN(error);
4666 4667
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4668 4669 4670 4671 4672 4673 4674 4675
//
// 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
//
4676
int ha_tokudb::rnd_init(bool scan) {
4677
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_init");
4678
    int error = 0;
4679
    read_lock_wait_time = get_read_lock_wait_time(ha_thd());
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4680
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4681 4682
    if (scan) {
        DB* db = share->key_file[primary_key];
4683
        lockretryN(read_lock_wait_time){
4684 4685 4686
            error = db->pre_acquire_read_lock(db, transaction, db->dbt_neg_infty(), NULL, db->dbt_pos_infty(), NULL);
            lockretry_wait;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4687
        if (error) { last_cursor_error = error; goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4688 4689
    }
    error = index_init(primary_key, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4690 4691 4692
    if (error) { goto cleanup;}

    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4693
    // only want to set range_lock_grabbed to true after index_init
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4694 4695 4696 4697
    // 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
4698
    if (scan) { range_lock_grabbed = true; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4699
    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4700 4701
cleanup:
    TOKUDB_DBUG_RETURN(error);
4702 4703
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4704 4705 4706
//
// End a scan of the table
//
4707
int ha_tokudb::rnd_end() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4708
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_end");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4709
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4710
    TOKUDB_DBUG_RETURN(index_end());
4711 4712
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4713

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4714 4715 4716 4717 4718 4719 4720 4721 4722
//
// 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
//
4723
int ha_tokudb::rnd_next(uchar * buf) {
4724
    TOKUDB_DBUG_ENTER("ha_tokudb::ha_tokudb::rnd_next");
4725
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4726
    u_int32_t flags = SET_READ_FLAG(0);
4727 4728 4729
    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
4730
    struct smart_dbt_info info;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4731

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4732
    HANDLE_INVALID_CURSOR();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4733 4734 4735 4736
    //
    // The reason we do not just call index_next is that index_next 
    // increments a different variable than we do here
    //
4737
    statistic_increment(table->in_use->status_var.ha_read_rnd_next_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4738 4739 4740 4741

    info.ha = this;
    info.buf = buf;
    info.keynr = primary_key;
4742

4743
    lockretryN(read_lock_wait_time){
4744 4745 4746 4747
        error = cursor->c_getf_next(cursor, flags, SMART_DBT_CALLBACK, &info);
        lockretry_wait;
    }
    error = handle_cursor_error(error, HA_ERR_END_OF_FILE,primary_key);
4748 4749 4750

    trx->stmt_progress.queried++;
    track_progress(thd);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4751
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4752
    TOKUDB_DBUG_RETURN(error);
4753 4754 4755
}


4756 4757 4758
void ha_tokudb::track_progress(THD* thd) {
    tokudb_trx_data* trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
    if (trx) {
4759
        ulonglong num_written = trx->stmt_progress.inserted + trx->stmt_progress.updated + trx->stmt_progress.deleted;
4760
        bool update_status = 
4761
            (trx->stmt_progress.queried && tokudb_read_status_frequency && (trx->stmt_progress.queried % tokudb_read_status_frequency) == 0) ||
4762
            (num_written && tokudb_write_status_frequency && (num_written % tokudb_write_status_frequency) == 0);
4763
        if (update_status) {
4764 4765 4766 4767
            char *next_status = write_status_msg;
            bool first = true;
            int r;
            if (trx->stmt_progress.queried) {
4768 4769
                r = sprintf(next_status, "Queried about %llu row%s", trx->stmt_progress.queried, trx->stmt_progress.queried == 1 ? "" : "s"); 
                assert(r >= 0);
4770 4771 4772 4773
                next_status += r;
                first = false;
            }
            if (trx->stmt_progress.inserted) {
4774 4775
                r = sprintf(next_status, "%sInserted about %llu row%s", first ? "" : ", ", trx->stmt_progress.inserted, trx->stmt_progress.inserted == 1 ? "" : "s"); 
                assert(r >= 0);
4776 4777 4778 4779
                next_status += r;
                first = false;
            }
            if (trx->stmt_progress.updated) {
4780 4781
                r = sprintf(next_status, "%sUpdated about %llu row%s", first ? "" : ", ", trx->stmt_progress.updated, trx->stmt_progress.updated == 1 ? "" : "s"); 
                assert(r >= 0);
4782 4783 4784 4785
                next_status += r;
                first = false;
            }
            if (trx->stmt_progress.deleted) {
4786 4787
                r = sprintf(next_status, "%sDeleted about %llu row%s", first ? "" : ", ", trx->stmt_progress.deleted, trx->stmt_progress.deleted == 1 ? "" : "s"); 
                assert(r >= 0);
4788 4789 4790
                next_status += r;
                first = false;
            }
4791 4792
            if (!first)
	        thd_proc_info(thd, write_status_msg);
4793 4794 4795 4796 4797
        }
    }
}


4798
DBT *ha_tokudb::get_pos(DBT * to, uchar * pos) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4799
    TOKUDB_DBUG_ENTER("ha_tokudb::get_pos");
4800 4801
    /* We don't need to set app_data here */
    bzero((void *) to, sizeof(*to));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4802 4803
    to->data = pos + sizeof(u_int32_t);
    to->size = *(u_int32_t *)pos;
4804
    DBUG_DUMP("key", (const uchar *) to->data, to->size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4805
    DBUG_RETURN(to);
4806 4807
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4808
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4809
// Retrieves a row with based on the primary key saved in pos
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4810 4811 4812 4813 4814
// Returns:
//      0 on success
//      HA_ERR_KEY_NOT_FOUND if not found
//      error otherwise
//
4815
int ha_tokudb::rnd_pos(uchar * buf, uchar * pos) {
4816
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_pos");
4817
    DBT db_pos;
4818
    int error = 0;
4819 4820 4821
    struct smart_dbt_info info;
    bool old_unpack_entire_row = unpack_entire_row;
    DBT* key = get_pos(&db_pos, pos); 
4822
    read_lock_wait_time = get_read_lock_wait_time(ha_thd());
4823 4824

    unpack_entire_row = true;
4825 4826
    statistic_increment(table->in_use->status_var.ha_read_rnd_count, &LOCK_status);
    active_index = MAX_KEY;
4827 4828 4829 4830 4831

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

4832
    lockretryN(read_lock_wait_time) {
4833 4834 4835 4836
        error = share->file->getf_set(share->file, transaction, 0, key, smart_dbt_callback_rowread_ptquery, &info);
        lockretry_wait;
    }

Barry Perlman's avatar
Barry Perlman committed
4837
    if (error == DB_NOTFOUND) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4838 4839 4840 4841
        error = HA_ERR_KEY_NOT_FOUND;
        goto cleanup;
    }
cleanup:
4842
    unpack_entire_row = old_unpack_entire_row;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4843
    TOKUDB_DBUG_RETURN(error);
4844 4845
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4846

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4847
int ha_tokudb::prelock_range( const key_range *start_key, const key_range *end_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4848
    TOKUDB_DBUG_ENTER("ha_tokudb::read_range_first");
4849 4850

    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4851 4852 4853 4854
    DBT start_dbt_key;
    const DBT* start_dbt_data = NULL;
    DBT end_dbt_key;
    const DBT* end_dbt_data = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4855 4856
    uchar* start_key_buff  = key_buff2;
    uchar* end_key_buff = key_buff3;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4857

Zardosht Kasheff's avatar
Zardosht Kasheff committed
4858 4859 4860 4861 4862 4863
    bzero((void *) &start_dbt_key, sizeof(start_dbt_key));
    bzero((void *) &end_dbt_key, sizeof(end_dbt_key));

    if (start_key) {
        switch (start_key->flag) {
        case HA_READ_AFTER_KEY:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4864
            pack_key(&start_dbt_key, active_index, start_key_buff, start_key->key, start_key->length, COL_POS_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4865 4866 4867
            start_dbt_data = share->key_file[active_index]->dbt_pos_infty();
            break;
        default:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4868
            pack_key(&start_dbt_key, active_index, start_key_buff, start_key->key, start_key->length, COL_NEG_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879
            start_dbt_data = share->key_file[active_index]->dbt_neg_infty();
            break;
        }
    }
    else {
        start_dbt_data = share->key_file[active_index]->dbt_neg_infty();
    }

    if (end_key) {
        switch (end_key->flag) {
        case HA_READ_BEFORE_KEY:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4880
            pack_key(&end_dbt_key, active_index, end_key_buff, end_key->key, end_key->length, COL_NEG_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4881 4882 4883
            end_dbt_data = share->key_file[active_index]->dbt_neg_infty();
            break;
        default:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4884
            pack_key(&end_dbt_key, active_index, end_key_buff, end_key->key, end_key->length, COL_POS_INF);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4885 4886 4887 4888 4889 4890 4891 4892 4893
            end_dbt_data = share->key_file[active_index]->dbt_pos_infty();
            break;
        }
        
    }
    else {
        end_dbt_data = share->key_file[active_index]->dbt_pos_infty();
    }

4894
    lockretryN(read_lock_wait_time){
4895 4896 4897 4898 4899 4900 4901 4902 4903 4904
        error = share->key_file[active_index]->pre_acquire_read_lock(
            share->key_file[active_index], 
            transaction, 
            start_key ? &start_dbt_key : share->key_file[active_index]->dbt_neg_infty(), 
            start_dbt_data, 
            end_key ? &end_dbt_key : share->key_file[active_index]->dbt_pos_infty(), 
            end_dbt_data
            );
        lockretry_wait;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4905 4906 4907 4908 4909 4910
    if (error){ 
        last_cursor_error = error;
        //
        // cursor should be initialized here, but in case it is not, we still check
        //
        if (cursor) {
4911 4912
            int r = cursor->c_close(cursor);
            assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4913 4914 4915 4916
            cursor = NULL;
        }
        goto cleanup; 
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4917 4918 4919

cleanup:
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4920
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4921

4922 4923 4924 4925 4926 4927
//
// 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
4928
int ha_tokudb::prepare_range_scan( const key_range *start_key, const key_range *end_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4929 4930 4931 4932 4933
    int error = prelock_range(start_key, end_key);
    if (!error) {
        range_lock_grabbed = true;
    }
    return error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950
}

int ha_tokudb::read_range_first(
    const key_range *start_key,
    const key_range *end_key,
    bool eq_range, 
    bool sorted) 
{
    int error;
    error = prelock_range(start_key, end_key);
    if (error) { goto cleanup; }
    range_lock_grabbed = true;
    
    error = handler::read_range_first(start_key, end_key, eq_range, sorted);
cleanup:
    return error;
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4951 4952
int ha_tokudb::read_range_next()
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4953 4954 4955 4956
    TOKUDB_DBUG_ENTER("ha_tokudb::read_range_next");
    int error;
    error = handler::read_range_next();
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4957
        range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4958 4959
    }
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4960 4961 4962 4963
}



4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984
/*
  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
*/
4985
void ha_tokudb::position(const uchar * record) {
4986
    TOKUDB_DBUG_ENTER("ha_tokudb::position");
4987 4988
    DBT key;
    if (hidden_primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4989 4990
        DBUG_ASSERT(ref_length == (TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH + sizeof(u_int32_t)));
        memcpy_fixed(ref + sizeof(u_int32_t), current_ident, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4991
        *(u_int32_t *)ref = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4992 4993
    } 
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
4994 4995 4996 4997 4998 4999 5000 5001 5002
        bool has_null;
        //
        // save the data
        //
        create_dbt_key_from_table(&key, primary_key, ref + sizeof(u_int32_t), record, &has_null);
        //
        // save the size of data in the first four bytes of ref
        //
        memcpy(ref, &key.size, sizeof(u_int32_t));
5003 5004 5005 5006
    }
    DBUG_VOID_RETURN;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5007 5008 5009 5010 5011 5012
//
// Per InnoDB: Returns statistics information of the table to the MySQL interpreter,
// in various fields of the handle object. 
// Return:
//      0, always success
//
5013
int ha_tokudb::info(uint flag) {
5014
    TOKUDB_DBUG_ENTER("ha_tokudb::info %p %d %lld", this, flag, (long long) share->rows);
5015 5016 5017 5018
    int error;
    DB_TXN* txn = NULL;
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
    DB_BTREE_STAT64 dict_stats;
5019 5020
    if (flag & HA_STATUS_VARIABLE) {
        // Just to get optimizations right
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5021
        stats.records = share->rows + share->rows_from_locked_table;
5022
        stats.deleted = 0;
5023
        if (!(flag & HA_STATUS_NO_LOCK)) {
5024
            u_int64_t num_rows = 0;
5025
            TOKU_DB_FRAGMENTATION_S frag_info = {0};
5026

5027 5028 5029
            error = db_env->txn_begin(db_env, NULL, &txn, DB_READ_UNCOMMITTED);
            if (error) { goto cleanup; }

5030 5031 5032 5033 5034 5035 5036 5037
            error = estimate_num_rows(share->file,&num_rows, txn);
            if (error == 0) {
                share->rows = num_rows;
                stats.records = num_rows;
            }
            else {
                goto cleanup;
            }
5038 5039 5040 5041 5042 5043
            error = share->file->get_fragmentation(
                share->file,
                &frag_info
                );
            if (error) { goto cleanup; }
            stats.delete_length = frag_info.unused_bytes;
5044

5045 5046 5047 5048 5049 5050
            error = share->file->stat64(
                share->file, 
                txn, 
                &dict_stats
                );
            if (error) { goto cleanup; }
5051
            
5052
            stats.data_file_length = dict_stats.bt_dsize;
5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068
            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
                //
                u_int64_t hpk_space = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH*dict_stats.bt_ndata;
                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
                //
                u_int64_t inf_byte_space = dict_stats.bt_ndata;
                stats.data_file_length = (inf_byte_space > stats.data_file_length) ? 0 : stats.data_file_length - inf_byte_space;
            }

5069
            stats.mean_rec_length = stats.records ? (ulong)(stats.data_file_length/stats.records) : 0;
5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081
            stats.index_file_length = 0;
            for (uint i = 0; i < curr_num_DBs; i++) {
                if (i == primary_key) {
                    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;
5082 5083 5084 5085 5086 5087 5088

                error = share->file->get_fragmentation(
                    share->file,
                    &frag_info
                    );
                if (error) { goto cleanup; }
                stats.delete_length += frag_info.unused_bytes;
5089 5090
            }
        }
5091
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5092
    if ((flag & HA_STATUS_CONST)) {
5093
        stats.max_data_file_length=  9223372036854775807ULL;
5094
        for (uint i = 0; i < table_share->keys; i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5095
            table->key_info[i].rec_per_key[table->key_info[i].key_parts - 1] = 0;
5096 5097 5098
        }
    }
    /* Don't return key if we got an error for the internal primary key */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5099
    if (flag & HA_STATUS_ERRKEY && last_dup_key < table_share->keys) {
5100
        errkey = last_dup_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5101
    }    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5102
    if (flag & HA_STATUS_AUTO && table->found_next_number_field) {        
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5103 5104
        THD *thd= table->in_use;
        struct system_variables *variables= &thd->variables;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5105 5106
        stats.auto_increment_value = share->last_auto_increment + variables->auto_increment_increment;
    }
5107 5108 5109
    error = 0;
cleanup:
    if (txn != NULL) {
5110
        commit_txn(txn, DB_TXN_NOSYNC);
5111 5112 5113
        txn = NULL;
    }
    TOKUDB_DBUG_RETURN(error);
5114 5115
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5116 5117 5118
//
//  Per InnoDB: Tells something additional to the handler about how to do things.
//
5119
int ha_tokudb::extra(enum ha_extra_function operation) {
5120
    TOKUDB_DBUG_ENTER("extra %p %d", this, operation);
5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139
    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;
    }
5140
    TOKUDB_DBUG_RETURN(0);
5141 5142 5143
}

int ha_tokudb::reset(void) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5144
    TOKUDB_DBUG_ENTER("ha_tokudb::reset");
5145 5146
    key_read = 0;
    using_ignore = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5147
    TOKUDB_DBUG_RETURN(0);
5148 5149
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172

//
// 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;
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
    if (lt == lock_read) {
        for (uint i = 0; i < curr_num_DBs; i++) {
            DB* db = share->key_file[i];
            error = db->pre_acquire_read_lock(
                db, 
                trans, 
                db->dbt_neg_infty(), db->dbt_neg_infty(), 
                db->dbt_pos_infty(), db->dbt_pos_infty()
                );
5173
            if (error) break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5174
        }
5175
        if (error) goto cleanup;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5176 5177
    }
    else if (lt == lock_write) {
5178 5179
        if (tokudb_debug & TOKUDB_DEBUG_LOCK)
            TOKUDB_TRACE("%s\n", __FUNCTION__);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5180 5181 5182
        for (uint i = 0; i < curr_num_DBs; i++) {
            DB* db = share->key_file[i];
            error = db->pre_acquire_table_lock(db, trans);
5183 5184 5185
            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
5186
        }
5187 5188 5189
        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
5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201
    }
    else {
        error = ENOSYS;
        goto cleanup;
    }

    error = 0;
cleanup:
    return error;
}


5202 5203 5204 5205 5206 5207
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);

    /* First table lock, start transaction */
5208
    if (thd_test_options(thd, OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN) && 
5209 5210 5211 5212 5213
         !trx->all &&
         (thd_sql_command(thd) != SQLCOM_CREATE_TABLE) &&
         (thd_sql_command(thd) != SQLCOM_DROP_TABLE) &&
         (thd_sql_command(thd) != SQLCOM_ALTER_TABLE)) {
        /* QQQ We have to start a master transaction */
5214
        // DBUG_PRINT("trans", ("starting transaction all "));
5215 5216 5217 5218 5219
        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) {
5220
            TOKUDB_TRACE("just created master:%p\n", trx->all);
5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231
        }
        trx->sp_level = trx->all;
        trans_register_ha(thd, TRUE, tokudb_hton);
    }
    DBUG_PRINT("trans", ("starting transaction stmt"));
    if (trx->stmt) { 
        if (tokudb_debug & TOKUDB_DEBUG_TXN) {
            TOKUDB_TRACE("warning:stmt=%p\n", trx->stmt);
        }
    }
    u_int32_t txn_begin_flags;
5232
    if (trx->all == NULL) {
5233 5234 5235
        txn_begin_flags = toku_iso_to_txn_flag(toku_iso_level);
    }
    else {
5236
        txn_begin_flags = DB_INHERIT_ISOLATION;
5237 5238 5239 5240 5241 5242
    }
    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;
    }
5243
    trx->sub_sp_level = trx->stmt;
5244
    if (tokudb_debug & TOKUDB_DEBUG_TXN) {
5245
        TOKUDB_TRACE("just created stmt:%p:%p\n", trx->sp_level, trx->stmt);
5246
    }
5247
    reset_stmt_progress(&trx->stmt_progress);
5248 5249 5250 5251 5252
    trans_register_ha(thd, FALSE, tokudb_hton);
cleanup:
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5253

5254 5255 5256 5257 5258 5259 5260 5261
/*
  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
5262 5263 5264 5265 5266 5267 5268 5269
//
// Parameters:
//      [in]    thd - handle to the user thread
//              lock_type - the type of lock
// Returns:
//      0 on success
//      error otherwise
//
5270
int ha_tokudb::external_lock(THD * thd, int lock_type) {
5271 5272 5273
    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);
5274
    // QQQ this is here to allow experiments without transactions
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5275 5276
    int error = 0;
    tokudb_trx_data *trx = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5277 5278


Zardosht Kasheff's avatar
Zardosht Kasheff committed
5279
    trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
5280
    if (!trx) {
5281 5282
        error = create_tokudb_trx_data_instance(&trx);
        if (error) { goto cleanup; }
5283
        thd_data_set(thd, tokudb_hton->slot, trx);
5284
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5285 5286
    if (trx->all == NULL) {
        trx->sp_level = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5287
    }
5288 5289 5290
    if (lock_type != F_UNLCK) {
        if (!trx->tokudb_lock_count++) {
            DBUG_ASSERT(trx->stmt == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5291
            transaction = NULL;    // Safety
5292 5293
            error = create_txn(thd, trx);
            if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5294
                goto cleanup;
5295 5296
            }
        }
5297 5298
        assert(thd->in_sub_stmt == 0);
        transaction = trx->sub_sp_level;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5299 5300
    }
    else {
5301
        lock.type = TL_UNLOCK;  // Unlocked
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313

        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
5314
        share->rows_from_locked_table = 0;
5315 5316 5317 5318 5319 5320 5321 5322
        if (!--trx->tokudb_lock_count) {
            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"));
5323
                commit_txn(trx->stmt, 0);
5324
                reset_stmt_progress(&trx->stmt_progress);
Rich Prohaska's avatar
Rich Prohaska committed
5325
                if (tokudb_debug & TOKUDB_DEBUG_TXN)
5326
                    TOKUDB_TRACE("commit:%p:%d\n", trx->stmt, error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5327
                trx->stmt = NULL;
5328
                trx->sub_sp_level = NULL;
5329 5330
            }
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5331
        transaction = NULL;
5332
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5333
cleanup:
5334 5335
    if (tokudb_debug & TOKUDB_DEBUG_LOCK)
        TOKUDB_TRACE("%s error=%d\n", __FUNCTION__, error);
5336
    TOKUDB_DBUG_RETURN(error);
5337 5338 5339 5340 5341 5342 5343 5344 5345
}


/*
  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.
*/

5346
int ha_tokudb::start_stmt(THD * thd, thr_lock_type lock_type) {
5347
    TOKUDB_DBUG_ENTER("ha_tokudb::start_stmt cmd=%d %d", thd_sql_command(thd), lock_type);
5348
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5349 5350


5351
    tokudb_trx_data *trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
5352 5353 5354 5355 5356 5357 5358
    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) {
5359 5360 5361 5362
        error = create_txn(thd, trx);
        if (error) {
            goto cleanup;
        }
5363
    }
5364 5365 5366 5367 5368
    else {
        if (tokudb_debug & TOKUDB_DEBUG_TXN) {
            TOKUDB_TRACE("trx->stmt already existed\n");
        }
    }
5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379
    //
    // 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) {
5380
        acquire_table_lock(trx->sub_sp_level,lock_read);
5381 5382
    }
    else {
5383
        acquire_table_lock(trx->sub_sp_level,lock_write);
5384
    }    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5385 5386 5387
    if (added_rows > deleted_rows) {
        share->rows_from_locked_table = added_rows - deleted_rows;
    }
5388 5389
    transaction = trx->sub_sp_level;
    trans_register_ha(thd, FALSE, tokudb_hton);
5390
cleanup:
5391
    TOKUDB_DBUG_RETURN(error);
5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421
}

/*
  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.
*/

5422
THR_LOCK_DATA **ha_tokudb::store_lock(THD * thd, THR_LOCK_DATA ** to, enum thr_lock_type lock_type) {
5423
    TOKUDB_DBUG_ENTER("ha_tokudb::store_lock, lock_type=%d cmd=%d", lock_type, thd_sql_command(thd));
5424 5425
    if (tokudb_debug & TOKUDB_DEBUG_LOCK)
        TOKUDB_TRACE("%s lock_type=%d cmd=%d\n", __FUNCTION__, lock_type, thd_sql_command(thd));
5426 5427
    if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK) {
        /* If we are not doing a LOCK TABLE, then allow multiple writers */
5428
        if ((lock_type >= TL_WRITE_CONCURRENT_INSERT && lock_type <= TL_WRITE) && 
5429
            !thd->in_lock_tables && thd_sql_command(thd) != SQLCOM_TRUNCATE && !thd_tablespace_op(thd)) {
5430
            lock_type = TL_WRITE_ALLOW_WRITE;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5431
        }
5432
        lock.type = lock_type;
5433
    }
5434
    *to++ = &lock;
5435 5436
    if (tokudb_debug & TOKUDB_DEBUG_LOCK)
        TOKUDB_TRACE("%s lock_type=%d\n", __FUNCTION__, lock_type);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5437
    DBUG_RETURN(to);
5438 5439
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5440 5441 5442 5443 5444 5445 5446
int toku_dbt_up(DB*,
                                 u_int32_t old_version, const DBT *old_descriptor, const DBT *old_key, const DBT *old_val,
                                 u_int32_t new_version, const DBT *new_descriptor, const DBT *new_key, const DBT *new_val) {
    assert(false);
    return 0;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5447

5448
static int create_sub_table(const char *table_name, DBT* row_descriptor, DB_TXN* txn) {
5449
    TOKUDB_DBUG_ENTER("create_sub_table");
5450
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5451
    DB *file = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5452
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5453 5454 5455 5456 5457 5458 5459 5460
    
    error = db_create(&file, db_env, 0);
    if (error) {
        DBUG_PRINT("error", ("Got error: %d when creating table", error));
        my_errno = error;
        goto exit;
    }
        
5461
    error = file->set_descriptor(file, 1, row_descriptor);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5462 5463 5464 5465 5466
    if (error) {
        DBUG_PRINT("error", ("Got error: %d when setting row descriptor for table '%s'", error, table_name));
        goto exit;
    }
    
5467
    error = file->open(file, txn, table_name, NULL, DB_BTREE, DB_THREAD | DB_CREATE | DB_EXCL, my_umask);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5468 5469 5470 5471
    if (error) {
        DBUG_PRINT("error", ("Got error: %d when opening table '%s'", error, table_name));
        goto exit;
    } 
5472

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5473 5474
    error = 0;
exit:
5475
    if (file) {
5476 5477
        int r = file->close(file, 0);
        assert(r==0);
5478
    }
5479
    TOKUDB_DBUG_RETURN(error);
5480 5481
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5482 5483 5484 5485 5486 5487 5488
void ha_tokudb::update_create_info(HA_CREATE_INFO* create_info) {
    if (share->has_auto_inc) {
        info(HA_STATUS_AUTO);
        create_info->auto_increment_value = stats.auto_increment_value;
    }
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5489
//
5490 5491 5492 5493
// 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
5494
//
5495
int ha_tokudb::remove_key_name_from_status(DB* status_block, char* key_name, DB_TXN* txn) {
5496
    int error;
5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515
    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
5516

5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543
//
// 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;
}
5544

5545 5546 5547 5548 5549
//
// 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
5550 5551 5552
    //
    // tracing information about what type of table we are creating
    //
5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569
    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);
            }
        }
    }
5570 5571 5572 5573 5574
}

//
// creates dictionary for secondary index, with key description key_info, all using txn
//
5575
int ha_tokudb::create_secondary_dictionary(const char* name, TABLE* form, KEY* key_info, DB_TXN* txn, KEY_AND_COL_INFO* kc_info, u_int32_t keynr) {
5576 5577 5578
    int error;
    DBT row_descriptor;
    uchar* row_desc_buff = NULL;
5579
    uchar* ptr = NULL;
5580 5581 5582
    char* newname = NULL;
    KEY* prim_key = NULL;
    char dict_name[MAX_DICT_NAME_LEN];
5583
    u_int32_t max_row_desc_buff_size;
5584

5585
    uint hpk= (form->s->primary_key >= MAX_KEY) ? TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH : 0;
5586

5587
    bzero(&row_descriptor, sizeof(row_descriptor));
5588 5589 5590 5591 5592 5593 5594
    
    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


    row_desc_buff = (uchar *)my_malloc(max_row_desc_buff_size, MYF(MY_WME));
5595
    if (row_desc_buff == NULL){ error = ENOMEM; goto cleanup;}
5596
    ptr = row_desc_buff;
5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609

    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;
5610 5611 5612 5613
    //
    // save data necessary for key comparisons
    //
    ptr += create_toku_key_descriptor(
5614 5615 5616 5617 5618 5619
        row_desc_buff,
        false,
        key_info,
        hpk,
        prim_key
        );
5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643

    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
        );

    row_descriptor.size = ptr - row_desc_buff;
    assert(row_descriptor.size <= max_row_desc_buff_size);

5644
    error = create_sub_table(newname, &row_descriptor, txn);
5645 5646 5647 5648 5649 5650 5651 5652 5653 5654
cleanup:    
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    my_free(row_desc_buff, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}

//
// create and close the main dictionarr with name of "name" using table form, all within
// transaction txn.
//
5655
int ha_tokudb::create_main_dictionary(const char* name, TABLE* form, DB_TXN* txn, KEY_AND_COL_INFO* kc_info) {
5656 5657 5658
    int error;
    DBT row_descriptor;
    uchar* row_desc_buff = NULL;
5659
    uchar* ptr = NULL;
5660 5661
    char* newname = NULL;
    KEY* prim_key = NULL;
5662
    u_int32_t max_row_desc_buff_size;
5663 5664 5665 5666

    uint hpk= (form->s->primary_key >= MAX_KEY) ? TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH : 0;

    bzero(&row_descriptor, sizeof(row_descriptor));
5667 5668 5669 5670 5671
    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

    row_desc_buff = (uchar *)my_malloc(max_row_desc_buff_size, MYF(MY_WME));
5672
    if (row_desc_buff == NULL){ error = ENOMEM; goto cleanup;}
5673
    ptr = row_desc_buff;
5674 5675 5676

    newname = (char *)my_malloc(get_max_dict_name_path_length(name),MYF(MY_WME));
    if (newname == NULL){ error = ENOMEM; goto cleanup;}
5677

5678
    make_name(newname, name, "main");
5679 5680

    prim_key = (hpk) ? NULL : &form->s->key_info[primary_key];
5681

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5682 5683 5684 5685
    //
    // setup the row descriptor
    //
    row_descriptor.data = row_desc_buff;
5686 5687 5688 5689
    //
    // save data necessary for key comparisons
    //
    ptr += create_toku_key_descriptor(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5690
        row_desc_buff, 
5691
        hpk,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5692 5693 5694 5695
        prim_key,
        false,
        NULL
        );
5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712
    
    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
        );


    row_descriptor.size = ptr - row_desc_buff;
    assert(row_descriptor.size <= max_row_desc_buff_size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5713

5714
    /* Create the main table that will hold the real rows */
5715
    error = create_sub_table(newname, &row_descriptor, txn);
5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739
cleanup:    
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
    my_free(row_desc_buff, MYF(MY_ALLOW_ZERO_PTR));
    return error;
}

//
// 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;
    char* newname = NULL;
5740 5741
    KEY_AND_COL_INFO kc_info;
    bzero(&kc_info, sizeof(kc_info));
5742

5743
    pthread_mutex_lock(&tokudb_meta_mutex);
5744

5745 5746
    newname = (char *)my_malloc(get_max_dict_name_path_length(name),MYF(MY_WME));
    if (newname == NULL){ error = ENOMEM; goto cleanup;}
5747

5748
    error = db_env->txn_begin(db_env, 0, &txn, 0);
5749
    if (error) { goto cleanup; }
5750

5751 5752
    primary_key = form->s->primary_key;
    hidden_primary_key = (primary_key  >= MAX_KEY) ? TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH : 0;
5753 5754 5755
    if (hidden_primary_key) {
        primary_key = form->s->keys;
    }
5756 5757 5758 5759

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

Zardosht Kasheff's avatar
Zardosht Kasheff committed
5760
    /* Create status.tokudb and save relevant metadata */
5761
    make_name(newname, name, "status");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5762

5763 5764 5765
    error = db_create(&status_block, db_env, 0);
    if (error) { goto cleanup; }

5766
    error = status_block->open(status_block, txn, newname, NULL, DB_BTREE, DB_CREATE | DB_EXCL, 0);
5767
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5768

5769 5770 5771
    version = HA_TOKU_VERSION;
    capabilities = HA_TOKU_CAP;
    
5772
    error = write_to_status(status_block, hatoku_version,&version,sizeof(version), txn);
5773
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5774

5775
    error = write_to_status(status_block, hatoku_capabilities,&capabilities,sizeof(capabilities), txn);
5776 5777
    if (error) { goto cleanup; }

5778
    error = write_auto_inc_create(status_block, create_info->auto_increment_value, txn);
5779
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5780

5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791
    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; }
5792

5793
    error = create_main_dictionary(name, form, txn, &kc_info);
5794 5795 5796 5797
    if (error) {
        goto cleanup;
    }

5798 5799 5800

    for (uint i = 0; i < form->s->keys; i++) {
        if (i != primary_key) {
5801
            error = create_secondary_dictionary(name, form, &form->key_info[i], txn, &kc_info, i);
5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813
            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; }

5814
    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5815 5816
cleanup:
    if (status_block != NULL) {
5817 5818
        int r = status_block->close(status_block, 0);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5819
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5820
    free_key_and_col_info(&kc_info);
5821 5822 5823 5824 5825 5826 5827
    if (txn) {
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn,0);
        }
5828
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5829
    my_free(newname, MYF(MY_ALLOW_ZERO_PTR));
5830
    pthread_mutex_unlock(&tokudb_meta_mutex);
5831
    TOKUDB_DBUG_RETURN(error);
5832 5833
}

5834 5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846
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
5847
//
5848 5849 5850 5851
// 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
5852
//
5853
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) {
5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903
    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; }
5904

5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916
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) {
5917
    int error;
5918
    DB* status_db = NULL;
5919 5920 5921 5922 5923 5924
    DBC* status_cursor = NULL;
    DB_TXN* txn = NULL;
    DBT curr_key;
    DBT curr_val;
    bzero(&curr_key, sizeof(curr_key));
    bzero(&curr_val, sizeof(curr_val));
5925
    pthread_mutex_lock(&tokudb_meta_mutex);
5926

5927 5928 5929
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }

5930
    //
5931
    // modify metadata db
5932
    //
5933 5934
    if (is_delete) {
        error = drop_table_from_metadata(from_name, txn);
5935
    }
5936 5937
    else {
        error = rename_table_in_metadata(from_name, to_name, txn);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5938
    }
5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971 5972 5973 5974 5975 5976
    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);
5977
    assert(error==0);
5978
    status_cursor = NULL;
5979
    if (error) { goto cleanup; }
5980 5981

    error = status_db->close(status_db, 0);
5982
    assert(error == 0);
5983 5984 5985 5986 5987 5988 5989 5990
    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; }

5991
    my_errno = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
5992
cleanup:
5993
    if (status_cursor) {
5994 5995
        int r = status_cursor->c_close(status_cursor);
        assert(r==0);
5996 5997
    }
    if (status_db) {
5998 5999
        int r = status_db->close(status_db, 0);
        assert(r==0);
6000 6001 6002 6003 6004 6005 6006 6007
    }
    if (txn) {
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn, 0);
        }
6008
    }
6009
    pthread_mutex_unlock(&tokudb_meta_mutex);
6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023
    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");
6024 6025
    int error;
    error = delete_or_rename_table(name, NULL, true);
6026
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6027 6028 6029 6030 6031
        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);
6032 6033 6034
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
6035 6036 6037 6038 6039 6040 6041 6042 6043
//
// 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
//
6044
int ha_tokudb::rename_table(const char *from, const char *to) {
6045
    TOKUDB_DBUG_ENTER("%s %s %s", __FUNCTION__, from, to);
6046
    int error;
6047
    error = delete_or_rename_table(from, to, false);
6048
    if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6049 6050 6051 6052
        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);
    }
6053
    TOKUDB_DBUG_RETURN(error);
6054 6055 6056 6057
}


/*
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6058
  Returns estimate on number of seeks it will take to read through the table
6059 6060 6061
  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.
*/
6062
/// QQQ why divide by 3
6063
double ha_tokudb::scan_time() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6064
    TOKUDB_DBUG_ENTER("ha_tokudb::scan_time");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6065
    double ret_val = (double)stats.records / 3;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6066
    DBUG_RETURN(ret_val);
6067 6068
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080
//
// 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(
6081 6082 6083 6084
    uint    index,
    uint    ranges,
    ha_rows rows
    )
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6085 6086 6087 6088
{
    double total_scan;
    double ret_val; 

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6089 6090 6091 6092 6093 6094 6095 6096 6097
    //
    // 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
6098 6099 6100 6101
    //
    // if it is not the primary key, and it is not a clustering key, then return handler::read_time
    //
    if (index != primary_key && !(table->key_info[index].flags & HA_CLUSTERING)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6102 6103 6104 6105
        ret_val = handler::read_time(index, ranges, rows);
        goto cleanup;
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6106 6107 6108
    //
    // for primary key and for clustered keys, return a fraction of scan_time()
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124
    total_scan = scan_time();

    if (stats.records < rows) {
        ret_val = total_scan;
        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);
cleanup:
    return ret_val;
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
6125
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136
// 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
6137
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6138
ha_rows ha_tokudb::records_in_range(uint keynr, key_range* start_key, key_range* end_key) {
6139
    TOKUDB_DBUG_ENTER("ha_tokudb::records_in_range");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6140
    DBT key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6141
    ha_rows ret_val = HA_TOKUDB_RANGE_COUNT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6142
    DB *kfile = share->key_file[keynr];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6143 6144 6145
    u_int64_t less, equal, greater;
    u_int64_t start_rows, end_rows, rows;
    int is_exact;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6146
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6147
    uchar inf_byte;
6148

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6149
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6150
    // get start_rows and end_rows values so that we can estimate range
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6151 6152 6153 6154
    // 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
6155 6156
    //
    if (start_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6157 6158 6159 6160 6161 6162 6163 6164 6165 6166
        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
6167
        error = kfile->key_range64(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6168 6169
            kfile, 
            transaction, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6170
            &key,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6171 6172 6173 6174
            &less,
            &equal,
            &greater,
            &is_exact
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6175 6176 6177 6178 6179
            );
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6180
        start_rows= less;
6181
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6182
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6183
        start_rows= 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6184 6185 6186
    }

    if (end_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6187 6188 6189 6190 6191 6192 6193 6194 6195 6196
        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
6197
        error = kfile->key_range64(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6198 6199
            kfile, 
            transaction, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6200
            &key,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6201 6202 6203 6204
            &less,
            &equal,
            &greater,
            &is_exact
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6205 6206 6207 6208 6209
            );
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6210
        end_rows= less;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6211 6212
    }
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6213
        end_rows = stats.records;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6214 6215
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6216
    rows = (end_rows > start_rows) ? end_rows - start_rows : 1;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6217

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6218
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6219 6220
    // 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
6221
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6222
    ret_val = (ha_rows) (rows <= 1 ? 1 : rows);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6223
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6224
    DBUG_RETURN(ret_val);
6225 6226
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6227 6228

//
6229 6230 6231
// 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
6232 6233 6234 6235 6236 6237 6238 6239 6240 6241
//
void ha_tokudb::init_auto_increment() {
    DBT key;
    DBT value;
    int error;
    HA_METADATA_KEY key_val = hatoku_max_ai;
    bzero(&key, sizeof(key));
    bzero(&value, sizeof(value));
    key.data = &key_val;
    key.size = sizeof(key_val);
6242
    value.flags = DB_DBT_USERMEM;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254
    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))
        //
6255 6256
        value.ulen = sizeof(share->last_auto_increment);
        value.data = &share->last_auto_increment;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6257 6258 6259 6260 6261 6262 6263 6264
        error = share->status_block->get(
            share->status_block, 
            txn, 
            &key, 
            &value, 
            0
            );
        
6265
        if (error || value.size != sizeof(share->last_auto_increment)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6266 6267
            share->last_auto_increment = 0;
        }
6268

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6269 6270 6271 6272 6273 6274
        //
        // 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;
6275 6276
        value.ulen = sizeof(share->auto_inc_create_value);
        value.data = &share->auto_inc_create_value;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6277 6278 6279 6280 6281 6282 6283 6284
        error = share->status_block->get(
            share->status_block, 
            txn, 
            &key, 
            &value, 
            0
            );
        
6285
        if (error || value.size != sizeof(share->auto_inc_create_value)) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6286 6287 6288
            share->auto_inc_create_value = 0;
        }

6289
        commit_txn(txn, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6290 6291 6292 6293 6294 6295
    }
    if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT) {
        TOKUDB_TRACE("init auto increment:%lld\n", share->last_auto_increment);
    }
}

6296
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
6297
    TOKUDB_DBUG_ENTER("ha_tokudb::get_auto_increment");
6298 6299
    ulonglong nr;

6300
    pthread_mutex_lock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6301 6302 6303 6304

    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;
6305
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6306 6307 6308 6309
    else {
        nr = share->last_auto_increment + increment;
    }
    share->last_auto_increment = nr + (nb_desired_values - 1)*increment;
6310 6311
    if (delay_updating_ai_metadata) {
        ai_metadata_update_required = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6312 6313 6314 6315
    }
    else {
        update_max_auto_inc(share->status_block, share->last_auto_increment);
    }
6316

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6317
    if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT) {
6318 6319
        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
6320
    }
6321
    *first_value = nr;
6322
    *nb_reserved_values = nb_desired_values;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6323
    pthread_mutex_unlock(&share->mutex);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6324
    DBUG_VOID_RETURN;
6325 6326
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6327 6328 6329 6330
bool ha_tokudb::is_auto_inc_singleton(){
    return false;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6331 6332 6333 6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348
//
// 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
//
int ha_tokudb::add_index(TABLE *table_arg, KEY *key_info, uint num_of_keys) {
    TOKUDB_DBUG_ENTER("ha_tokudb::add_index");
    int error;
    uint curr_index = 0;
    DBC* tmp_cursor = NULL;
    int cursor_ret_val = 0;
6349
    DBT curr_pk_key, curr_pk_val;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6350
    DB_TXN* txn = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6351
    THD* thd = ha_thd(); 
6352
    DB_LOADER* loader = NULL;
6353
    u_int32_t loader_flags = (get_load_save_space(thd)) ? LOADER_USE_PUTS : 0;
6354 6355 6356 6357 6358 6359
    u_int32_t mult_put_flags[MAX_KEY + 1] = {DB_YESOVERWRITE};
    u_int32_t mult_dbt_flags[MAX_KEY + 1] = {DB_DBT_REALLOC};
    struct loader_context lc = {0};
    lc.thd = thd;
    lc.ha = this;
    loader_error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6360 6361 6362 6363
    //
    // 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
6364

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6365 6366 6367 6368 6369
    //
    // 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
6370
    read_lock_wait_time = get_read_lock_wait_time(ha_thd());
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6371 6372
    thd_proc_info(thd, "Adding indexes");

6373 6374
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6375 6376 6377 6378 6379
    
    //
    // in unpack_row, MySQL passes a buffer that is this long,
    // so this length should be good enough for us as well
    //
6380 6381
    bzero((void *) &curr_pk_key, sizeof(curr_pk_key));
    bzero((void *) &curr_pk_val, sizeof(curr_pk_val));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397

    //
    // 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;
            }
        }
    }
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6398 6399 6400 6401
    //
    // open all the DB files and set the appropriate variables in share
    // they go to the end of share->key_file
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6402
    curr_index = curr_num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6403
    for (uint i = 0; i < num_of_keys; i++, curr_index++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6404 6405
        if (key_info[i].flags & HA_CLUSTERING) {
            set_key_filter(
6406
                &share->kc_info.key_filters[curr_index],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6407 6408 6409 6410 6411 6412
                &key_info[i],
                table_arg,
                false
                );                
            if (!hidden_primary_key) {
                set_key_filter(
6413
                    &share->kc_info.key_filters[curr_index],
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6414 6415 6416 6417 6418 6419
                    &table_arg->key_info[primary_key],
                    table_arg,
                    false
                    );
            }

6420
            error = initialize_col_pack_info(&share->kc_info,table_arg->s,curr_index);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6421 6422 6423 6424 6425 6426
            if (error) {
                goto cleanup;
            }
        }


6427 6428 6429
        error = create_secondary_dictionary(share->table_name, table_arg, &key_info[i], txn, &share->kc_info, curr_index);
        if (error) { goto cleanup; }

6430
        error = open_secondary_dictionary(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6431 6432 6433
            &share->key_file[curr_index], 
            &key_info[i],
            share->table_name,
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6434
            false,
6435
            txn
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6436 6437 6438 6439
            );
        if (error) { goto cleanup; }
    }
    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6440 6441 6442 6443 6444
    //
    // 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
    //
6445
    lockretryN(read_lock_wait_time){
6446
        error = share->file->pre_acquire_read_lock(
6447 6448 6449 6450 6451
            share->file,
            txn,
            share->file->dbt_neg_infty(),
            NULL,
            share->file->dbt_pos_infty(),
6452 6453 6454 6455
            NULL
            );
        lockretry_wait;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6456
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6457

6458 6459 6460 6461 6462 6463 6464 6465 6466
    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,
6467
        loader_flags
6468 6469 6470 6471 6472 6473 6474 6475
        );
    if (error) { goto cleanup; }

    error = loader->set_poll_function(loader, poll_fun, &lc);
    if (error) { goto cleanup; }

    error = loader->set_error_callback(loader, loader_ai_err_fun, &lc);
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6476

6477 6478 6479
    //
    // scan primary table, create each secondary key, add to each DB
    //    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6480 6481 6482 6483 6484
    if ((error = share->file->cursor(share->file, txn, &tmp_cursor, 0))) {
        tmp_cursor = NULL;             // Safety
        goto cleanup;
    }

6485
    cursor_ret_val = tmp_cursor->c_get(tmp_cursor, &curr_pk_key, &curr_pk_val, DB_NEXT | DB_PRELOCKED);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6486

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6487 6488 6489 6490 6491
    while (cursor_ret_val != DB_NOTFOUND) {
        if (cursor_ret_val) {
            error = cursor_ret_val;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6492

6493 6494
        error = loader->put(loader, &curr_pk_key, &curr_pk_val);
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6495

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6496 6497 6498
        num_processed++; 

        if ((num_processed % 1000) == 0) {
6499
            sprintf(status_msg, "Adding indexes: Processed %llu of about %llu rows.", num_processed, (long long unsigned) share->rows);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6500
            thd_proc_info(thd, status_msg);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6501 6502 6503 6504
            if (thd->killed) {
                error = ER_ABORTING_CONNECTION;
                goto cleanup;
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6505
        }
6506
        cursor_ret_val = tmp_cursor->c_get(tmp_cursor, &curr_pk_key, &curr_pk_val, DB_NEXT | DB_PRELOCKED);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6507
    }
6508 6509
    error = tmp_cursor->c_close(tmp_cursor);
    assert(error==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6510 6511
    tmp_cursor = NULL;

6512 6513 6514
    error = loader->close(loader);
    loader = NULL;
    if (error) goto cleanup;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6515

6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527 6528 6529
    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
6530 6531 6532 6533 6534
                goto cleanup;
            }
        }
    }

6535 6536 6537 6538 6539 6540
    //
    // We have an accurate row count, might as well update share->rows
    //
    pthread_mutex_lock(&share->mutex);
    share->rows = num_processed;
    pthread_mutex_unlock(&share->mutex);
6541 6542 6543 6544 6545 6546 6547 6548 6549

    //
    // 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
6550 6551 6552
    
    error = 0;
cleanup:
6553
    if (tmp_cursor) {            
6554 6555
        int r = tmp_cursor->c_close(tmp_cursor);
        assert(r==0);
6556 6557
        tmp_cursor = NULL;
    }
6558
    if (loader != NULL) {
6559 6560
        sprintf(status_msg, "aborting creation of indexes.");
        thd_proc_info(thd, status_msg);
6561 6562
        loader->abort(loader);
    }
6563 6564 6565 6566 6567
    if (txn) {
        if (error) {
            curr_index = curr_num_DBs;
            for (uint i = 0; i < num_of_keys; i++, curr_index++) {
                if (share->key_file[curr_index]) {
6568
                    int r = share->key_file[curr_index]->close(
6569 6570 6571
                        share->key_file[curr_index],
                        0
                        );
6572
                    assert(r==0);
6573 6574
                    share->key_file[curr_index] = NULL;
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6575
            }
6576 6577 6578 6579
            abort_txn(txn);
        }
        else {
            commit_txn(txn,0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6580 6581
        }
    }
6582
            if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6583 6584 6585 6586
                sql_print_error("Could not add indexes to table %s because \
another transaction has accessed the table. \
To add indexes, make sure no transactions touch the table.", share->table_name);
            }
6587
    TOKUDB_DBUG_RETURN(error ? error : loader_error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605 6606 6607 6608 6609
}

//
// Prepares to drop indexes to the table. For each value, i, in the array key_num,
// table->key_info[i] is a key that is to be dropped.
//  ***********NOTE*******************
// Although prepare_drop_index is supposed to just get the DB's ready for removal,
// and not actually do the removal, we are doing it here and not in final_drop_index
// For the flags we expose in alter_table_flags, namely xxx_NO_WRITES, this is allowed
// Changes for "future-proofing" this so that it works when we have the equivalent flags
// that are not NO_WRITES are not worth it at the moments
// Parameters:
//      [in]    table_arg - table that is being modified, seems to be identical to this->table
//      [in]    key_num - array of indexes that specify which keys of the array table->key_info
//                  are to be dropped
//              num_of_keys - size of array, key_num
//  Returns:
//      0 on success, error otherwise
//
int ha_tokudb::prepare_drop_index(TABLE *table_arg, uint *key_num, uint num_of_keys) {
    TOKUDB_DBUG_ENTER("ha_tokudb::prepare_drop_index");
    int error;
6610
    DB_TXN* txn = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6611

6612 6613
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6614 6615 6616
    
    for (uint i = 0; i < num_of_keys; i++) {
        uint curr_index = key_num[i];
6617 6618
        int r = share->key_file[curr_index]->close(share->key_file[curr_index],0);
        assert(r==0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6619 6620
        share->key_file[curr_index] = NULL;

6621 6622 6623 6624 6625
        error = remove_key_name_from_status(share->status_block, table_arg->key_info[curr_index].name, txn);
        if (error) { goto cleanup; }
        
        error = delete_or_rename_dictionary(share->table_name, NULL, table_arg->key_info[curr_index].name, true, txn, true);
        if (error) { goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6626 6627
    }
cleanup:
6628 6629 6630 6631 6632 6633 6634 6635
    if (txn) {
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn,0);
        }
    }
6636
        if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6637 6638 6639 6640
            sql_print_error("Could not drop indexes from table %s because \
another transaction has accessed the table. \
To drop indexes, make sure no transactions touch the table.", share->table_name);
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656
    TOKUDB_DBUG_RETURN(error);
}


//  ***********NOTE*******************
// Although prepare_drop_index is supposed to just get the DB's ready for removal,
// and not actually do the removal, we are doing it here and not in final_drop_index
// For the flags we expose in alter_table_flags, namely xxx_NO_WRITES, this is allowed
// Changes for "future-proofing" this so that it works when we have the equivalent flags
// that are not NO_WRITES are not worth it at the moments, therefore, we can make
// this function just return
int ha_tokudb::final_drop_index(TABLE *table_arg) {
    TOKUDB_DBUG_ENTER("ha_tokudb::final_drop_index");
    TOKUDB_DBUG_RETURN(0);
}

6657
void ha_tokudb::print_error(int error, myf errflag) {
6658
    if (error == DB_LOCK_DEADLOCK)
6659
        error = HA_ERR_LOCK_DEADLOCK;
6660
    if (error == DB_LOCK_NOTGRANTED)
6661
        error = HA_ERR_LOCK_WAIT_TIMEOUT;
6662
#if defined(HA_ERR_DISK_FULL)
6663 6664 6665
    if (error == ENOSPC) {
        error = HA_ERR_DISK_FULL;
    }
6666
#endif
6667 6668 6669
    if (error == DB_KEYEXIST) {
        error = HA_ERR_FOUND_DUPP_KEY;
    }
6670 6671 6672
    handler::print_error(error, errflag);
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6673
#if 0 // QQQ use default
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6674 6675 6676 6677
//
// This function will probably need to be redone from scratch
// if we ever choose to implement it
//
6678
int ha_tokudb::analyze(THD * thd, HA_CHECK_OPT * check_opt) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6679 6680 6681 6682 6683 6684 6685 6686 6687 6688 6689 6690 6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715
    uint i;
    DB_BTREE_STAT *stat = 0;
    DB_TXN_STAT *txn_stat_ptr = 0;
    tokudb_trx_data *trx = (tokudb_trx_data *) thd->ha_data[tokudb_hton->slot];
    DBUG_ASSERT(trx);

    for (i = 0; i < table_share->keys; i++) {
        if (stat) {
            free(stat);
            stat = 0;
        }
        if ((key_file[i]->stat) (key_file[i], trx->all, (void *) &stat, 0))
            goto err;
        share->rec_per_key[i] = (stat->bt_ndata / (stat->bt_nkeys ? stat->bt_nkeys : 1));
    }
    /* A hidden primary key is not in key_file[] */
    if (hidden_primary_key) {
        if (stat) {
            free(stat);
            stat = 0;
        }
        if ((file->stat) (file, trx->all, (void *) &stat, 0))
            goto err;
    }
    pthread_mutex_lock(&share->mutex);
    share->status |= STATUS_TOKUDB_ANALYZE;        // Save status on close
    share->version++;           // Update stat in table
    pthread_mutex_unlock(&share->mutex);
    update_status(share, table);        // Write status to file
    if (stat)
        free(stat);
    return ((share->status & STATUS_TOKUDB_ANALYZE) ? HA_ADMIN_FAILED : HA_ADMIN_OK);

  err:
    if (stat)
        free(stat);
    return HA_ADMIN_FAILED;
6716
}
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6717
#endif
6718

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6719 6720 6721
//
// flatten all DB's in this table, to do so, just do a full scan on every DB
//
6722
int ha_tokudb::optimize(THD * thd, HA_CHECK_OPT * check_opt) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6723
    TOKUDB_DBUG_ENTER("ha_tokudb::optimize");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6724 6725
    int error;
    DBC* tmp_cursor = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6726 6727 6728
    tokudb_trx_data *trx = NULL;
    DB_TXN* txn = NULL;
    bool do_commit = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6729
    uint curr_num_DBs = table->s->keys + test(hidden_primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741

    trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
    if (trx == NULL) {
        error = HA_ERR_UNSUPPORTED;
        goto cleanup;
    }

    //
    // optimize may be called without a valid transaction, so we have to do this
    // in order to get a valid transaction
    // this is a bit hacky, but it is the best we have right now
    //
6742
    txn = trx->sub_sp_level ? trx->sub_sp_level : trx->sp_level;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6743 6744 6745 6746 6747 6748 6749
    if (txn == NULL) {        
        error = db_env->txn_begin(db_env, NULL, &txn, 0);
        if (error) {
            goto cleanup;
        }
        do_commit = true;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6750 6751 6752
    //
    // prelock so each scan goes faster
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6753
    error = acquire_table_lock(txn,lock_read);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6754 6755 6756 6757 6758 6759 6760 6761
    if (error) {
        goto cleanup;
    }

    //
    // for each DB, scan through entire table and do nothing
    //
    for (uint i = 0; i < curr_num_DBs; i++) {
6762 6763
        error = share->key_file[i]->cursor(share->key_file[i], txn, &tmp_cursor, 0);
        if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6764 6765 6766 6767
            tmp_cursor = NULL;
            goto cleanup;
        }
        while (error != DB_NOTFOUND) {
6768
            error = tmp_cursor->c_getf_next(tmp_cursor, DB_PRELOCKED, smart_dbt_do_nothing, NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6769 6770 6771 6772
            if (error && error != DB_NOTFOUND) {
                goto cleanup;
            }
        }
6773 6774
        error = tmp_cursor->c_close(tmp_cursor);
        assert(error==0);
6775
        tmp_cursor = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6776 6777 6778 6779
    }

    error = 0;
cleanup:
6780
    if (tmp_cursor) {
6781 6782
        int r = tmp_cursor->c_close(tmp_cursor);
        assert(r==0);
6783 6784
        tmp_cursor = NULL;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6785
    if (do_commit) {
6786
        commit_txn(txn, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6787
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6788
    TOKUDB_DBUG_RETURN(error);
6789 6790
}

6791 6792 6793 6794 6795 6796 6797 6798 6799 6800
//
// 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);
    
    error = share->key_file[keynr]->close(share->key_file[keynr], 0);
6801
    assert(error == 0);
6802 6803 6804 6805 6806 6807 6808 6809 6810

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

    if (is_pk) {
        error = delete_or_rename_dictionary(
            share->table_name, 
            NULL,
            "main", 
6811
            false, //is_key
6812 6813 6814 6815 6816 6817 6818 6819 6820 6821
            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, 
6822
            true, //is_key
6823 6824 6825 6826 6827 6828 6829
            txn,
            true // is a delete
            );
        if (error) { goto cleanup; }
    }

    if (is_pk) {
6830
        error = create_main_dictionary(share->table_name, table, txn, &share->kc_info);
6831 6832 6833 6834 6835 6836
    }
    else {
        error = create_secondary_dictionary(
            share->table_name, 
            table, 
            &table_share->key_info[keynr], 
6837 6838 6839
            txn,
            &share->kc_info,
            keynr
6840 6841 6842 6843 6844 6845 6846 6847
            );
    }
    if (error) { goto cleanup; }

cleanup:
    return error;
}

6848 6849 6850 6851 6852 6853 6854 6855 6856 6857 6858 6859
// 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;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6860
    uint curr_num_DBs = 0;
6861 6862 6863 6864
    DB_TXN* txn = NULL;

    error = db_env->txn_begin(db_env, 0, &txn, 0);
    if (error) { goto cleanup; }
6865

6866
    if (thd_sql_command(ha_thd()) != SQLCOM_TRUNCATE) {
6867
        share->try_table_lock = true;
6868 6869 6870 6871
        error = HA_ERR_WRONG_COMMAND;
        goto cleanup;
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
6872
    curr_num_DBs = table->s->keys + test(hidden_primary_key);
6873
    for (uint i = 0; i < curr_num_DBs; i++) {
6874 6875
        error = truncate_dictionary(i, txn);
        if (error) { goto cleanup; }
6876 6877 6878
    }

    // zap the row count
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6879
    if (error == 0) {
6880
        share->rows = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6881
    }
6882

6883
    share->try_table_lock = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6884
cleanup:
6885 6886 6887 6888 6889 6890 6891 6892 6893
    if (txn) {
        if (error) {
            abort_txn(txn);
        }
        else {
            commit_txn(txn,0);
        }
    }

6894
            if (error == DB_LOCK_NOTGRANTED && ((tokudb_debug & TOKUDB_DEBUG_HIDE_DDL_LOCK_ERRORS) == 0)) {
6895 6896 6897 6898
                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);
            }
6899 6900 6901 6902 6903 6904 6905 6906 6907 6908 6909
    //
    // 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
6910
                    false, // 
6911 6912 6913 6914 6915 6916 6917
                    NULL
                    );
                assert(!r);
            }
            else {
                r = open_main_dictionary(
                    share->table_name, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
6918
                    false, 
6919 6920 6921 6922 6923 6924
                    NULL
                    );
                assert(!r);
            }
        }
    }
6925 6926 6927
    TOKUDB_DBUG_RETURN(error);
}

6928 6929 6930 6931 6932 6933 6934 6935
void ha_tokudb::set_loader_error(int err) {
    loader_error = err;
}
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;
}
6936