ha_tokudb.cc 147 KB
Newer Older
1
#ifdef USE_PRAGMA_IMPLEMENTATION
2
#pragma implementation          // gcc: Class implementation
3 4 5 6 7
#endif

#define MYSQL_SERVER 1
#include "mysql_priv.h"

8 9 10 11
#if !defined(HA_END_SPACE_KEY) || HA_END_SPACE_KEY != 0
#error
#endif

12 13 14 15
unsigned long my_getphyspages() {
    return sysconf(_SC_PHYS_PAGES);
}

16 17 18 19 20 21
#include <syscall.h>

unsigned int my_tid() {
    return syscall(__NR_gettid);
}

22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
static inline void *thd_data_get(THD *thd, int slot) {
#if MYSQL_VERSION_ID <= 50123
    return thd->ha_data[slot];
#else
    return thd->ha_data[slot].ha_ptr;
#endif
}

static inline void thd_data_set(THD *thd, int slot, void *data) {
#if MYSQL_VERSION_ID <= 50123
    thd->ha_data[slot] = data;
#else
    thd->ha_data[slot].ha_ptr = data;
#endif
}

38 39 40 41 42
#undef PACKAGE
#undef VERSION
#undef HAVE_DTRACE
#undef _DTRACE_VERSION

43
//#include "tokudb_config.h"
44 45 46 47 48 49 50 51 52 53 54 55

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

#include "tokudb_probes.h"

#include "ha_tokudb.h"
#include <mysql/plugin.h>

56
static handler *tokudb_create_handler(handlerton * hton, TABLE_SHARE * table, MEM_ROOT * mem_root);
57

58
handlerton *tokudb_hton;
59 60

typedef struct st_tokudb_trx_data {
61 62 63 64
    DB_TXN *all;
    DB_TXN *stmt;
    DB_TXN *sp_level;
    uint tokudb_lock_count;
65 66
} tokudb_trx_data;

67 68
// QQQ how to tune these?
#define HA_TOKUDB_ROWS_IN_TABLE 10000   /* to get optimization right */
69
#define HA_TOKUDB_RANGE_COUNT   100
70
#define HA_TOKUDB_MAX_ROWS      10000000        /* Max rows in table */
71 72 73 74 75 76
/* extra rows for estimate_rows_upper_bound() */
#define HA_TOKUDB_EXTRA_ROWS    100

/* Bits for share->status */
#define STATUS_PRIMARY_KEY_INIT 1
#define STATUS_ROW_COUNT_INIT   2
77
#define STATUS_TOKUDB_ANALYZE      4
78
#define STATUS_AUTO_INCREMENT_INIT 8
79

80
// tokudb debug tracing
Rich Prohaska's avatar
Rich Prohaska committed
81 82
#define TOKUDB_DEBUG_INIT 1
#define TOKUDB_DEBUG_OPEN 2
83 84 85 86 87
#define TOKUDB_DEBUG_ENTER 4
#define TOKUDB_DEBUG_RETURN 8
#define TOKUDB_DEBUG_ERROR 16
#define TOKUDB_DEBUG_TXN 32
#define TOKUDB_DEBUG_AUTO_INCREMENT 64
Yoni Fogel's avatar
Yoni Fogel committed
88
#define TOKUDB_DEBUG_SAVE_TRACE 128
89

90 91 92
#define TOKUDB_TRACE(f, ...) \
    printf("%d:%s:%d:" f, my_tid(), __FILE__, __LINE__, ##__VA_ARGS__);

93
#define TOKUDB_DBUG_ENTER(f, ...)      \
94
{ \
95
    if (tokudb_debug & TOKUDB_DEBUG_ENTER) { \
96
        TOKUDB_TRACE(f "\n", ##__VA_ARGS__); \
97
    } \
98
} \
99
    DBUG_ENTER(__FUNCTION__);
100

101 102 103 104

#define TOKUDB_DBUG_RETURN(r) \
{ \
    int rr = (r); \
105 106 107
    if ((tokudb_debug & TOKUDB_DEBUG_RETURN) || (rr != 0 && (tokudb_debug & TOKUDB_DEBUG_ERROR))) { \
        TOKUDB_TRACE("%s:return %d\n", __FUNCTION__, rr); \
    } \
108 109
    DBUG_RETURN(rr); \
}
110

111 112
#define TOKUDB_DBUG_DUMP(s, p, len) \
{ \
113
    TOKUDB_TRACE("%s:%s", __FUNCTION__, s); \
114 115 116 117 118
    uint i;                                                             \
    for (i=0; i<len; i++) {                                             \
        printf("%2.2x", ((uchar*)p)[i]);                                \
    }                                                                   \
    printf("\n");                                                       \
119 120
}

121 122
const char *ha_tokudb_ext = ".tokudb";

Rich Prohaska's avatar
Rich Prohaska committed
123
//static my_bool tokudb_shared_data = FALSE;
124 125 126 127 128
static u_int32_t tokudb_init_flags = 
    DB_CREATE | DB_THREAD | DB_PRIVATE | 
    DB_INIT_LOCK | 
    DB_INIT_MPOOL |
    DB_INIT_TXN | 
129 130
    0 | // disabled for 1.0.2 DB_INIT_LOG |
    0;  // disabled for 1.0.1 DB_RECOVER;
131
static u_int32_t tokudb_env_flags = DB_LOG_AUTOREMOVE;
Rich Prohaska's avatar
Rich Prohaska committed
132
//static u_int32_t tokudb_lock_type = DB_LOCK_DEFAULT;
133 134
//static ulong tokudb_log_buffer_size = 0;
//static ulong tokudb_log_file_size = 0;
135
static ulonglong tokudb_cache_size = 0;
136
static uint tokudb_cache_memory_percent = 50;
137
static char *tokudb_home;
138 139 140
//static char *tokudb_tmpdir;
static char *tokudb_data_dir;
static char *tokudb_log_dir;
Rich Prohaska's avatar
Rich Prohaska committed
141 142
//static long tokudb_lock_scan_time = 0;
//static ulong tokudb_region_size = 0;
143
//static ulong tokudb_cache_parts = 1;
144
static ulong tokudb_trans_retry = 1;
145
static ulong tokudb_max_lock;
146
static ulong tokudb_debug;
147 148 149 150 151
#ifdef TOKUDB_VERSION
static char *tokudb_version = TOKUDB_VERSION;
#else
static char *tokudb_version;
#endif
152 153 154

static DB_ENV *db_env;

155 156 157
static const char tokudb_hton_name[] = "TokuDB";
static const int tokudb_hton_name_length = sizeof(tokudb_hton_name) - 1;

158
// thread variables
159

160 161
static MYSQL_THDVAR_BOOL(commit_sync, PLUGIN_VAR_THDLOCAL, "sync on txn commit", 
                         /* check */ NULL, /* update */ NULL, /* default*/ TRUE);
162 163

static void tokudb_print_error(const DB_ENV * db_env, const char *db_errpfx, const char *buffer);
164
static void tokudb_cleanup_log_files(void);
165 166 167 168 169 170 171 172 173 174
static TOKUDB_SHARE *get_share(const char *table_name, TABLE * table);
static int free_share(TOKUDB_SHARE * share, TABLE * table, uint hidden_primary_key, bool mutex_is_locked);
static int write_status(DB * status_block, char *buff, uint length);
static void update_status(TOKUDB_SHARE * share, TABLE * table);
static int tokudb_end(handlerton * hton, ha_panic_function type);
static bool tokudb_flush_logs(handlerton * hton);
static bool tokudb_show_status(handlerton * hton, THD * thd, stat_print_fn * print, enum ha_stat_type);
static int tokudb_close_connection(handlerton * hton, THD * thd);
static int tokudb_commit(handlerton * hton, THD * thd, bool all);
static int tokudb_rollback(handlerton * hton, THD * thd, bool all);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
175
static uint tokudb_alter_table_flags(uint flags);
176
#if 0
177 178 179
static int tokudb_rollback_to_savepoint(handlerton * hton, THD * thd, void *savepoint);
static int tokudb_savepoint(handlerton * hton, THD * thd, void *savepoint);
static int tokudb_release_savepoint(handlerton * hton, THD * thd, void *savepoint);
180
#endif
181
static bool tokudb_show_logs(THD * thd, stat_print_fn * stat_print);
182 183 184 185

static HASH tokudb_open_tables;
pthread_mutex_t tokudb_mutex;

186 187 188
static uchar *tokudb_get_key(TOKUDB_SHARE * share, size_t * length, my_bool not_used __attribute__ ((unused))) {
    *length = share->table_name_length;
    return (uchar *) share->table_name;
189 190
}

191
static int tokudb_init_func(void *p) {
192
    TOKUDB_DBUG_ENTER("tokudb_init_func");
193

194
    tokudb_hton = (handlerton *) p;
195

196 197
    VOID(pthread_mutex_init(&tokudb_mutex, MY_MUTEX_INIT_FAST));
    (void) hash_init(&tokudb_open_tables, system_charset_info, 32, 0, 0, (hash_get_key) tokudb_get_key, 0, 0);
198

199
    tokudb_hton->state = SHOW_OPTION_YES;
200
    // tokudb_hton->flags= HTON_CAN_RECREATE;  // QQQ this came from skeleton
201
    tokudb_hton->flags = HTON_CLOSE_CURSORS_AT_COMMIT | HTON_FLUSH_AFTER_RENAME;
202
#ifdef DB_TYPE_TOKUDB
203 204 205
    tokudb_hton->db_type = DB_TYPE_TOKUDB;
#else
    tokudb_hton->db_type = DB_TYPE_UNKNOWN;
206 207
#endif

208 209
    tokudb_hton->create = tokudb_create_handler;
    tokudb_hton->close_connection = tokudb_close_connection;
210 211
#if 0
    tokudb_hton->savepoint_offset = sizeof(DB_TXN *);
212 213 214
    tokudb_hton->savepoint_set = tokudb_savepoint;
    tokudb_hton->savepoint_rollback = tokudb_rollback_to_savepoint;
    tokudb_hton->savepoint_release = tokudb_release_savepoint;
215
#endif
216 217 218 219
    if (tokudb_init_flags & DB_INIT_TXN) {
        tokudb_hton->commit = tokudb_commit;
        tokudb_hton->rollback = tokudb_rollback;
    }
220 221 222
    tokudb_hton->panic = tokudb_end;
    tokudb_hton->flush_logs = tokudb_flush_logs;
    tokudb_hton->show_status = tokudb_show_status;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
223
    tokudb_hton->alter_table_flags = tokudb_alter_table_flags;
224
#if 0
225 226 227
    if (!tokudb_tmpdir)
        tokudb_tmpdir = mysql_tmpdir;
    DBUG_PRINT("info", ("tokudb_tmpdir: %s", tokudb_tmpdir));
228
#endif
229 230 231
    if (!tokudb_home)
        tokudb_home = mysql_real_data_home;
    DBUG_PRINT("info", ("tokudb_home: %s", tokudb_home));
232
#if 0
233
    if (!tokudb_log_buffer_size) { // QQQ
234 235 236 237 238 239 240
        tokudb_log_buffer_size = max(table_cache_size * 512, 32 * 1024);
        DBUG_PRINT("info", ("computing tokudb_log_buffer_size %ld\n", tokudb_log_buffer_size));
    }
    tokudb_log_file_size = tokudb_log_buffer_size * 4;
    tokudb_log_file_size = MY_ALIGN(tokudb_log_file_size, 1024 * 1024L);
    tokudb_log_file_size = max(tokudb_log_file_size, 10 * 1024 * 1024L);
    DBUG_PRINT("info", ("computing tokudb_log_file_size: %ld\n", tokudb_log_file_size));
241
#endif
242 243 244
    int r;
    if ((r = db_env_create(&db_env, 0))) {
        DBUG_PRINT("info", ("db_env_create %d\n", r));
245
        goto error;
246 247
    }

Rich Prohaska's avatar
Rich Prohaska committed
248
    DBUG_PRINT("info", ("tokudb_env_flags: 0x%x\n", tokudb_env_flags));
249
    r = db_env->set_flags(db_env, tokudb_env_flags, 1);
250
    if (r) { // QQQ
Rich Prohaska's avatar
Rich Prohaska committed
251
        if (tokudb_debug & TOKUDB_DEBUG_INIT) 
252 253
            TOKUDB_TRACE("%s:WARNING: flags=%x r=%d\n", __FUNCTION__, tokudb_env_flags, r); 
        // goto error;
254
    }
255 256

    // config error handling
257
    db_env->set_errcall(db_env, tokudb_print_error);
258 259 260
    db_env->set_errpfx(db_env, "TokuDB");

    // config directories
261
#if 0
262 263
    DBUG_PRINT("info", ("tokudb_tmpdir: %s\n", tokudb_tmpdir));
    db_env->set_tmp_dir(db_env, tokudb_tmpdir);
264 265 266 267 268 269 270 271 272 273 274 275 276
#endif

    {
    char *data_dir = tokudb_data_dir;
    if (data_dir == 0) 
        data_dir = mysql_data_home;
    DBUG_PRINT("info", ("tokudb_data_dir: %s\n", data_dir));
    db_env->set_data_dir(db_env, data_dir);
    }

    if (tokudb_log_dir) {
        DBUG_PRINT("info", ("tokudb_log_dir: %s\n", tokudb_log_dir));
        db_env->set_lg_dir(db_env, tokudb_log_dir);
277 278
    }

279
    // config the cache table
280 281 282 283
    if (tokudb_cache_size == 0) {
        unsigned long pagesize = my_getpagesize();
        unsigned long long npages = my_getphyspages();
        unsigned long long physmem = npages * pagesize;
284
        tokudb_cache_size = (ulonglong) (physmem * (tokudb_cache_memory_percent / 100.0));
285
    }
286 287
    if (tokudb_cache_size) {
        DBUG_PRINT("info", ("tokudb_cache_size: %lld\n", tokudb_cache_size));
288
        r = db_env->set_cachesize(db_env, tokudb_cache_size / (1024 * 1024L * 1024L), tokudb_cache_size % (1024L * 1024L * 1024L), 1);
289 290 291 292
        if (r) {
            DBUG_PRINT("info", ("set_cachesize %d\n", r));
            goto error; 
        }
293 294 295 296
    }
    u_int32_t gbytes, bytes; int parts;
    r = db_env->get_cachesize(db_env, &gbytes, &bytes, &parts);
    if (r == 0) 
Rich Prohaska's avatar
Rich Prohaska committed
297
        if (tokudb_debug & TOKUDB_DEBUG_INIT) 
298
            TOKUDB_TRACE("%s:tokudb_cache_size=%lld\n", __FUNCTION__, ((unsigned long long) gbytes << 30) + bytes);
299

300 301
#if 0
    // QQQ config the logs
302 303 304 305 306 307
    DBUG_PRINT("info", ("tokudb_log_file_size: %ld\n", tokudb_log_file_size));
    db_env->set_lg_max(db_env, tokudb_log_file_size);
    DBUG_PRINT("info", ("tokudb_log_buffer_size: %ld\n", tokudb_log_buffer_size));
    db_env->set_lg_bsize(db_env, tokudb_log_buffer_size);
    // DBUG_PRINT("info",("tokudb_region_size: %ld\n", tokudb_region_size));
    // db_env->set_lg_regionmax(db_env, tokudb_region_size);
308
#endif
309

310 311
    // config the locks
#if 0 // QQQ no lock types yet
312 313
    DBUG_PRINT("info", ("tokudb_lock_type: 0x%lx\n", tokudb_lock_type));
    db_env->set_lk_detect(db_env, tokudb_lock_type);
314 315 316 317 318 319 320 321 322
#endif
    if (tokudb_max_lock) {
        DBUG_PRINT("info",("tokudb_max_lock: %ld\n", tokudb_max_lock));
        r = db_env->set_lk_max_locks(db_env, tokudb_max_lock);
        if (r) {
            DBUG_PRINT("info", ("tokudb_set_max_locks %d\n", r));
            goto error;
        }
    }
323

324 325 326 327 328 329 330
    if (tokudb_debug & TOKUDB_DEBUG_INIT) TOKUDB_TRACE("%s:env open:flags=%x\n", __FUNCTION__, tokudb_init_flags);

    r = db_env->open(db_env, tokudb_home, tokudb_init_flags, 0666);

    if (tokudb_debug & TOKUDB_DEBUG_INIT) TOKUDB_TRACE("%s:env opened:return=%d\n", __FUNCTION__, r);

    if (r) {
331
        DBUG_PRINT("info", ("env->open %d\n", r));
332 333
        goto error;
    }
334

335
    DBUG_RETURN(FALSE);
336 337 338 339 340 341

error:
    if (db_env) {
        db_env->close(db_env, 0);
        db_env = 0;
    }
342 343 344 345
    DBUG_RETURN(TRUE);
}

static int tokudb_done_func(void *p) {
346
    TOKUDB_DBUG_ENTER("tokudb_done_func");
347 348 349 350 351 352
    int error = 0;

    if (tokudb_open_tables.records)
        error = 1;
    hash_free(&tokudb_open_tables);
    pthread_mutex_destroy(&tokudb_mutex);
353
    TOKUDB_DBUG_RETURN(0);
354 355 356 357 358 359 360
}

/** @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.
*/
361 362 363
static TOKUDB_SHARE *get_share(const char *table_name, TABLE * table) {
    TOKUDB_SHARE *share;
    uint length;
364

365 366 367 368 369 370
    pthread_mutex_lock(&tokudb_mutex);
    length = (uint) strlen(table_name);

    if (!(share = (TOKUDB_SHARE *) hash_search(&tokudb_open_tables, (uchar *) table_name, length))) {
        ulong *rec_per_key;
        char *tmp_name;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
371
        uint num_keys = table->s->keys;
372

373 374 375 376
        if (!(share = (TOKUDB_SHARE *) 
            my_multi_malloc(MYF(MY_WME | MY_ZEROFILL), 
                            &share, sizeof(*share),
                            &tmp_name, length + 1, 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
377
                            &rec_per_key, num_keys * sizeof(ha_rows), 
378
                            NullS))) {
379 380 381 382 383 384 385
            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);
386

387
        share->rec_per_key = rec_per_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
388
        bzero((void *) share->key_file, sizeof(share->key_file));
389 390 391 392 393

        if (my_hash_insert(&tokudb_open_tables, (uchar *) share))
            goto error;
        thr_lock_init(&share->lock);
        pthread_mutex_init(&share->mutex, MY_MUTEX_INIT_FAST);
394
    }
395
    pthread_mutex_unlock(&tokudb_mutex);
396

397
    return share;
398

399
  error:
400 401 402
    pthread_mutex_destroy(&share->mutex);
    my_free((uchar *) share, MYF(0));

403
    return NULL;
404 405
}

406 407
static int free_share(TOKUDB_SHARE * share, TABLE * table, uint hidden_primary_key, bool mutex_is_locked) {
    int error, result = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
408
    uint num_keys = table->s->keys + test(hidden_primary_key);
409 410 411 412 413 414 415 416 417 418 419 420

    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));
        DB **key_file = share->key_file;

        /* this does share->file->close() implicitly */
        update_status(share, table);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
421
        for (uint i = 0; i < num_keys; i++) {
422
            if (tokudb_debug & TOKUDB_DEBUG_OPEN)
423
                TOKUDB_TRACE("dbclose:%p\n", key_file[i]);
424 425 426 427 428 429 430 431 432 433 434 435 436 437 438
            if (key_file[i] && (error = key_file[i]->close(key_file[i], 0)))
                result = error;
        }

        if (share->status_block && (error = share->status_block->close(share->status_block, 0)))
            result = error;

        hash_delete(&tokudb_open_tables, (uchar *) share);
        thr_lock_delete(&share->lock);
        pthread_mutex_destroy(&share->mutex);
        my_free((uchar *) share, MYF(0));
    }
    pthread_mutex_unlock(&tokudb_mutex);

    return result;
439 440
}

441 442 443 444 445
static handler *tokudb_create_handler(handlerton * hton, TABLE_SHARE * table, MEM_ROOT * mem_root) {
    return new(mem_root) ha_tokudb(hton, table);
}

int tokudb_end(handlerton * hton, ha_panic_function type) {
446
    TOKUDB_DBUG_ENTER("tokudb_end");
447 448
    int error = 0;
    if (db_env) {
449 450
        if (tokudb_init_flags & DB_INIT_LOG)
            tokudb_cleanup_log_files();
451 452 453
        error = db_env->close(db_env, 0);       // Error is logged
        db_env = 0;
    }
454
    TOKUDB_DBUG_RETURN(error);
455 456 457
}

static int tokudb_close_connection(handlerton * hton, THD * thd) {
458
    my_free(thd_data_get(thd, hton->slot), MYF(0));
459 460 461 462
    return 0;
}

bool tokudb_flush_logs(handlerton * hton) {
463
    TOKUDB_DBUG_ENTER("tokudb_flush_logs");
464 465
    int error;
    bool result = 0;
466 467 468 469 470 471 472 473 474
    if (tokudb_init_flags & DB_INIT_LOG) {
        if ((error = db_env->log_flush(db_env, 0))) {
            my_error(ER_ERROR_DURING_FLUSH_LOGS, MYF(0), error);
            result = 1;
        }
        if ((error = db_env->txn_checkpoint(db_env, 0, 0, 0))) {
            my_error(ER_ERROR_DURING_CHECKPOINT, MYF(0), error);
            result = 1;
        }
475
    }
476
    TOKUDB_DBUG_RETURN(result);
477 478 479
}

static int tokudb_commit(handlerton * hton, THD * thd, bool all) {
480
    TOKUDB_DBUG_ENTER("tokudb_commit");
481
    DBUG_PRINT("trans", ("ending transaction %s", all ? "all" : "stmt"));
482
    u_int32_t syncflag = THDVAR(thd, commit_sync) ? 0 : DB_TXN_NOSYNC;
483
    tokudb_trx_data *trx = (tokudb_trx_data *) thd_data_get(thd, hton->slot);
484
    DB_TXN **txn = all ? &trx->all : &trx->stmt;
485 486
    int error = 0;
    if (*txn) {
487
        if (tokudb_debug & TOKUDB_DEBUG_TXN) 
488
            TOKUDB_TRACE("commit:%d:%p\n", all, *txn);
489
        error = (*txn)->commit(*txn, syncflag);
490 491 492 493
        if (*txn == trx->sp_level)
            trx->sp_level = 0;
        *txn = 0;
    } else
494
        if (tokudb_debug & TOKUDB_DEBUG_TXN) 
495
            TOKUDB_TRACE("commit0\n");
496
    TOKUDB_DBUG_RETURN(error);
497 498 499
}

static int tokudb_rollback(handlerton * hton, THD * thd, bool all) {
500
    TOKUDB_DBUG_ENTER("tokudb_rollback");
501
    DBUG_PRINT("trans", ("aborting transaction %s", all ? "all" : "stmt"));
502
    tokudb_trx_data *trx = (tokudb_trx_data *) thd_data_get(thd, hton->slot);
503
    DB_TXN **txn = all ? &trx->all : &trx->stmt;
504 505
    int error = 0;
    if (*txn) {
506
        if (tokudb_debug & TOKUDB_DEBUG_TXN)
507
            TOKUDB_TRACE("rollback:%p\n", *txn);
508 509 510 511 512
        error = (*txn)->abort(*txn);
	if (*txn == trx->sp_level)
	    trx->sp_level = 0;
	*txn = 0;
    } else
513
        if (tokudb_debug & TOKUDB_DEBUG_TXN) 
514
            TOKUDB_TRACE("abort0\n");
515
    TOKUDB_DBUG_RETURN(error);
516 517
}

518 519
#if 0

520
static int tokudb_savepoint(handlerton * hton, THD * thd, void *savepoint) {
521
    TOKUDB_DBUG_ENTER("tokudb_savepoint");
522 523 524 525 526 527
    int error;
    DB_TXN **save_txn = (DB_TXN **) savepoint;
    tokudb_trx_data *trx = (tokudb_trx_data *) thd->ha_data[hton->slot];
    if (!(error = db_env->txn_begin(db_env, trx->sp_level, save_txn, 0))) {
        trx->sp_level = *save_txn;
    }
528
    TOKUDB_DBUG_RETURN(error);
529 530 531
}

static int tokudb_rollback_to_savepoint(handlerton * hton, THD * thd, void *savepoint) {
532
    TOKUDB_DBUG_ENTER("tokudb_rollback_to_savepoint");
533 534 535 536 537 538 539 540
    int error;
    DB_TXN *parent, **save_txn = (DB_TXN **) savepoint;
    tokudb_trx_data *trx = (tokudb_trx_data *) thd->ha_data[hton->slot];
    parent = (*save_txn)->parent;
    if (!(error = (*save_txn)->abort(*save_txn))) {
        trx->sp_level = parent;
        error = tokudb_savepoint(hton, thd, savepoint);
    }
541
    TOKUDB_DBUG_RETURN(error);
542 543 544
}

static int tokudb_release_savepoint(handlerton * hton, THD * thd, void *savepoint) {
545
    TOKUDB_DBUG_ENTER("tokudb_release_savepoint");
546 547 548 549 550 551 552 553
    int error;
    DB_TXN *parent, **save_txn = (DB_TXN **) savepoint;
    tokudb_trx_data *trx = (tokudb_trx_data *) thd->ha_data[hton->slot];
    parent = (*save_txn)->parent;
    if (!(error = (*save_txn)->commit(*save_txn, 0))) {
        trx->sp_level = parent;
        *save_txn = 0;
    }
554
    TOKUDB_DBUG_RETURN(error);
555
}
556

557
#endif
558 559

static bool tokudb_show_logs(THD * thd, stat_print_fn * stat_print) {
560
    TOKUDB_DBUG_ENTER("tokudb_show_logs");
561 562 563 564 565 566 567 568 569
    char **all_logs, **free_logs, **a, **f;
    int error = 1;
    MEM_ROOT **root_ptr = my_pthread_getspecific_ptr(MEM_ROOT **, THR_MALLOC);
    MEM_ROOT show_logs_root, *old_mem_root = *root_ptr;

    init_sql_alloc(&show_logs_root, BDB_LOG_ALLOC_BLOCK_SIZE, BDB_LOG_ALLOC_BLOCK_SIZE);
    *root_ptr = &show_logs_root;
    all_logs = free_logs = 0;

570 571
    error = db_env->log_archive(db_env, &all_logs, 0);
    if (error) {
572
        DBUG_PRINT("error", ("log_archive failed (error %d)", error));
573
        db_env->err(db_env, error, "log_archive");
574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597
        if (error == DB_NOTFOUND)
            error = 0;          // No log files
        goto err;
    }
    /* Error is 0 here */
    if (all_logs) {
        for (a = all_logs, f = free_logs; *a; ++a) {
            if (f && *f && strcmp(*a, *f) == 0) {
                f++;
                if ((error = stat_print(thd, tokudb_hton_name, tokudb_hton_name_length, *a, strlen(*a), STRING_WITH_LEN(SHOW_LOG_STATUS_FREE))))
                    break;
            } else {
                if ((error = stat_print(thd, tokudb_hton_name, tokudb_hton_name_length, *a, strlen(*a), STRING_WITH_LEN(SHOW_LOG_STATUS_INUSE))))
                    break;
            }
        }
    }
  err:
    if (all_logs)
        free(all_logs);
    if (free_logs)
        free(free_logs);
    free_root(&show_logs_root, MYF(0));
    *root_ptr = old_mem_root;
598
    TOKUDB_DBUG_RETURN(error);
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614
}

bool tokudb_show_status(handlerton * hton, THD * thd, stat_print_fn * stat_print, enum ha_stat_type stat_type) {
    switch (stat_type) {
    case HA_ENGINE_LOGS:
        return tokudb_show_logs(thd, stat_print);
    default:
        return FALSE;
    }
}

static void tokudb_print_error(const DB_ENV * db_env, const char *db_errpfx, const char *buffer) {
    sql_print_error("%s:  %s", db_errpfx, buffer);
}

void tokudb_cleanup_log_files(void) {
615
    TOKUDB_DBUG_ENTER("tokudb_cleanup_log_files");
616 617 618 619 620
    char **names;
    int error;

    if ((error = db_env->txn_checkpoint(db_env, 0, 0, 0)))
        my_error(ER_ERROR_DURING_CHECKPOINT, MYF(0), error);
621

622
    if ((error = db_env->log_archive(db_env, &names, 0)) != 0) {
623
        DBUG_PRINT("error", ("log_archive failed (error %d)", error));
624
        db_env->err(db_env, error, "log_archive");
625 626
        DBUG_VOID_RETURN;
    }
627

628 629
    if (names) {
        char **np;
630
        for (np = names; *np; ++np) {
631 632
#if 1
            if (tokudb_debug)
633
                TOKUDB_TRACE("%s:cleanup:%s\n", __FUNCTION__, *np);
634 635 636
#else
            my_delete(*np, MYF(MY_WME));
#endif
637
        }
638

639 640
        free(names);
    }
641

642
    DBUG_VOID_RETURN;
643 644
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
645 646 647 648 649 650 651 652 653 654 655 656 657 658
//
// *******NOTE*****
// If the flags HA_ONLINE_DROP_INDEX and HA_ONLINE_DROP_UNIQUE_INDEX
// are ever added, prepare_drop_index and final_drop_index will need to be modified
// so that the actual deletion of DB's is done in final_drop_index and not prepare_drop_index
//
static uint tokudb_alter_table_flags(uint flags)
{
    return (HA_ONLINE_ADD_INDEX_NO_WRITES| HA_ONLINE_DROP_INDEX_NO_WRITES |
            HA_ONLINE_ADD_UNIQUE_INDEX_NO_WRITES| HA_ONLINE_DROP_UNIQUE_INDEX_NO_WRITES);

}


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
static int get_name_length(const char *name) {
    int n = 0;
    const char *newname = name;
    if (tokudb_data_dir) {
        n += strlen(tokudb_data_dir) + 1;
        if (strncmp("./", name, 2) == 0) 
            newname = name + 2;
    }
    n += strlen(newname);
    n += strlen(ha_tokudb_ext);
    return n;
}

static void make_name(char *newname, const char *tablename, const char *dictname) {
    const char *newtablename = tablename;
    char *nn = newname;
    if (tokudb_data_dir) {
        nn += sprintf(nn, "%s/", tokudb_data_dir);
        if (strncmp("./", tablename, 2) == 0)
            newtablename = tablename + 2;
    }
    nn += sprintf(nn, "%s%s", newtablename, ha_tokudb_ext);
    if (dictname)
        nn += sprintf(nn, "/%s%s", dictname, ha_tokudb_ext);
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
685

Zardosht Kasheff's avatar
Zardosht Kasheff committed
686 687 688 689 690 691 692

#define CHECK_VALID_CURSOR() \
    if (cursor == NULL) { \
        error = last_cursor_error; \
        goto cleanup; \
    }

693 694
ha_tokudb::ha_tokudb(handlerton * hton, TABLE_SHARE * table_arg)
:  
Zardosht Kasheff's avatar
Zardosht Kasheff committed
695
    handler(hton, table_arg), alloc_ptr(0), rec_buff(0),
Zardosht Kasheff's avatar
Zardosht Kasheff committed
696
    // flags defined in sql\handler.h
697 698
    int_table_flags(HA_REC_NOT_IN_SEQ | HA_FAST_KEY_READ | HA_NULL_IN_KEY | HA_CAN_INDEX_BLOBS | HA_PRIMARY_KEY_IN_READ_INDEX | 
                    HA_FILE_BASED | HA_CAN_GEOMETRY | HA_AUTO_PART_KEY | HA_TABLE_SCAN_ON_INDEX), 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
699
    changed_rows(0), last_dup_key((uint) - 1), version(0), using_ignore(0), last_cursor_error(0),range_lock_grabbed(false), primary_key_offsets(NULL) {
700
}
701

702 703 704
static const char *ha_tokudb_exts[] = {
    ha_tokudb_ext,
    NullS
705 706
};

Zardosht Kasheff's avatar
Zardosht Kasheff committed
707 708 709
/* 
 *  returns NULL terminated file extension string
 */
710
const char **ha_tokudb::bas_ext() const {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
711 712
    TOKUDB_DBUG_ENTER("ha_tokudb::bas_ext");
    DBUG_RETURN(ha_tokudb_exts);
713 714
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
715 716 717 718
//
// Returns a bit mask of capabilities of the key or its part specified by 
// the arguments. The capabilities are defined in sql/handler.h.
//
719
ulong ha_tokudb::index_flags(uint idx, uint part, bool all_parts) const {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
720
    TOKUDB_DBUG_ENTER("ha_tokudb::index_flags");
721
    ulong flags = (HA_READ_NEXT | HA_READ_PREV | HA_READ_ORDER | HA_KEYREAD_ONLY | HA_READ_RANGE);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
722
    DBUG_RETURN(flags);
723 724 725 726 727 728 729 730
}

static int tokudb_cmp_hidden_key(DB * file, const DBT * new_key, const DBT * saved_key) {
    ulonglong a = uint5korr((char *) new_key->data);
    ulonglong b = uint5korr((char *) saved_key->data);
    return a < b ? -1 : (a > b ? 1 : 0);
}

Yoni Fogel's avatar
Yoni Fogel committed
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
/*
    Things that are required for ALL data types:
        key_part->field->null_bit
        key_part->length
        key_part->field->packed_col_length(...)
            DEFAULT: virtual uint packed_col_length(const uchar *to, uint length)
                { return length;}
            All integer types use this.
            String types MIGHT use different one, espescially the varchars
        key_part->field->pack_cmp(...)
            DEFAULT: virtual int pack_cmp(...)
                { return cmp(a,b); }
            All integer types use the obvious one.
            Assume X byte bytestream, int =:
            ((u_int64_t)((u_int8_t)bytes[0])) << 0 | 
            ((u_int64_t)((u_int8_t)bytes[1])) << 8 | 
            ((u_int64_t)((u_int8_t)bytes[2])) << 16 | 
            ((u_int64_t)((u_int8_t)bytes[3])) << 24 | 
            ((u_int64_t)((u_int8_t)bytes[4])) << 32 | 
            ((u_int64_t)((u_int8_t)bytes[5])) << 40 | 
            ((u_int64_t)((u_int8_t)bytes[6])) << 48 | 
            ((u_int64_t)((u_int8_t)bytes[7])) << 56
            If the integer type is < 8 bytes, just skip the unneeded ones.
            Then compare the integers in the obvious way.
        Strings:
            Empty space differences at end are ignored.
            i.e. delete all empty space at end first, and then compare.
    Possible prerequisites:
        key_part->field->cmp
            NO DEFAULT
*/

typedef enum {
    TOKUTRACE_SIGNED_INTEGER   = 0,
    TOKUTRACE_UNSIGNED_INTEGER = 1,
    TOKUTRACE_CHAR = 2
} tokutrace_field_type;

typedef struct {
    tokutrace_field_type    type;
    bool                    null_bit;
    u_int32_t               length;
} tokutrace_field;

typedef struct {
    u_int16_t           version;
    u_int32_t           num_fields;
    tokutrace_field     fields[0];
} tokutrace_cmp_fun;

static int tokutrace_db_get_cmp_byte_stream(DB* db, DBT* byte_stream) {
    int r      = ENOSYS;
    void* data = NULL;
    KEY* key   = NULL;
    if (byte_stream->flags != DB_DBT_MALLOC) { return EINVAL; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
786
    bzero((void *) byte_stream, sizeof(*byte_stream));
Yoni Fogel's avatar
Yoni Fogel committed
787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850

    u_int32_t num_fields = 0;
    if (!db->app_private) { num_fields = 1; }
    else {
        key = (KEY*)db->app_private;
        num_fields = key->key_parts;
    }
    size_t need_size = sizeof(tokutrace_cmp_fun) +
                       num_fields * sizeof(tokutrace_field);

    data = my_malloc(need_size, MYF(MY_FAE | MY_ZEROFILL | MY_WME));
    if (!data) { return ENOMEM; }

    tokutrace_cmp_fun* info = (tokutrace_cmp_fun*)data;
    info->version     = 1;
    info->num_fields  = num_fields;
    
    if (!db->app_private) {
        info->fields[0].type     = TOKUTRACE_UNSIGNED_INTEGER;
        info->fields[0].null_bit = false;
        info->fields[0].length   = 40 / 8;
        goto finish;
    }
    assert(db->app_private);
    assert(key);
    u_int32_t i;
    for (i = 0; i < num_fields; i++) {
        info->fields[i].null_bit = key->key_part[i].null_bit;
        info->fields[i].length   = key->key_part[i].length;
        enum_field_types type    = key->key_part[i].field->type();
        switch (type) {
#ifdef HAVE_LONG_LONG
            case (MYSQL_TYPE_LONGLONG):
#endif
            case (MYSQL_TYPE_LONG):
            case (MYSQL_TYPE_INT24):
            case (MYSQL_TYPE_SHORT):
            case (MYSQL_TYPE_TINY): {
                /* Integer */
                Field_num* field = static_cast<Field_num*>(key->key_part[i].field);
                if (field->unsigned_flag) {
                    info->fields[i].type = TOKUTRACE_UNSIGNED_INTEGER; }
                else {
                    info->fields[i].type = TOKUTRACE_SIGNED_INTEGER; }
                break;
            }
            default: {
                fprintf(stderr, "Cannot save cmp function for type %d.\n", type);
                r = ENOSYS;
                goto cleanup;
            }
        }
    }
finish:
    byte_stream->data = data;
    byte_stream->size = need_size;
    r = 0;
cleanup:
    if (r!=0) {
        if (data) { my_free(data, MYF(0)); }
    }
    return r;
}

851
static int tokudb_compare_two_keys(KEY *key, const DBT * new_key, const DBT * saved_key, bool cmp_prefix) {
852 853 854 855
    uchar *new_key_ptr = (uchar *) new_key->data;
    uchar *saved_key_ptr = (uchar *) saved_key->data;
    KEY_PART_INFO *key_part = key->key_part, *end = key_part + key->key_parts;
    uint key_length = new_key->size;
856
    uint saved_key_length = saved_key->size;
857 858

    //DBUG_DUMP("key_in_index", saved_key_ptr, saved_key->size);
859
    for (; key_part != end && (int) key_length > 0 && (int) saved_key_length > 0; key_part++) {
860
        int cmp;
Yoni Fogel's avatar
Yoni Fogel committed
861 862
        uint new_key_field_length;
        uint saved_key_field_length;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
863
        if (key_part->field->null_bit) {
Yoni Fogel's avatar
Yoni Fogel committed
864
            assert(new_key_ptr   < (uchar *) new_key->data   + new_key->size);
865
            assert(saved_key_ptr < (uchar *) saved_key->data + saved_key->size);
Yoni Fogel's avatar
Yoni Fogel committed
866 867 868
            if (*new_key_ptr != *saved_key_ptr) {
                return ((int) *new_key_ptr - (int) *saved_key_ptr); }
            saved_key_ptr++;
869
            key_length--;
870
            saved_key_length--;
Yoni Fogel's avatar
Yoni Fogel committed
871
            if (!*new_key_ptr++) { continue; }
872
        }
Yoni Fogel's avatar
Yoni Fogel committed
873 874 875 876
        new_key_field_length     = key_part->field->packed_col_length(new_key_ptr,   key_part->length);
        saved_key_field_length   = key_part->field->packed_col_length(saved_key_ptr, key_part->length);
        assert(      key_length >= new_key_field_length);
        assert(saved_key_length >= saved_key_field_length);
877 878
        if ((cmp = key_part->field->pack_cmp(new_key_ptr, saved_key_ptr, key_part->length, 0)))
            return cmp;
Yoni Fogel's avatar
Yoni Fogel committed
879 880 881 882
        new_key_ptr      += new_key_field_length;
        key_length       -= new_key_field_length;
        saved_key_ptr    += saved_key_field_length;
        saved_key_length -= saved_key_field_length;
883
    }
884
    return cmp_prefix ? 0 : key_length - saved_key_length;
885 886
}

887
static int tokudb_cmp_packed_key(DB *file, const DBT *keya, const DBT *keyb) {
888 889
    assert(file->app_private != 0);
    KEY *key = (KEY *) file->app_private;
890 891
    return tokudb_compare_two_keys(key, keya, keyb, false);
}
892

893 894 895 896 897 898 899 900 901 902 903
static int tokudb_cmp_primary_key(DB *file, const DBT *keya, const DBT *keyb) {
    assert(file->app_private != 0);
    KEY *key = (KEY *) file->api_internal;
    return tokudb_compare_two_keys(key, keya, keyb, false);
}

//TODO: QQQ Only do one direction for prefix.
static int tokudb_prefix_cmp_packed_key(DB *file, const DBT *keya, const DBT *keyb) {
    assert(file->app_private != 0);
    KEY *key = (KEY *) file->app_private;
    return tokudb_compare_two_keys(key, keya, keyb, true);
904 905
}

906
#if 0
907
/* Compare key against row */
908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933
static bool tokudb_key_cmp(TABLE * table, KEY * key_info, const uchar * key, uint key_length) {
    KEY_PART_INFO *key_part = key_info->key_part, *end = key_part + key_info->key_parts;

    for (; key_part != end && (int) key_length > 0; key_part++) {
        int cmp;
        uint length;
        if (key_part->null_bit) {
            key_length--;
            /*
               With the current usage, the following case will always be FALSE,
               because NULL keys are sorted before any other key
             */
            if (*key != (table->record[0][key_part->null_offset] & key_part->null_bit) ? 0 : 1)
                return 1;
            if (!*key++)        // Null value
                continue;
        }
        /*
           Last argument has to be 0 as we are also using this to function to see
           if a key like 'a  ' matched a row with 'a'
         */
        if ((cmp = key_part->field->pack_cmp(key, key_part->length, 0)))
            return cmp;
        length = key_part->field->packed_col_length(key, key_part->length);
        key += length;
        key_length -= length;
934
    }
935 936
    return 0;                   // Identical keys
}
937
#endif
938

Zardosht Kasheff's avatar
Zardosht Kasheff committed
939 940 941 942 943 944
int primary_key_part_compare (const void* left, const void* right) {
    PRIM_KEY_PART_INFO* left_part= (PRIM_KEY_PART_INFO *)left;
    PRIM_KEY_PART_INFO* right_part = (PRIM_KEY_PART_INFO *)right;
    return left_part->offset - right_part->offset;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998
//
// Open a secondary table, the key will be a secondary index, the data will be a primary key
//
int ha_tokudb::open_secondary_table(DB** ptr, KEY* key_info, const char* name, int mode, u_int32_t* key_type) {
    int error = ENOSYS;
    char part[MAX_ALIAS_NAME + 10];
    char name_buff[FN_REFLEN];
    uint open_flags = (mode == O_RDONLY ? DB_RDONLY : 0) | DB_THREAD;
    char newname[strlen(name) + 32];
    DBT cmp_byte_stream;

    if (tokudb_init_flags & DB_INIT_TXN)
        open_flags += DB_AUTO_COMMIT;

    if ((error = db_create(ptr, db_env, 0))) {
        my_errno = error;
        goto cleanup;
    }
    sprintf(part, "key-%s", key_info->name);
    make_name(newname, name, part);
    fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
    *key_type = key_info->flags & HA_NOSAME ? DB_NOOVERWRITE : DB_YESOVERWRITE;
    (*ptr)->app_private = (void *) (key_info);
    if (tokudb_debug & TOKUDB_DEBUG_SAVE_TRACE) {
        bzero((void *) &cmp_byte_stream, sizeof(cmp_byte_stream));
        cmp_byte_stream.flags = DB_DBT_MALLOC;
        if ((error = tokutrace_db_get_cmp_byte_stream(*ptr, &cmp_byte_stream))) {
            my_errno = error;
            goto cleanup;
        }
        (*ptr)->set_bt_compare(*ptr, tokudb_cmp_packed_key);
        my_free(cmp_byte_stream.data, MYF(0));
    }
    else
        (*ptr)->set_bt_compare(*ptr, tokudb_cmp_packed_key);    
    if (!(key_info->flags & HA_NOSAME)) {
        DBUG_PRINT("info", ("Setting DB_DUP+DB_DUPSORT for key %s\n", key_info->name));
        (*ptr)->set_flags(*ptr, DB_DUP + DB_DUPSORT);
        (*ptr)->api_internal = share->file->app_private;
        (*ptr)->set_dup_compare(*ptr, hidden_primary_key ? tokudb_cmp_hidden_key : tokudb_cmp_primary_key);
    }
    if ((error = (*ptr)->open(*ptr, 0, name_buff, NULL, DB_BTREE, open_flags, 0))) {
        my_errno = error;
        goto cleanup;
    }
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
        TOKUDB_TRACE("open:%s:file=%p\n", newname, *ptr);
    }
cleanup:
    return error;
}



Zardosht Kasheff's avatar
Zardosht Kasheff committed
999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
//
// 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
//
1010
int ha_tokudb::open(const char *name, int mode, uint test_if_locked) {
1011
    TOKUDB_DBUG_ENTER("ha_tokudb::open %p %s", this, name);
1012 1013 1014
    TOKUDB_OPEN();

    char name_buff[FN_REFLEN];
1015
    uint open_flags = (mode == O_RDONLY ? DB_RDONLY : 0) | DB_THREAD;
1016 1017 1018
    uint max_key_length;
    int error;

1019 1020 1021
    if (tokudb_init_flags & DB_INIT_TXN)
        open_flags += DB_AUTO_COMMIT;

1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
    /* 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;
        ref_length = hidden_primary_key = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
    } else
        key_used_on_scan = primary_key;

    /* Need some extra memory in case of packed keys */
    max_key_length = table_share->max_key_length + MAX_REF_PARTS * 3;
    if (!(alloc_ptr =
          my_multi_malloc(MYF(MY_WME),
1036 1037 1038 1039
                          &key_buff, max_key_length, 
                          &key_buff2, max_key_length, 
                          &primary_key_buff, (hidden_primary_key ? 0 : table_share->key_info[table_share->primary_key].key_length), 
                          NullS)))
1040
        TOKUDB_DBUG_RETURN(1);
1041 1042
    if (!(rec_buff = (uchar *) my_malloc((alloced_rec_buff_length = table_share->rec_buff_length), MYF(MY_WME)))) {
        my_free(alloc_ptr, MYF(0));
1043
        TOKUDB_DBUG_RETURN(1);
1044 1045 1046 1047 1048 1049
    }

    /* Init shared structure */
    if (!(share = get_share(name, table))) {
        my_free((char *) rec_buff, MYF(0));
        my_free(alloc_ptr, MYF(0));
1050
        TOKUDB_DBUG_RETURN(1);
1051
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
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
    /* Make sorted list of primary key parts, if they exist*/
    if (!hidden_primary_key) {
        uint num_prim_key_parts = table_share->key_info[table_share->primary_key].key_parts;
        primary_key_offsets = (PRIM_KEY_PART_INFO *)my_malloc(
            num_prim_key_parts*sizeof(*primary_key_offsets), 
            MYF(MY_WME)
            );
        
        if (!primary_key_offsets) {
            free_share(share, table, hidden_primary_key, 1);
            my_free((char *) rec_buff, MYF(0));
            my_free(alloc_ptr, MYF(0));
            TOKUDB_DBUG_RETURN(1);
        }
        for (uint i = 0; i < table_share->key_info[table_share->primary_key].key_parts; i++) {
            primary_key_offsets[i].offset = table_share->key_info[table_share->primary_key].key_part[i].offset;
            primary_key_offsets[i].part_index = i;
        }
        qsort(
            primary_key_offsets, // start of array
            num_prim_key_parts, //num elements
            sizeof(*primary_key_offsets), //size of each element
            primary_key_part_compare
            );
    }

1078 1079 1080 1081 1082
    thr_lock_data_init(&share->lock, &lock, NULL);
    bzero((void *) &current_row, sizeof(current_row));

    /* Fill in shared structure, if needed */
    pthread_mutex_lock(&share->mutex);
Rich Prohaska's avatar
Rich Prohaska committed
1083
    if (tokudb_debug & TOKUDB_DEBUG_OPEN)
1084 1085
        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);
1086 1087
    if (!share->use_count++) {
        DBUG_PRINT("info", ("share->use_count %u", share->use_count));
Yoni Fogel's avatar
Yoni Fogel committed
1088
        DBT cmp_byte_stream;
1089

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1090
        if ((error = db_create(&share->file, db_env, 0))) {
1091 1092 1093
            free_share(share, table, hidden_primary_key, 1);
            my_free((char *) rec_buff, MYF(0));
            my_free(alloc_ptr, MYF(0));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1094
            if (primary_key_offsets) my_free(primary_key_offsets, MYF(0));
1095
            my_errno = error;
1096
            TOKUDB_DBUG_RETURN(1);
1097 1098 1099
        }

        if (!hidden_primary_key)
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1100
            share->file->app_private = (void *) (table_share->key_info + table_share->primary_key);
Yoni Fogel's avatar
Yoni Fogel committed
1101 1102 1103
        if (tokudb_debug & TOKUDB_DEBUG_SAVE_TRACE) {
            bzero((void *) &cmp_byte_stream, sizeof(cmp_byte_stream));
            cmp_byte_stream.flags = DB_DBT_MALLOC;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1104
            if ((error = tokutrace_db_get_cmp_byte_stream(share->file, &cmp_byte_stream))) {
Yoni Fogel's avatar
Yoni Fogel committed
1105 1106 1107
                free_share(share, table, hidden_primary_key, 1);
                my_free((char *) rec_buff, MYF(0));
                my_free(alloc_ptr, MYF(0));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1108
                if (primary_key_offsets) my_free(primary_key_offsets, MYF(0));
Yoni Fogel's avatar
Yoni Fogel committed
1109 1110 1111
                my_errno = error;
                TOKUDB_DBUG_RETURN(1);
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1112
            share->file->set_bt_compare(share->file, (hidden_primary_key ? tokudb_cmp_hidden_key : tokudb_cmp_packed_key));
Yoni Fogel's avatar
Yoni Fogel committed
1113 1114 1115
            my_free(cmp_byte_stream.data, MYF(0));
        }
        else
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1116
            share->file->set_bt_compare(share->file, (hidden_primary_key ? tokudb_cmp_hidden_key : tokudb_cmp_packed_key));
Yoni Fogel's avatar
Yoni Fogel committed
1117
        
1118
        char newname[strlen(name) + 32];
1119 1120
        make_name(newname, name, "main");
        fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1121
        if ((error = share->file->open(share->file, 0, name_buff, NULL, DB_BTREE, open_flags, 0))) {
1122 1123 1124
            free_share(share, table, hidden_primary_key, 1);
            my_free((char *) rec_buff, MYF(0));
            my_free(alloc_ptr, MYF(0));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1125
            if (primary_key_offsets) my_free(primary_key_offsets, MYF(0));
1126
            my_errno = error;
1127
            TOKUDB_DBUG_RETURN(1);
1128
        }
Rich Prohaska's avatar
Rich Prohaska committed
1129
        if (tokudb_debug & TOKUDB_DEBUG_OPEN)
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1130
            TOKUDB_TRACE("open:%s:file=%p\n", newname, share->file);
1131 1132

        /* Open other keys;  These are part of the share structure */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1133 1134
        share->key_file[primary_key] = share->file;
        share->key_type[primary_key] = hidden_primary_key ? 0 : DB_NOOVERWRITE;
1135

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1136 1137
        DB **ptr = share->key_file;
        for (uint i = 0; i < table_share->keys; i++, ptr++) {
1138
            if (i != primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1139
                if ((error = open_secondary_table(ptr,&table_share->key_info[i],name,mode,&share->key_type[i]))) {
1140
                    __close(1);
1141
                    TOKUDB_DBUG_RETURN(1);
1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160
                }
            }
        }
        /* Calculate pack_length of primary key */
        share->fixed_length_primary_key = 1;
        if (!hidden_primary_key) {
            ref_length = 0;
            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->fixed_length_primary_key = (ref_length == table->key_info[primary_key].key_length);
            share->status |= STATUS_PRIMARY_KEY_INIT;
        }
        share->ref_length = ref_length;
    }
    ref_length = share->ref_length;     // If second open
    pthread_mutex_unlock(&share->mutex);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1161 1162 1163
    transaction = NULL;
    cursor = NULL;
    key_read = false;
1164
    stats.block_size = 1<<20;    // QQQ Tokudb DB block size
1165 1166
    share->fixed_length_row = !(table_share->db_create_options & HA_OPTION_PACK_RECORD);

1167
    // QQQ what happens if get_status fails
1168 1169 1170
    get_status();
    info(HA_STATUS_NO_LOCK | HA_STATUS_VARIABLE | HA_STATUS_CONST);

1171
    TOKUDB_DBUG_RETURN(0);
1172 1173
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1174 1175 1176
//
// Closes a handle to a table. 
//
1177
int ha_tokudb::close(void) {
1178
    TOKUDB_DBUG_ENTER("ha_tokudb::close %p", this);
1179
    TOKUDB_CLOSE();
1180
    TOKUDB_DBUG_RETURN(__close(0));
1181 1182 1183
}

int ha_tokudb::__close(int mutex_is_locked) {
1184
    TOKUDB_DBUG_ENTER("ha_tokudb::__close %p", this);
1185
    if (tokudb_debug & TOKUDB_DEBUG_OPEN) 
1186
        TOKUDB_TRACE("close:%p\n", this);
1187 1188
    my_free(rec_buff, MYF(MY_ALLOW_ZERO_PTR));
    my_free(alloc_ptr, MYF(MY_ALLOW_ZERO_PTR));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1189
    my_free(primary_key_offsets, MYF(MY_ALLOW_ZERO_PTR));
1190
    ha_tokudb::reset();         // current_row buffer
1191
    TOKUDB_DBUG_RETURN(free_share(share, table, hidden_primary_key, mutex_is_locked));
1192 1193
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1194 1195 1196 1197 1198 1199
//
// Reallocate record buffer (rec_buff) if needed
// If not needed, does nothing
// Parameters:
//          length - size of buffer required for rec_buff
//
1200 1201 1202 1203 1204 1205 1206 1207 1208
bool ha_tokudb::fix_rec_buff_for_blob(ulong length) {
    if (!rec_buff || length > alloced_rec_buff_length) {
        uchar *newptr;
        if (!(newptr = (uchar *) my_realloc((void *) rec_buff, length, MYF(MY_ALLOW_ZERO_PTR))))
            return 1;
        rec_buff = newptr;
        alloced_rec_buff_length = length;
    }
    return 0;
1209 1210 1211
}

/* Calculate max length needed for row */
1212 1213 1214 1215 1216
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]);
1217
        length += blob->get_length((uchar *) (buf + field_offset(blob))) + 2;
1218 1219
    }
    return length;
1220 1221 1222 1223
}

/*
*/
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235
//
// 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
//      [in]    record - row in MySQL format
//
1236

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1237
int ha_tokudb::pack_row(DBT * row, const uchar * record) {
1238
    uchar *ptr;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1239
    int r = ENOSYS;
1240
    bzero((void *) row, sizeof(*row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1241 1242 1243 1244 1245 1246 1247 1248 1249
    uint curr_skip_index;

    KEY *key_info = table->key_info + primary_key;
    my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);

    //
    // two cases, fixed length row, and variable length row
    // fixed length row is first below
    //
1250
    if (share->fixed_length_row) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291
        if (hidden_primary_key) {
            row->data = (void *)record;
            row->size = table_share->reclength;
            r = 0;
            goto cleanup;
        }
        else {
            //
            // if the primary key is not hidden, then it is part of the record
            // because primary key information is already stored in the key
            // that will be passed to the fractal tree, we do not copy
            // components that belong to the primary key
            //
            if (fix_rec_buff_for_blob(table_share->reclength)) {
                r = HA_ERR_OUT_OF_MEM;
                goto cleanup;
            }

            uchar* tmp_dest = rec_buff;
            const uchar* tmp_src = record;
            uint i = 0;
            //
            // say we have 100 bytes in record, and bytes 25-50 and 75-90 belong to the primary key
            // this for loop will do a memcpy [0,25], [51,75] and [90,100]
            //
            for (i =0; i < key_info->key_parts; i++){
                uint curr_index = primary_key_offsets[i].part_index;
                uint bytes_to_copy = record + key_info->key_part[curr_index].offset - tmp_src;
                memcpy(tmp_dest,tmp_src, bytes_to_copy);
                tmp_dest += bytes_to_copy;
                tmp_src = record + key_info->key_part[curr_index].offset + key_info->key_part[curr_index].length;
            }
            memcpy(tmp_dest,tmp_src, record + table_share->reclength - tmp_src);
            tmp_dest += record + table_share->reclength - tmp_src;

            row->data = rec_buff;
            row->size = (size_t) (tmp_dest - rec_buff);

            r = 0;
            goto cleanup;
        }
1292
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1293
    
1294
    if (table_share->blob_fields) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1295 1296 1297 1298
        if (fix_rec_buff_for_blob(max_row_length(record))) {
            r = HA_ERR_OUT_OF_MEM;
            goto cleanup;
        }
1299 1300
    }

1301
    /* Copy null bits */
1302 1303 1304
    memcpy(rec_buff, record, table_share->null_bytes);
    ptr = rec_buff + table_share->null_bytes;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1305 1306 1307 1308 1309
    //
    // assert that when the hidden primary key exists, primary_key_offsets is NULL
    //
    assert( (hidden_primary_key != 0) == (primary_key_offsets == NULL));
    curr_skip_index = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1310
    for (Field ** field = table->field; *field; field++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334
        uint curr_field_offset = field_offset(*field);
        //
        // if the primary key is hidden, primary_key_offsets will be NULL and
        // this clause will not execute
        //
        if (primary_key_offsets) {
            uint curr_skip_offset = primary_key_offsets[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 = primary_key_offsets[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
                //
                if (table->key_info[primary_key].key_part[curr_key_index].length == (*field)->field_length) {
                    continue;
                }
            }
        }
1335
        ptr = (*field)->pack(ptr, (const uchar *)
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1336
                             (record + curr_field_offset));
1337
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1338

1339 1340
    row->data = rec_buff;
    row->size = (size_t) (ptr - rec_buff);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1341 1342 1343 1344 1345 1346
    r = 0;

cleanup:
    dbug_tmp_restore_column_map(table->write_set, old_map);

    return r;
1347 1348
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1349 1350 1351 1352 1353 1354
//
// 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
1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
void ha_tokudb::unpack_row(uchar * record, DBT* row, DBT* key) {
    //
    // two cases, fixed length row, and variable length row
    // fixed length row is first below
    //
    if (share->fixed_length_row) {
        if (hidden_primary_key) {
            memcpy(record, (void *) row->data, table_share->reclength);
        }
        else {
            my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);
            KEY *key_info = table_share->key_info + primary_key;

            uchar* tmp_dest = record;
            uchar* tmp_src = (uchar *)row->data;
            uint i = 0;

            //
            // unpack_key will fill in parts of record that are part of the primary key
            //
            unpack_key(record, key, primary_key);

            //
            // produces the opposite effect to what happened in pack_row
            // first we fill in the parts of record that are not part of the key
            //
            for (i =0; i < key_info->key_parts; i++){
                uint curr_index = primary_key_offsets[i].part_index;
                uint bytes_to_copy = record + key_info->key_part[curr_index].offset - tmp_dest;
                memcpy(tmp_dest,tmp_src, bytes_to_copy);
                tmp_src += bytes_to_copy;
                tmp_dest = record + key_info->key_part[curr_index].offset + key_info->key_part[curr_index].length;
            }
            memcpy(tmp_dest,tmp_src, record + table_share->reclength - tmp_dest);
            dbug_tmp_restore_column_map(table->write_set, old_map);
        }
    }
1392 1393 1394 1395 1396 1397
    else {
        /* Copy null bits */
        my_bitmap_map *old_map = dbug_tmp_use_all_columns(table, table->write_set);
        const uchar *ptr = (const uchar *) row->data;
        memcpy(record, ptr, table_share->null_bytes);
        ptr += table_share->null_bytes;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429
        if (primary_key_offsets) {
            //
            // unpack_key will fill in parts of record that are part of the primary key
            //
            unpack_key(record, key, primary_key);
        }

        //
        // fill in parts of record that are not part of the key
        //
        uint curr_skip_index = 0;
        for (Field ** field = table->field; *field; field++) {
            uint curr_field_offset = field_offset(*field);
            if (primary_key_offsets) {
                uint curr_skip_offset = primary_key_offsets[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 = primary_key_offsets[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
                    //
                    if (table->key_info[primary_key].key_part[curr_key_index].length == (*field)->field_length) {
                        continue;
                    }
                }
            }
1430
            ptr = (*field)->unpack(record + field_offset(*field), ptr);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1431
        }
1432 1433
        dbug_tmp_restore_column_map(table->write_set, old_map);
    }
1434 1435
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1436

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1437 1438 1439 1440 1441 1442 1443 1444
//
// Store the key and the primary key into the row
// Parameters:
//      [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
//
1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461
void ha_tokudb::unpack_key(uchar * record, DBT * key, uint index) {
    KEY *key_info = table->key_info + index;
    KEY_PART_INFO *key_part = key_info->key_part, *end = key_part + key_info->key_parts;
    uchar *pos = (uchar *) key->data;

    for (; key_part != end; key_part++) {
        if (key_part->null_bit) {
            if (!*pos++) {        // Null value
                /*
                   We don't need to reset the record data as we will not access it
                   if the null data is set
                 */
                record[key_part->null_offset] |= key_part->null_bit;
                continue;
            }
            record[key_part->null_offset] &= ~key_part->null_bit;
        }
1462 1463 1464 1465 1466 1467 1468
        /* tokutek change to make pack_key and unpack_key work for
           decimals */
        uint unpack_length = key_part->length;
        if (key_part->field->type() == MYSQL_TYPE_NEWDECIMAL) {
            Field_new_decimal *field_nd = (Field_new_decimal *) key_part->field;
            unpack_length += field_nd->precision << 8;
        }
1469
        pos = (uchar *) key_part->field->unpack_key(record + field_offset(key_part->field), pos,
1470
#if MYSQL_VERSION_ID < 50123
1471
                                                    unpack_length);
1472
#else
1473
                                                    unpack_length, table->s->db_low_byte_first);
1474
#endif
1475
    }
1476 1477
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1478

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1479 1480 1481 1482 1483
//
// 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
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1484
//      [in]    key_info - holds data about the key, such as it's length and offset into record
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1485 1486 1487 1488 1489 1490 1491
//      [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
//
1492

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1493
DBT* ha_tokudb::create_dbt_key_from_key(DBT * key, KEY* key_info, uchar * buff, const uchar * record, int key_length) {
1494 1495 1496 1497 1498 1499
    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);

    key->data = buff;
    for (; key_part != end && key_length > 0; key_part++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1500 1501 1502 1503 1504 1505
        //
        // 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) {
1506
            /* Store 0 if the key part is a NULL part */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1507
            if (record[key_part->null_offset] & key_part->field->null_bit) {
1508
                *buff++ = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1509 1510 1511 1512 1513
                //
                // fractal tree does not handle this falg at the moment
                // so commenting out for now
                //
                //key->flags |= DB_DBT_DUPOK;
1514 1515 1516 1517
                continue;
            }
            *buff++ = 1;        // Store NOT NULL marker
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1518 1519 1520 1521 1522 1523
        //
        // 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
        //
        buff = key_part->field->pack_key(buff, (uchar *) (record + field_offset(key_part->field)),
1524
#if MYSQL_VERSION_ID < 50123
1525
                                         key_part->length);
1526
#else
1527
                                         key_part->length, table->s->db_low_byte_first);
1528
#endif
1529 1530 1531 1532 1533
        key_length -= key_part->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);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559
    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
//              key_length - currently set to MAX_KEY_LENGTH, is it size of buff?
// Returns:
//      the parameter key
//
DBT *ha_tokudb::create_dbt_key_from_table(DBT * key, uint keynr, uchar * buff, const uchar * record, int key_length) {
    TOKUDB_DBUG_ENTER("ha_tokudb::create_dbt_key_from_table");
    bzero((void *) key, sizeof(*key));
    if (hidden_primary_key && keynr == primary_key) {
        key->data = current_ident;
        key->size = TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH;
        DBUG_RETURN(key);
    }
    DBUG_RETURN(create_dbt_key_from_key(key, &table->key_info[keynr],buff,record,key_length));
1560 1561 1562
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574
//
// 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
//
1575
DBT *ha_tokudb::pack_key(DBT * key, uint keynr, uchar * buff, const uchar * key_ptr, uint key_length) {
1576
    TOKUDB_DBUG_ENTER("ha_tokudb::pack_key");
1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591
    KEY *key_info = table->key_info + keynr;
    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;

    for (; key_part != end && (int) key_length > 0; key_part++) {
        uint offset = 0;
        if (key_part->null_bit) {
            if (!(*buff++ = (*key_ptr == 0)))   // Store 0 if NULL
            {
                key_length -= key_part->store_length;
                key_ptr += key_part->store_length;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1592 1593 1594 1595 1596
                //
                // fractal tree does not handle this falg at the moment
                // so commenting out for now
                //
                //key->flags |= DB_DBT_DUPOK;
1597 1598 1599 1600 1601
                continue;
            }
            offset = 1;         // Data is at key_ptr+1
        }
        buff = key_part->field->pack_key_from_key_image(buff, (uchar *) key_ptr + offset,
1602
#if MYSQL_VERSION_ID < 50123
1603
                                                        key_part->length);
1604
#else
1605
                                                        key_part->length, table->s->db_low_byte_first);
1606
#endif
1607 1608 1609 1610 1611 1612 1613
        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);
1614 1615
}

1616
int ha_tokudb::read_last() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1617
    TOKUDB_DBUG_ENTER("ha_tokudb::read_last");
1618
    int do_commit = 0;
1619
    if (transaction == NULL && (tokudb_init_flags & DB_INIT_TXN)) {
1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630
        int r = db_env->txn_begin(db_env, 0, &transaction, 0);
        assert(r == 0);
        do_commit = 1;
    }
    int error = index_init(primary_key, 0);
    if (error == 0)
        error = index_last(table->record[1]);
    index_end();
    if (do_commit) {
        int r = transaction->commit(transaction, 0);
        assert(r == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1631
        transaction = NULL;
1632
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1633
    TOKUDB_DBUG_RETURN(error);
1634 1635
}

1636 1637 1638 1639
/** @brief
    Get status information that is stored in the 'status' sub database
    and the max used value for the hidden primary key.
*/
1640
void ha_tokudb::get_status() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1641
    TOKUDB_DBUG_ENTER("ha_tokudb::get_status");
1642

1643 1644
    if (!test_all_bits(share->status, (STATUS_PRIMARY_KEY_INIT | STATUS_ROW_COUNT_INIT))) {
        pthread_mutex_lock(&share->mutex);
1645 1646 1647 1648 1649
        if (!(share->status & STATUS_PRIMARY_KEY_INIT)) {
            (void) extra(HA_EXTRA_KEYREAD);
            int error = read_last();
            (void) extra(HA_EXTRA_NO_KEYREAD);
            if (error == 0) {
1650
                share->auto_ident = uint5korr(current_ident);
1651

1652 1653 1654 1655 1656 1657 1658
                // mysql may not initialize the next_number_field here
                // so we do this in the get_auto_increment method
                // index_last uses record[1]
                assert(table->next_number_field == 0);
                if (table->next_number_field) {
                    share->last_auto_increment = table->next_number_field->val_int_offset(table->s->rec_buff_length);
                    if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT) 
1659
                        TOKUDB_TRACE("init auto increment:%lld\n", share->last_auto_increment);
1660
                }
1661
            }
1662
        }
1663

1664 1665
        if (!share->status_block) {
            char name_buff[FN_REFLEN];
1666 1667 1668
            char newname[get_name_length(share->table_name) + 32];
            make_name(newname, share->table_name, "status");
            fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
1669 1670
            uint open_mode = (((table->db_stat & HA_READ_ONLY) ? DB_RDONLY : 0)
                              | DB_THREAD);
Rich Prohaska's avatar
Rich Prohaska committed
1671
            if (tokudb_debug & TOKUDB_DEBUG_OPEN)
1672
                TOKUDB_TRACE("open:%s\n", newname);
1673
            if (!db_create(&share->status_block, db_env, 0)) {
1674
                if (share->status_block->open(share->status_block, NULL, name_buff, NULL, DB_BTREE, open_mode, 0)) {
1675 1676 1677 1678 1679
                    share->status_block->close(share->status_block, 0);
                    share->status_block = 0;
                }
            }
        }
1680

1681 1682
        if (!(share->status & STATUS_ROW_COUNT_INIT) && share->status_block) {
            share->org_rows = share->rows = table_share->max_rows ? table_share->max_rows : HA_TOKUDB_MAX_ROWS;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1683
            DB_TXN *transaction = NULL;
1684 1685 1686
            int r = 0;
            if (tokudb_init_flags & DB_INIT_TXN)
                r = db_env->txn_begin(db_env, 0, &transaction, 0);
1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
            if (r == 0) {
                r = share->status_block->cursor(share->status_block, transaction, &cursor, 0);
                if (r == 0) {
                    DBT row;
                    char rec_buff[64];
                    bzero((void *) &row, sizeof(row));
                    bzero((void *) &last_key, sizeof(last_key));
                    row.data = rec_buff;
                    row.ulen = sizeof(rec_buff);
                    row.flags = DB_DBT_USERMEM;
                    if (!cursor->c_get(cursor, &last_key, &row, DB_FIRST)) {
                        uint i;
                        uchar *pos = (uchar *) row.data;
                        share->org_rows = share->rows = uint4korr(pos);
1701
                        pos += 4;
1702 1703 1704 1705
                        for (i = 0; i < table_share->keys; i++) {
                            share->rec_per_key[i] = uint4korr(pos);
                            pos += 4;
                        }
1706
                    }
1707
                    cursor->c_close(cursor);
1708
                }
1709 1710 1711
                if (transaction) {
                    r = transaction->commit(transaction, 0);
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1712
                transaction = NULL;
1713
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1714
            cursor = NULL;
1715 1716 1717 1718
        }
        share->status |= STATUS_PRIMARY_KEY_INIT | STATUS_ROW_COUNT_INIT;
        pthread_mutex_unlock(&share->mutex);
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1719
    DBUG_VOID_RETURN;
1720 1721 1722
}

static int write_status(DB * status_block, char *buff, uint length) {
1723
    TOKUDB_DBUG_ENTER("write_status");
1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734
    DBT row, key;
    int error;
    const char *key_buff = "status";

    bzero((void *) &row, sizeof(row));
    bzero((void *) &key, sizeof(key));
    row.data = buff;
    key.data = (void *) key_buff;
    key.size = sizeof(key_buff);
    row.size = length;
    error = status_block->put(status_block, 0, &key, &row, 0);
1735
    TOKUDB_DBUG_RETURN(error);
1736 1737 1738
}

static void update_status(TOKUDB_SHARE * share, TABLE * table) {
1739
    TOKUDB_DBUG_ENTER("update_status");
1740 1741 1742 1743 1744 1745 1746 1747 1748
    if (share->rows != share->org_rows || (share->status & STATUS_TOKUDB_ANALYZE)) {
        pthread_mutex_lock(&share->mutex);
        if (!share->status_block) {
            /*
               Create sub database 'status' if it doesn't exist from before
               (This '*should*' always exist for table created with MySQL)
             */

            char name_buff[FN_REFLEN];
1749 1750 1751
            char newname[get_name_length(share->table_name) + 32];
            make_name(newname, share->table_name, "status");
            fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
1752 1753 1754
            if (db_create(&share->status_block, db_env, 0))
                goto end;
            share->status_block->set_flags(share->status_block, 0);
1755
            if (share->status_block->open(share->status_block, NULL, name_buff, NULL, DB_BTREE, DB_THREAD | DB_CREATE, my_umask))
1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774
                goto end;
        }
        {
            char rec_buff[4 + MAX_KEY * 4], *pos = rec_buff;
            int4store(pos, share->rows);
            pos += 4;
            for (uint i = 0; i < table->s->keys; i++) {
                int4store(pos, share->rec_per_key[i]);
                pos += 4;
            }
            DBUG_PRINT("info", ("updating status for %s", share->table_name));
            (void) write_status(share->status_block, rec_buff, (uint) (pos - rec_buff));
            share->status &= ~STATUS_TOKUDB_ANALYZE;
            share->org_rows = share->rows;
        }
      end:
        pthread_mutex_unlock(&share->mutex);
    }
    DBUG_VOID_RETURN;
1775 1776 1777 1778 1779
}

/** @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
1780
    This is used in filesort.cc. 
1781
*/
1782
ha_rows ha_tokudb::estimate_rows_upper_bound() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1783 1784
    TOKUDB_DBUG_ENTER("ha_tokudb::estimate_rows_upper_bound");
    DBUG_RETURN(share->rows + HA_TOKUDB_EXTRA_ROWS);
1785
}
1786

1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806
int ha_tokudb::cmp_ref(const uchar * ref1, const uchar * ref2) {
    if (hidden_primary_key)
        return memcmp(ref1, ref2, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);

    int result;
    Field *field;
    KEY *key_info = table->key_info + table_share->primary_key;
    KEY_PART_INFO *key_part = key_info->key_part;
    KEY_PART_INFO *end = key_part + key_info->key_parts;

    for (; key_part != end; key_part++) {
        field = key_part->field;
        result = field->pack_cmp((const uchar *) ref1, (const uchar *) ref2, key_part->length, 0);
        if (result)
            return result;
        ref1 += field->packed_col_length((const uchar *) ref1, key_part->length);
        ref2 += field->packed_col_length((const uchar *) ref2, key_part->length);
    }

    return 0;
1807 1808
}

1809 1810 1811 1812 1813 1814
bool ha_tokudb::check_if_incompatible_data(HA_CREATE_INFO * info, uint table_changes) {
    if (table_changes < IS_EQUAL_YES)
        return COMPATIBLE_DATA_NO;
    return COMPATIBLE_DATA_YES;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1815 1816 1817 1818 1819 1820 1821 1822
//
// 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
//
1823
int ha_tokudb::write_row(uchar * record) {
1824
    TOKUDB_DBUG_ENTER("ha_tokudb::write_row");
1825 1826 1827 1828
    DBT row, prim_key, key;
    int error;

    statistic_increment(table->in_use->status_var.ha_write_count, &LOCK_status);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1829
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT) {
1830
        table->timestamp_field->set_time();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1831 1832
    }
    if (table->next_number_field && record == table->record[0]) {
1833
        update_auto_increment();
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1834 1835
    }
    if ((error = pack_row(&row, (const uchar *) record))){
1836
        TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1837 1838 1839 1840 1841
    }
    
    if (hidden_primary_key) {
        get_auto_primary_key(current_ident);
    }
1842

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1843
    u_int32_t put_flags = share->key_type[primary_key];
1844
    THD *thd = ha_thd();
1845
    if (thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS)) {
1846
        put_flags = DB_YESOVERWRITE;
1847
    }
1848

1849
    if (table_share->keys + test(hidden_primary_key) == 1) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1850
        error = share->file->put(share->file, transaction, create_dbt_key_from_table(&prim_key, primary_key, key_buff, record), &row, put_flags);
1851 1852 1853
        last_dup_key = primary_key;
    } else {
        DB_TXN *sub_trans = transaction;
1854
        /* QQQ Don't use sub transactions in temporary tables */
1855 1856
        for (uint retry = 0; retry < tokudb_trans_retry; retry++) {
            key_map changed_keys(0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1857
            if (!(error = share->file->put(share->file, sub_trans, create_dbt_key_from_table(&prim_key, primary_key, key_buff, record), &row, put_flags))) {
1858 1859 1860 1861
                changed_keys.set_bit(primary_key);
                for (uint keynr = 0; keynr < table_share->keys; keynr++) {
                    if (keynr == primary_key)
                        continue;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1862
                    put_flags = share->key_type[keynr];
1863 1864
                    if (put_flags == DB_NOOVERWRITE && thd_test_options(thd, OPTION_RELAXED_UNIQUE_CHECKS))
                        put_flags = DB_YESOVERWRITE;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1865
                    if ((error = share->key_file[keynr]->put(share->key_file[keynr], sub_trans, create_dbt_key_from_table(&key, keynr, key_buff2, record), &prim_key, put_flags))) {
1866 1867 1868 1869 1870
                        last_dup_key = keynr;
                        break;
                    }
                    changed_keys.set_bit(keynr);
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1871 1872
            } 
            else {
1873
                last_dup_key = primary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1874
            }
1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894
            if (error) {
                /* Remove inserted row */
                DBUG_PRINT("error", ("Got error %d", error));
                if (using_ignore) {
                    int new_error = 0;
                    if (!changed_keys.is_clear_all()) {
                        new_error = 0;
                        for (uint keynr = 0; keynr < table_share->keys + test(hidden_primary_key); keynr++) {
                            if (changed_keys.is_set(keynr)) {
                                if ((new_error = remove_key(sub_trans, keynr, record, &prim_key)))
                                    break;
                            }
                        }
                    }
                    if (new_error) {
                        error = new_error;      // This shouldn't happen
                        break;
                    }
                }
            }
1895
            if (error != DB_LOCK_DEADLOCK && error != DB_LOCK_NOTGRANTED)
1896 1897
                break;
        }
1898
    }
1899 1900 1901 1902
    if (error == DB_KEYEXIST)
        error = HA_ERR_FOUND_DUPP_KEY;
    else if (!error)
        changed_rows++;
1903
    TOKUDB_DBUG_RETURN(error);
1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923
}

/* 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;
        }
1924
    }
1925
    return 0;
1926 1927 1928 1929 1930 1931 1932
}


/*
  Update a row from one value to another.
  Clobbers key_buff2
*/
1933
int ha_tokudb::update_primary_key(DB_TXN * trans, bool primary_key_changed, const uchar * old_row, DBT * old_key, const uchar * new_row, DBT * new_key, bool local_using_ignore) {
1934
    TOKUDB_DBUG_ENTER("update_primary_key");
1935 1936 1937 1938 1939 1940 1941
    DBT row;
    int error;

    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
        if (!(error = remove_key(trans, primary_key, old_row, old_key))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1942
            if (!(error = pack_row(&row, new_row))) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1943
                if ((error = share->file->put(share->file, trans, new_key, &row, share->key_type[primary_key]))) {
1944 1945 1946 1947
                    // Probably a duplicated key; restore old key and row if needed
                    last_dup_key = primary_key;
                    if (local_using_ignore) {
                        int new_error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1948
                        if ((new_error = pack_row(&row, old_row)) || (new_error = share->file->put(share->file, trans, old_key, &row, share->key_type[primary_key])))
1949 1950 1951 1952 1953 1954 1955
                            error = new_error;  // fatal error
                    }
                }
            }
        }
    } else {
        // Primary key didn't change;  just update the row data
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1956
        if (!(error = pack_row(&row, new_row)))
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1957
            error = share->file->put(share->file, trans, new_key, &row, 0);
1958
    }
1959
    TOKUDB_DBUG_RETURN(error);
1960 1961 1962 1963 1964 1965 1966
}

/*
  Restore changed keys, when a non-fatal error aborts the insert/update
  of one row.
  Clobbers keybuff2
*/
1967
int ha_tokudb::restore_keys(DB_TXN * trans, key_map * changed_keys, uint primary_key, const uchar * old_row, DBT * old_key, const uchar * new_row, DBT * new_key) {
1968
    TOKUDB_DBUG_ENTER("restore_keys");
1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987
    int error;
    DBT tmp_key;
    uint keynr;

    /* Restore the old primary key, and the old row, but don't ignore
       duplicate key failure */
    if ((error = update_primary_key(trans, TRUE, new_row, new_key, old_row, old_key, FALSE)))
        goto err;

    /* Remove the new key, and put back the old key
       changed_keys is a map of all non-primary keys that need to be
       rolled back.  The last key set in changed_keys is the one that
       triggered the duplicate key error (it wasn't inserted), so for
       that one just put back the old value. */
    if (!changed_keys->is_clear_all()) {
        for (keynr = 0; keynr < table_share->keys + test(hidden_primary_key); keynr++) {
            if (changed_keys->is_set(keynr)) {
                if (changed_keys->is_prefix(1) && (error = remove_key(trans, keynr, new_row, new_key)))
                    break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
1988
                if ((error = share->key_file[keynr]->put(share->key_file[keynr], trans, create_dbt_key_from_table(&tmp_key, keynr, key_buff2, old_row), old_key, share->key_type[keynr])))
1989 1990 1991
                    break;
            }
        }
1992
    }
1993 1994 1995

  err:
    DBUG_ASSERT(error != DB_KEYEXIST);
1996
    TOKUDB_DBUG_RETURN(error);
1997 1998
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
1999 2000 2001 2002 2003 2004 2005 2006 2007
//
// 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
//
2008
int ha_tokudb::update_row(const uchar * old_row, uchar * new_row) {
2009
    TOKUDB_DBUG_ENTER("update_row");
2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026
    DBT prim_key, key, old_prim_key;
    int error;
    DB_TXN *sub_trans;
    bool primary_key_changed;

    LINT_INIT(error);
    statistic_increment(table->in_use->status_var.ha_update_count, &LOCK_status);
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_UPDATE)
        table->timestamp_field->set_time();

    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;
    } else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2027
        create_dbt_key_from_table(&prim_key, primary_key, key_buff, new_row);
2028 2029

        if ((primary_key_changed = key_cmp(primary_key, old_row, new_row)))
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2030
            create_dbt_key_from_table(&old_prim_key, primary_key, primary_key_buff, old_row);
2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045
        else
            old_prim_key = prim_key;
    }

    sub_trans = transaction;
    for (uint retry = 0; retry < tokudb_trans_retry; retry++) {
        key_map changed_keys(0);
        /* Start by updating the primary key */
        if (!(error = update_primary_key(sub_trans, primary_key_changed, old_row, &old_prim_key, new_row, &prim_key, using_ignore))) {
            // Update all other keys
            for (uint keynr = 0; keynr < table_share->keys; keynr++) {
                if (keynr == primary_key)
                    continue;
                if (key_cmp(keynr, old_row, new_row) || primary_key_changed) {
                    if ((error = remove_key(sub_trans, keynr, old_row, &old_prim_key))) {
2046
                        TOKUDB_DBUG_RETURN(error);     // Fatal error
2047 2048
                    }
                    changed_keys.set_bit(keynr);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2049
                    if ((error = share->key_file[keynr]->put(share->key_file[keynr], sub_trans, create_dbt_key_from_table(&key, keynr, key_buff2, new_row), &prim_key, share->key_type[keynr]))) {
2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069
                        last_dup_key = keynr;
                        break;
                    }
                }
            }
        }
        if (error) {
            /* Remove inserted row */
            DBUG_PRINT("error", ("Got error %d", error));
            if (using_ignore) {
                int new_error = 0;
                if (!changed_keys.is_clear_all())
                    new_error = restore_keys(transaction, &changed_keys, primary_key, old_row, &old_prim_key, new_row, &prim_key);
                if (new_error) {
                    /* This shouldn't happen */
                    error = new_error;
                    break;
                }
            }
        }
2070
        if (error != DB_LOCK_DEADLOCK && error != DB_LOCK_NOTGRANTED)
2071 2072 2073 2074
            break;
    }
    if (error == DB_KEYEXIST)
        error = HA_ERR_FOUND_DUPP_KEY;
2075
    TOKUDB_DBUG_RETURN(error);
2076 2077
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091
//
//
// 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
//
2092
int ha_tokudb::remove_key(DB_TXN * trans, uint keynr, const uchar * record, DBT * prim_key) {
2093
    TOKUDB_DBUG_ENTER("ha_tokudb::remove_key");
2094 2095 2096
    int error;
    DBT key;
    DBUG_PRINT("enter", ("index: %d", keynr));
2097 2098
    DBUG_PRINT("primary", ("index: %d", primary_key));
    DBUG_DUMP("prim_key", (uchar *) prim_key->data, prim_key->size);
2099 2100 2101 2102

    if (keynr == active_index && cursor)
        error = cursor->c_del(cursor, 0);
    else if (keynr == primary_key || ((table->key_info[keynr].flags & (HA_NOSAME | HA_NULL_PART_KEY)) == HA_NOSAME)) {  // Unique key
2103
        DBUG_PRINT("Unique key", ("index: %d", keynr));
2104
        DBUG_ASSERT(keynr == primary_key || prim_key->data != key_buff2);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2105
        error = share->key_file[keynr]->del(share->key_file[keynr], trans, keynr == primary_key ? prim_key : create_dbt_key_from_table(&key, keynr, key_buff2, record), 0);
2106
    } else {
2107
        /* QQQ use toku_db_delboth(key_file[keynr], key, val, trans);
2108 2109 2110 2111 2112 2113
           To delete the not duplicated key, we need to open an cursor on the
           row to find the key to be delete and delete it.
           We will never come here with keynr = primary_key
         */
        DBUG_ASSERT(keynr != primary_key && prim_key->data != key_buff2);
        DBC *tmp_cursor;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2114 2115
        if (!(error = share->key_file[keynr]->cursor(share->key_file[keynr], trans, &tmp_cursor, 0))) {
            if (!(error = tmp_cursor->c_get(tmp_cursor, create_dbt_key_from_table(&key, keynr, key_buff2, record), prim_key, DB_GET_BOTH))) { 
2116
                DBUG_DUMP("cget key", (uchar *) key.data, key.size);
2117 2118 2119 2120 2121 2122 2123
                error = tmp_cursor->c_del(tmp_cursor, 0);
            }
            int result = tmp_cursor->c_close(tmp_cursor);
            if (!error)
                error = result;
        }
    }
2124
    TOKUDB_DBUG_RETURN(error);
2125 2126
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138
//
// 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
2139
int ha_tokudb::remove_keys(DB_TXN * trans, const uchar * record, DBT * prim_key, key_map * keys) {
2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152
    int result = 0;
    for (uint keynr = 0; keynr < table_share->keys + test(hidden_primary_key); keynr++) {
        if (keys->is_set(keynr)) {
            int new_error = remove_key(trans, keynr, record, prim_key);
            if (new_error) {
                result = new_error;     // Return last error
                break;          // Let rollback correct things
            }
        }
    }
    return result;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2153
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2154
// Deletes a row in the table, called when handling a DELETE query
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2155 2156 2157 2158 2159 2160
// Parameters:
//      [in]    record - row to be deleted, in MySQL format
// Returns:
//      0 on success
//      error otherwise
//
2161
int ha_tokudb::delete_row(const uchar * record) {
2162
    TOKUDB_DBUG_ENTER("ha_tokudb::delete_row");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2163 2164
    int error = ENOSYS;
    DBT prim_key;
2165 2166 2167
    key_map keys = table_share->keys_in_use;
    statistic_increment(table->in_use->status_var.ha_delete_count, &LOCK_status);

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2168
    create_dbt_key_from_table(&prim_key, primary_key, key_buff, record);
2169 2170 2171 2172 2173 2174 2175
    if (hidden_primary_key)
        keys.set_bit(primary_key);

    /* Subtransactions may be used in order to retry the delete in
       case we get a DB_LOCK_DEADLOCK error. */
    DB_TXN *sub_trans = transaction;
    for (uint retry = 0; retry < tokudb_trans_retry; retry++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2176
        error = remove_keys(sub_trans, record, &prim_key, &keys);
2177 2178 2179 2180
        if (error) {
            DBUG_PRINT("error", ("Got error %d", error));
            break;              // No retry - return error
        }
2181
        if (error != DB_LOCK_DEADLOCK && error != DB_LOCK_NOTGRANTED)
2182
            break;
2183 2184
    }
#ifdef CANT_COUNT_DELETED_ROWS
2185 2186
    if (!error)
        changed_rows--;
2187
#endif
2188
    TOKUDB_DBUG_RETURN(error);
2189 2190
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2191 2192 2193 2194 2195 2196 2197 2198 2199
//
// 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
//
2200
int ha_tokudb::index_init(uint keynr, bool sorted) {
2201
    TOKUDB_DBUG_ENTER("ha_tokudb::index_init %p %d", this, keynr);
2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
    int error;
    DBUG_PRINT("enter", ("table: '%s'  key: %d", table_share->table_name.str, keynr));

    /*
       Under some very rare conditions (like full joins) we may already have
       an active cursor at this point
     */
    if (cursor) {
        DBUG_PRINT("note", ("Closing active cursor"));
        cursor->c_close(cursor);
    }
    active_index = keynr;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2214
    last_cursor_error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2215
    range_lock_grabbed = false;
2216
    DBUG_ASSERT(keynr <= table->s->keys);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2217
    DBUG_ASSERT(share->key_file[keynr]);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2218 2219
    if ((error = share->key_file[keynr]->cursor(share->key_file[keynr], transaction, &cursor, table->reginfo.lock_type > TL_WRITE_ALLOW_READ ? 0 : 0))) {
        last_cursor_error = error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2220
        cursor = NULL;             // Safety
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2221
    }
2222
    bzero((void *) &last_key, sizeof(last_key));
2223
    TOKUDB_DBUG_RETURN(error);
2224 2225
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2226 2227 2228
//
// closes the local cursor
//
2229
int ha_tokudb::index_end() {
2230
    TOKUDB_DBUG_ENTER("ha_tokudb::index_end %p", this);
2231
    int error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2232
    range_lock_grabbed = false;
2233 2234 2235
    if (cursor) {
        DBUG_PRINT("enter", ("table: '%s'", table_share->table_name.str));
        error = cursor->c_close(cursor);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2236
        cursor = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2237
        last_cursor_error = 0;
2238
    }
2239
    active_index = MAX_KEY;
2240
    TOKUDB_DBUG_RETURN(error);
2241 2242
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258
//
// The funtion read_row checks whether the row was obtained from the primary table or 
// from an index table. If it was obtained from an index table, it further dereferences on
// the main table. In the end, the read_row function will manage to return the actual row
// of interest in the buf parameter.
//
// Parameters:
//              error - result of preceding DB call
//      [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
//              read_next - if true, DB_NOTFOUND and DB_KEYEMPTY map to HA_ERR_END_OF_FILE, 
//                  else HA_ERR_KEY_NOT_FOUND, this is a bad parameter to have and this funcitonality
//                  should not be here
//
2259
int ha_tokudb::read_row(int error, uchar * buf, uint keynr, DBT * row, DBT * found_key, bool read_next) {
2260
    TOKUDB_DBUG_ENTER("ha_tokudb::read_row");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2261 2262 2263
    //
    // Disreputable error translation: this makes us all puke
    //
2264
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2265
        last_cursor_error = error;
2266
        table->status = STATUS_NOT_FOUND;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2267 2268 2269 2270 2271 2272 2273 2274
        cursor->c_close(cursor);
        cursor = NULL;
        if (error == DB_NOTFOUND || error == DB_KEYEMPTY) {
            error = read_next ? HA_ERR_END_OF_FILE : HA_ERR_KEY_NOT_FOUND;
            if ((share->key_file[keynr]->cursor(share->key_file[keynr], transaction, &cursor, table->reginfo.lock_type > TL_WRITE_ALLOW_READ ? 0 : 0))) {
                cursor = NULL;             // Safety
            }
        }
2275
        TOKUDB_DBUG_RETURN(error);
2276
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2277
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2278
    // extract hidden primary key to current_ident
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2279 2280
    //
    if (hidden_primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2281 2282 2283 2284 2285 2286
        if (keynr == primary_key) {
            memcpy_fixed(current_ident, (char *) found_key->data, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
        }
        else {
            memcpy_fixed(current_ident, (char *) row->data, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2287
    }
2288
    table->status = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2289 2290 2291 2292 2293 2294 2295 2296
    //
    // if the index shows that the table we read the row from was indexed on the primary key,
    // that means we have our row and can skip
    // this entire if clause. All that is required is to unpack row.
    // if the index shows that what we read was from a table that was NOT indexed on the 
    // primary key, then we must still retrieve the row, as the "row" value is indeed just
    // a primary key, whose row we must still read
    //
2297 2298
    if (keynr != primary_key) {
        if (key_read && found_key) {
2299
            // TOKUDB_DBUG_DUMP("key=", found_key->data, found_key->size);
2300
            unpack_key(buf, found_key, keynr);
2301
            if (!hidden_primary_key) {
2302
                // TOKUDB_DBUG_DUMP("row=", row->data, row->size);
2303
                unpack_key(buf, row, primary_key);
2304
            }
2305
            TOKUDB_DBUG_RETURN(0);
2306
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2307 2308 2309
        //
        // create a DBT that has the same data as row,
        //
2310 2311 2312 2313 2314
        DBT key;
        bzero((void *) &key, sizeof(key));
        key.data = key_buff;
        key.size = row->size;
        memcpy(key_buff, row->data, row->size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2315 2316 2317
        //
        // Read the data into current_row
        //
2318
        current_row.flags = DB_DBT_REALLOC;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2319
        if ((error = share->file->get(share->file, transaction, &key, &current_row, 0))) {
2320
            table->status = STATUS_NOT_FOUND;
2321
            TOKUDB_DBUG_RETURN(error == DB_NOTFOUND ? HA_ERR_CRASHED : error);
2322
        }
2323 2324
        // TOKUDB_DBUG_DUMP("key=", key.data, key.size);
        // TOKUDB_DBUG_DUMP("row=", row->data, row->size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2325 2326 2327 2328
        unpack_row(buf, &current_row, &key);
    }
    else {
        unpack_row(buf, row, found_key);
2329
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2330
    if (found_key) { DBUG_DUMP("read row key", (uchar *) found_key->data, found_key->size); }
2331
    TOKUDB_DBUG_RETURN(0);
2332 2333
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350
//
// This is only used to read whole keys
// 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
//                   keynr - index to use
//      [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), 
//      error otherwise
//
2351
int ha_tokudb::index_read_idx(uchar * buf, uint keynr, const uchar * key, uint key_len, enum ha_rkey_function find_flag) {
2352
    TOKUDB_DBUG_ENTER("ha_tokudb::index_read_idx");
2353 2354 2355
    table->in_use->status_var.ha_read_key_count++;
    current_row.flags = DB_DBT_REALLOC;
    active_index = MAX_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2356
    TOKUDB_DBUG_RETURN(read_row(share->key_file[keynr]->get(share->key_file[keynr], transaction, pack_key(&last_key, keynr, key_buff, key, key_len), &current_row, 0), buf, keynr, &current_row, &last_key, 0));
2357 2358
}

Yoni Fogel's avatar
Yoni Fogel committed
2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420
//TODO: QQQ Function to tell if a key+keylen is the entire key (loop through the schema), see comparison function for ideas.
/*
if (full_key) {
    switch (find_flag) {
        case (HA_READ_PREFIX): //Synonym for HA_READ_KEY_EXACT
        case (HA_READ_KEY_EXACT):
            Just c_get DB_SET, return.
        case (HA_READ_AFTER_KEY):
            c_get DB_SET_RANGE.  If EQUAL to query, then do DB_NEXT (or is it DB_NEXT_NODUP?)
        case (HA_READ_KEY_OR_NEXT):
            c_get DB_SET_RANGE
        case (HA_READ_BEFORE_KEY):
            c_get DB_SET_RANGE, then do DB_PREV (or is it DB_PREV_NODUP)?
        case (HA_READ_KEY_OR_PREV):
            c_get DB_SET_RANGE.  If NOT EQUAL to query, then do DB_PREV (or is it DB_PREV_NODUP?)
        case (HA_READ_PREFIX_LAST_OR_PREV):
            c_get DB_SET_RANGE.  If NOT EQUAL to query, then do DB_PREV (or is it DB_PREV_NODUP?)
            if if WAS equal to the query, then
                if (NO_DUP db, just return it) else:
                do DB_NEXT_NODUP
                if found, do DB_PREV and return
                else      do DB_LAST and return
        case (HA_READ_PREFIX_LAST):
            c_get DB_SET.  if !found, return NOT FOUND
            if (NO_DUP db, just return it) else:
                do c_get DB_NEXT_NODUP
                    if found, do DB_PREV and return.
                    else      do DB_LAST and return.
        default: Crash a lot.
     }
else {
// Not full key
    switch (find_flag) {
        case (HA_READ_PREFIX): //Synonym for HA_READ_KEY_EXACT
        case (HA_READ_KEY_EXACT):
            c_get DB_SET_RANGE, then check a prefix
        case (HA_READ_AFTER_KEY):
            c_get DB_SET_RANGE, then:
                while (found && query is prefix of 'dbtfound') do:
                    c_get DB_NEXT_NODUP (Definitely NEXT_NODUP since we care about key only).
        case (HA_READ_KEY_OR_NEXT):
            c_get SET_RANGE
        case (HA_READ_BEFORE_KEY):
            c_get DB_SET_RANGE, then do DB_PREV (or is it DB_PREV_NODUP)?
        case (HA_READ_KEY_OR_PREV):
            c_get DB_SET_RANGE.  If query not a prefix of found, then DB_PREV (or is it DB_PREV_NODUP?)
        case (HA_READ_PREFIX_LAST_OR_PREV):
            c_get DB_SET_RANGE, then:
                if (found && query is prefix of whatever found) do:
                    c_get DB_NEXT till not prefix (and return the one that was)
                if (found originally but was not prefix of whatever found) do:
                    c_get DB_PREV
        case (HA_READ_PREFIX_LAST):
            c_get DB_SET_RANGE.  if !found, or query not prefix of what found, return NOT FOUND
            whlie query is prefix of whatfound, do c_get DB_NEXT till not.. then return the last one that was.
        default: Crash a lot.
     }
}
Note that sometimes if not found, will need things like DB_FIRST or DB_LAST
TODO: QQQ maybe need to pass true/1 as last parameter of read_row (this would make it
return END_OF_FILE instead of just NOT_FOUND
*/
2421

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438
//
// 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
//
2439 2440
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);
2441
    // TOKUDB_DBUG_DUMP("key=", key, key_len);
2442 2443 2444
    DBT row;
    int error;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2445 2446
    CHECK_VALID_CURSOR();

2447 2448 2449 2450
    table->in_use->status_var.ha_read_key_count++;
    bzero((void *) &row, sizeof(row));
    pack_key(&last_key, active_index, key_buff, key, key_len);

2451
    switch (find_flag) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2452
    case HA_READ_KEY_EXACT: /* Find first record else error */
2453 2454 2455 2456 2457 2458 2459
        error = cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE);
        if (error == 0) {
            DBT orig_key;
            pack_key(&orig_key, active_index, key_buff2, key, key_len);
            if (tokudb_prefix_cmp_packed_key(share->key_file[active_index], &orig_key, &last_key))
                error = DB_NOTFOUND;
        }
2460
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2461
    case HA_READ_AFTER_KEY: /* Find next rec. after key-record */
2462 2463 2464
        error = cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE);
        if (error == 0) {
            DBT orig_key;
Yoni Fogel's avatar
Yoni Fogel committed
2465
            pack_key(&orig_key, active_index, key_buff2, key, key_len);
2466
            for (;;) {
2467
                if (tokudb_prefix_cmp_packed_key(share->key_file[active_index], &orig_key, &last_key) != 0)
2468
                    break;
Yoni Fogel's avatar
Yoni Fogel committed
2469
                error = cursor->c_get(cursor, &last_key, &row, DB_NEXT_NODUP);
2470 2471 2472
                if (error != 0)
                    break;
            }
2473
        }
2474
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2475
    case HA_READ_BEFORE_KEY: /* Find next rec. before key-record */
2476 2477 2478
        error = cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE);
        if (error == 0)
            error = cursor->c_get(cursor, &last_key, &row, DB_PREV);
Yoni Fogel's avatar
Yoni Fogel committed
2479
        else if (error == DB_NOTFOUND)
2480 2481
            error = cursor->c_get(cursor, &last_key, &row, DB_LAST);
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2482
    case HA_READ_KEY_OR_NEXT: /* Record or next record */
2483 2484
        error = cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE);
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2485
    case HA_READ_KEY_OR_PREV: /* Record or previous */
2486
        error = cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE);
2487 2488 2489
        if (error == 0) {
            DBT orig_key; 
            pack_key(&orig_key, active_index, key_buff2, key, key_len);
2490
            if (tokudb_prefix_cmp_packed_key(share->key_file[active_index], &orig_key, &last_key) != 0)
2491
                error = cursor->c_get(cursor, &last_key, &row, DB_PREV);
Yoni Fogel's avatar
Yoni Fogel committed
2492 2493
        }
        else if (error == DB_NOTFOUND)
2494
            error = cursor->c_get(cursor, &last_key, &row, DB_LAST);
2495
        break;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2496
    case HA_READ_PREFIX_LAST_OR_PREV: /* Last or prev key with the same prefix */
2497 2498 2499
        error = cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE);
        if (error == 0) {
            DBT orig_key;
Yoni Fogel's avatar
Yoni Fogel committed
2500
            pack_key(&orig_key, active_index, key_buff2, key, key_len);
2501
            for (;;) {
2502
                if (tokudb_prefix_cmp_packed_key(share->key_file[active_index], &orig_key, &last_key) != 0)
2503
                    break;
Yoni Fogel's avatar
Yoni Fogel committed
2504
                error = cursor->c_get(cursor, &last_key, &row, DB_NEXT_NODUP);
2505 2506 2507 2508 2509
                if (error != 0)
                    break;
            }
            if (error == 0)
                error = cursor->c_get(cursor, &last_key, &row, DB_PREV);
Yoni Fogel's avatar
Yoni Fogel committed
2510 2511
            else if (error == DB_NOTFOUND)
                error = cursor->c_get(cursor, &last_key, &row, DB_LAST);
2512
        }
Yoni Fogel's avatar
Yoni Fogel committed
2513 2514
        else if (error == DB_NOTFOUND)
            error = cursor->c_get(cursor, &last_key, &row, DB_LAST);
2515 2516
        break;
    default:
2517
        TOKUDB_TRACE("unsupported:%d\n", find_flag);
2518
        error = HA_ERR_UNSUPPORTED;
2519
        break;
2520
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2521
    error = read_row(error, buf, active_index, &row, &last_key, 0);
2522
    if (error && (tokudb_debug & TOKUDB_DEBUG_ERROR))
2523
        TOKUDB_TRACE("error:%d:%d\n", error, find_flag);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2524
cleanup:
2525
    TOKUDB_DBUG_RETURN(error);
2526 2527
}

2528
#if 0
2529 2530 2531 2532
/*
  Read last key is solved by reading the next key and then reading
  the previous key
*/
2533
int ha_tokudb::index_read_last(uchar * buf, const uchar * key, uint key_len) {
2534
    TOKUDB_DBUG_ENTER("ha_tokudb::index_read_last");
2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545
    DBT row;
    int error;
    KEY *key_info = &table->key_info[active_index];

    statistic_increment(table->in_use->status_var.ha_read_key_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));

    /* read of partial key */
    pack_key(&last_key, active_index, key_buff, key, key_len);
    /* Store for compare */
    memcpy(key_buff2, key_buff, (key_len = last_key.size));
2546
    assert(0);
2547 2548 2549 2550 2551 2552
    key_info->handler.bdb_return_if_eq = 1;
    error = read_row(cursor->c_get(cursor, &last_key, &row, DB_SET_RANGE), buf, active_index, &row, (DBT *) 0, 0);
    key_info->handler.bdb_return_if_eq = 0;
    bzero((void *) &row, sizeof(row));
    if (read_row(cursor->c_get(cursor, &last_key, &row, DB_PREV), buf, active_index, &row, &last_key, 1) || tokudb_key_cmp(table, key_info, key_buff2, key_len))
        error = HA_ERR_KEY_NOT_FOUND;
2553
    TOKUDB_DBUG_RETURN(error);
2554
}
2555
#endif
2556

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2557 2558 2559 2560 2561 2562 2563 2564 2565
//
// 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
//
2566
int ha_tokudb::index_next(uchar * buf) {
2567
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2568
    int error; 
2569
    DBT row;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2570
    u_int32_t flags = range_lock_grabbed ? (DB_NEXT | DB_PRELOCKED) : DB_NEXT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2571 2572
    CHECK_VALID_CURSOR();
    
2573 2574
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2575
    error = read_row(cursor->c_get(cursor, &last_key, &row, flags), buf, active_index, &row, &last_key, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2576
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2577
    TOKUDB_DBUG_RETURN(error);
2578 2579
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590
//
// 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
//
2591
int ha_tokudb::index_next_same(uchar * buf, const uchar * key, uint keylen) {
2592
    TOKUDB_DBUG_ENTER("ha_tokudb::index_next_same %p", this);
2593 2594
    DBT row;
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2595 2596
    CHECK_VALID_CURSOR();
    
2597 2598
    statistic_increment(table->in_use->status_var.ha_read_next_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));
Rich Prohaska's avatar
Rich Prohaska committed
2599 2600 2601 2602 2603
    /* QQQ NEXT_DUP on nodup returns EINVAL for tokudb */
    if (keylen == table->key_info[active_index].key_length && 
        !(table->key_info[active_index].flags & HA_NOSAME) && 
        !(table->key_info[active_index].flags & HA_END_SPACE_KEY)) {

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2604
        u_int32_t flags = range_lock_grabbed ? (DB_NEXT_DUP | DB_PRELOCKED) : DB_NEXT_DUP;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2605
        error = cursor->c_get(cursor, &last_key, &row, flags);
2606 2607
        error = read_row(error, buf, active_index, &row, &last_key, 1);
    } else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2608
        u_int32_t flags = range_lock_grabbed ? (DB_NEXT | DB_PRELOCKED) : DB_NEXT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2609
        error = read_row(cursor->c_get(cursor, &last_key, &row, flags), buf, active_index, &row, &last_key, 1);
2610 2611 2612
        if (!error &&::key_cmp_if_same(table, key, active_index, keylen))
            error = HA_ERR_END_OF_FILE;
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2613
cleanup:
2614
    TOKUDB_DBUG_RETURN(error);
2615 2616
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2617 2618 2619 2620 2621 2622 2623 2624 2625
//
// 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
//
2626
int ha_tokudb::index_prev(uchar * buf) {
2627
    TOKUDB_DBUG_ENTER("ha_tokudb::index_prev");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2628
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2629
    u_int32_t flags = range_lock_grabbed ? (DB_PREV | DB_PRELOCKED) : DB_PREV;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2630
    CHECK_VALID_CURSOR();
2631 2632 2633
    DBT row;
    statistic_increment(table->in_use->status_var.ha_read_prev_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2634
    error = read_row(cursor->c_get(cursor, &last_key, &row, flags), buf, active_index, &row, &last_key, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2635
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2636
    TOKUDB_DBUG_RETURN(error);
2637 2638
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2639 2640 2641 2642 2643 2644 2645 2646 2647
//
// 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
//
2648
int ha_tokudb::index_first(uchar * buf) {
2649
    TOKUDB_DBUG_ENTER("ha_tokudb::index_first");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2650
    int error;
2651
    DBT row;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2652
    CHECK_VALID_CURSOR();
2653 2654
    statistic_increment(table->in_use->status_var.ha_read_first_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2655
    error = read_row(cursor->c_get(cursor, &last_key, &row, DB_FIRST), buf, active_index, &row, &last_key, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2656
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2657
    TOKUDB_DBUG_RETURN(error);
2658 2659
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2660 2661 2662 2663 2664 2665 2666 2667 2668
//
// 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
//
2669
int ha_tokudb::index_last(uchar * buf) {
2670
    TOKUDB_DBUG_ENTER("ha_tokudb::index_last");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2671
    int error;
2672
    DBT row;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2673
    CHECK_VALID_CURSOR();
2674 2675
    statistic_increment(table->in_use->status_var.ha_read_last_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2676
    error = read_row(cursor->c_get(cursor, &last_key, &row, DB_LAST), buf, active_index, &row, &last_key, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2677
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2678
    TOKUDB_DBUG_RETURN(error);
2679 2680
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2681 2682 2683 2684 2685 2686 2687 2688
//
// 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
//
2689
int ha_tokudb::rnd_init(bool scan) {
2690
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_init");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2691
    int error;
2692
    current_row.flags = DB_DBT_REALLOC;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2693
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2694 2695 2696
    if (scan) {
        DB* db = share->key_file[primary_key];
        error = db->pre_acquire_read_lock(db, transaction, db->dbt_neg_infty(), NULL, db->dbt_pos_infty(), NULL);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2697
        if (error) { last_cursor_error = error; goto cleanup; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2698 2699
    }
    error = index_init(primary_key, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2700 2701 2702
    if (error) { goto cleanup;}

    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2703
    // only want to set range_lock_grabbed to true after index_init
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2704 2705 2706 2707
    // 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
2708
    if (scan) { range_lock_grabbed = true; }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2709
    error = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2710 2711
cleanup:
    TOKUDB_DBUG_RETURN(error);
2712 2713
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2714 2715 2716
//
// End a scan of the table
//
2717
int ha_tokudb::rnd_end() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2718
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_end");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2719
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2720
    TOKUDB_DBUG_RETURN(index_end());
2721 2722
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2723 2724 2725 2726 2727 2728 2729 2730 2731
//
// 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
//
2732
int ha_tokudb::rnd_next(uchar * buf) {
2733
    TOKUDB_DBUG_ENTER("ha_tokudb::ha_tokudb::rnd_next");
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2734
    int error;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2735
    DBT row;    
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2736
    u_int32_t flags = range_lock_grabbed ? (DB_NEXT | DB_PRELOCKED) : DB_NEXT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2737

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2738
    CHECK_VALID_CURSOR()
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2739 2740 2741 2742
    //
    // The reason we do not just call index_next is that index_next 
    // increments a different variable than we do here
    //
2743 2744
    statistic_increment(table->in_use->status_var.ha_read_rnd_next_count, &LOCK_status);
    bzero((void *) &row, sizeof(row));
2745
    DBUG_DUMP("last_key", (uchar *) last_key.data, last_key.size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2746
    error = read_row(cursor->c_get(cursor, &last_key, &row, flags), buf, primary_key, &row, &last_key, 1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2747
cleanup:
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2748
    TOKUDB_DBUG_RETURN(error);
2749 2750 2751 2752
}


DBT *ha_tokudb::get_pos(DBT * to, uchar * pos) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2753
    TOKUDB_DBUG_ENTER("ha_tokudb::get_pos");
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768
    /* We don't need to set app_data here */
    bzero((void *) to, sizeof(*to));

    to->data = pos;
    if (share->fixed_length_primary_key)
        to->size = ref_length;
    else {
        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++)
            pos += key_part->field->packed_col_length(pos, key_part->length);
        to->size = (uint) (pos - (uchar *) to->data);
    }
    DBUG_DUMP("key", (const uchar *) to->data, to->size);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2769
    DBUG_RETURN(to);
2770 2771
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2772
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2773
// Retrieves a row with based on the primary key saved in pos
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2774 2775 2776 2777 2778
// Returns:
//      0 on success
//      HA_ERR_KEY_NOT_FOUND if not found
//      error otherwise
//
2779
int ha_tokudb::rnd_pos(uchar * buf, uchar * pos) {
2780
    TOKUDB_DBUG_ENTER("ha_tokudb::rnd_pos");
2781 2782 2783
    DBT db_pos;
    statistic_increment(table->in_use->status_var.ha_read_rnd_count, &LOCK_status);
    active_index = MAX_KEY;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2784
    DBT* key = get_pos(&db_pos, pos); 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2785
    TOKUDB_DBUG_RETURN(read_row(share->file->get(share->file, transaction, key, &current_row, 0), buf, primary_key, &current_row, key, 0));
2786 2787
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2788 2789 2790 2791 2792 2793 2794

int ha_tokudb::read_range_first(
    const key_range *start_key,
    const key_range *end_key,
    bool eq_range, 
    bool sorted) 
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2795 2796 2797 2798 2799 2800 2801 2802 2803 2804
    TOKUDB_DBUG_ENTER("ha_tokudb::read_range_first");
    int error;
    DBT start_dbt_key;
    const DBT* start_dbt_data = NULL;
    DBT end_dbt_key;
    const DBT* end_dbt_data = NULL;
    uchar start_key_buff [table_share->max_key_length + MAX_REF_PARTS * 3];
    uchar end_key_buff [table_share->max_key_length + MAX_REF_PARTS * 3];
    bzero((void *) &start_dbt_key, sizeof(start_dbt_key));
    bzero((void *) &end_dbt_key, sizeof(end_dbt_key));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2805
    range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
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 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848


    if (start_key) {
        pack_key(&start_dbt_key, active_index, start_key_buff, start_key->key, start_key->length);
        switch (start_key->flag) {
        case HA_READ_AFTER_KEY:
            start_dbt_data = share->key_file[active_index]->dbt_pos_infty();
            break;
        default:
            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) {
        pack_key(&end_dbt_key, active_index, end_key_buff, end_key->key, end_key->length);
        switch (end_key->flag) {
        case HA_READ_BEFORE_KEY:
            end_dbt_data = share->key_file[active_index]->dbt_neg_infty();
            break;
        default:
            end_dbt_data = share->key_file[active_index]->dbt_pos_infty();
            break;
        }
        
    }
    else {
        end_dbt_data = share->key_file[active_index]->dbt_pos_infty();
    }

    

    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
        );
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859
    if (error){ 
        last_cursor_error = error;
        //
        // cursor should be initialized here, but in case it is not, we still check
        //
        if (cursor) {
            cursor->c_close(cursor);
            cursor = NULL;
        }
        goto cleanup; 
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2860
    range_lock_grabbed = true;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2861 2862 2863 2864
    error = handler::read_range_first(start_key, end_key, eq_range, sorted);

cleanup:
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2865 2866 2867
}
int ha_tokudb::read_range_next()
{
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2868 2869 2870 2871
    TOKUDB_DBUG_ENTER("ha_tokudb::read_range_next");
    int error;
    error = handler::read_range_next();
    if (error) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2872
        range_lock_grabbed = false;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2873 2874
    }
    TOKUDB_DBUG_RETURN(error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2875 2876 2877 2878
}



2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899
/*
  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
*/
2900
void ha_tokudb::position(const uchar * record) {
2901
    TOKUDB_DBUG_ENTER("ha_tokudb::position");
2902 2903 2904 2905 2906
    DBT key;
    if (hidden_primary_key) {
        DBUG_ASSERT(ref_length == TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
        memcpy_fixed(ref, (char *) current_ident, TOKUDB_HIDDEN_PRIMARY_KEY_LENGTH);
    } else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2907
        create_dbt_key_from_table(&key, primary_key, ref, record);
2908 2909 2910 2911 2912 2913
        if (key.size < ref_length)
            bzero(ref + key.size, ref_length - key.size);
    }
    DBUG_VOID_RETURN;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2914 2915 2916 2917 2918 2919
//
// Per InnoDB: Returns statistics information of the table to the MySQL interpreter,
// in various fields of the handle object. 
// Return:
//      0, always success
//
2920
int ha_tokudb::info(uint flag) {
2921
    TOKUDB_DBUG_ENTER("ha_tokudb::info %p %d %lld %ld", this, flag, share->rows, changed_rows);
2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935
    if (flag & HA_STATUS_VARIABLE) {
        // Just to get optimizations right
        stats.records = share->rows + changed_rows;
        stats.deleted = 0;
    }
    if ((flag & HA_STATUS_CONST) || version != share->version) {
        version = share->version;
        for (uint i = 0; i < table_share->keys; i++) {
            table->key_info[i].rec_per_key[table->key_info[i].key_parts - 1] = share->rec_per_key[i];
        }
    }
    /* Don't return key if we got an error for the internal primary key */
    if (flag & HA_STATUS_ERRKEY && last_dup_key < table_share->keys)
        errkey = last_dup_key;
2936
    TOKUDB_DBUG_RETURN(0);
2937 2938
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
2939 2940 2941
//
//  Per InnoDB: Tells something additional to the handler about how to do things.
//
2942
int ha_tokudb::extra(enum ha_extra_function operation) {
2943
    TOKUDB_DBUG_ENTER("extra %p %d", this, operation);
2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962
    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;
    }
2963
    TOKUDB_DBUG_RETURN(0);
2964 2965 2966
}

int ha_tokudb::reset(void) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2967
    TOKUDB_DBUG_ENTER("ha_tokudb::reset");
2968 2969 2970 2971 2972 2973 2974 2975 2976
    key_read = 0;
    using_ignore = 0;
    if (current_row.flags & (DB_DBT_MALLOC | DB_DBT_REALLOC)) {
        current_row.flags = 0;
        if (current_row.data) {
            free(current_row.data);
            current_row.data = 0;
        }
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2977
    TOKUDB_DBUG_RETURN(0);
2978 2979 2980 2981 2982 2983 2984 2985 2986 2987
}

/*
  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
2988 2989 2990 2991 2992 2993 2994 2995
//
// Parameters:
//      [in]    thd - handle to the user thread
//              lock_type - the type of lock
// Returns:
//      0 on success
//      error otherwise
//
2996
int ha_tokudb::external_lock(THD * thd, int lock_type) {
2997
    TOKUDB_DBUG_ENTER("ha_tokudb::external_lock %d", thd_sql_command(thd));
2998
    // QQQ this is here to allow experiments without transactions
Zardosht Kasheff's avatar
Zardosht Kasheff committed
2999
    if ((tokudb_init_flags & DB_INIT_TXN) == 0) {
3000
        TOKUDB_DBUG_RETURN(0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3001
    }
3002
    int error = 0;
3003
    tokudb_trx_data *trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
3004
    if (!trx) {
3005
        trx = (tokudb_trx_data *)
3006
            my_malloc(sizeof(*trx), MYF(MY_ZEROFILL));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3007
        if (!trx) {
3008
            TOKUDB_DBUG_RETURN(1);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3009
        }
3010
        thd_data_set(thd, tokudb_hton->slot, trx);
3011
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3012
    if (trx->all == 0) {
3013
        trx->sp_level = 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3014
    }
3015 3016 3017
    if (lock_type != F_UNLCK) {
        if (!trx->tokudb_lock_count++) {
            DBUG_ASSERT(trx->stmt == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3018
            transaction = NULL;    // Safety
3019 3020
            /* First table lock, start transaction */
            if ((thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN | OPTION_TABLE_LOCK)) && !trx->all) {
3021
                /* QQQ We have to start a master transaction */
3022 3023 3024
                DBUG_PRINT("trans", ("starting transaction all:  options: 0x%lx", (ulong) thd->options));
                if ((error = db_env->txn_begin(db_env, NULL, &trx->all, 0))) {
                    trx->tokudb_lock_count--;      // We didn't get the lock
3025
                    TOKUDB_DBUG_RETURN(error);
3026
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3027
                if (tokudb_debug & TOKUDB_DEBUG_TXN) {
3028
                    TOKUDB_TRACE("master:%p\n", trx->all);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3029
                }
3030 3031
                trx->sp_level = trx->all;
                trans_register_ha(thd, TRUE, tokudb_hton);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3032
                if (thd->in_lock_tables) {
3033
                    TOKUDB_DBUG_RETURN(0);     // Don't create stmt trans
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3034
                }
3035 3036
            }
            DBUG_PRINT("trans", ("starting transaction stmt"));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3037 3038
            if (trx->stmt) { 
                if (tokudb_debug & TOKUDB_DEBUG_TXN) {
3039
                    TOKUDB_TRACE("warning:stmt=%p\n", trx->stmt);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3040 3041
                }
            }
3042 3043 3044
            if ((error = db_env->txn_begin(db_env, trx->sp_level, &trx->stmt, 0))) {
                /* We leave the possible master transaction open */
                trx->tokudb_lock_count--;  // We didn't get the lock
3045
                TOKUDB_DBUG_RETURN(error);
3046
            }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3047
            if (tokudb_debug & TOKUDB_DEBUG_TXN) {
3048
                TOKUDB_TRACE("stmt:%p:%p\n", trx->sp_level, trx->stmt);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3049
            }
3050 3051 3052
            trans_register_ha(thd, FALSE, tokudb_hton);
        }
        transaction = trx->stmt;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3053 3054
    }
    else {
3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066
        lock.type = TL_UNLOCK;  // Unlocked
        thread_safe_add(share->rows, changed_rows, &share->mutex);
        changed_rows = 0;
        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"));
                error = trx->stmt->commit(trx->stmt, 0);
Rich Prohaska's avatar
Rich Prohaska committed
3067
                if (tokudb_debug & TOKUDB_DEBUG_TXN)
3068
                    TOKUDB_TRACE("commit:%p:%d\n", trx->stmt, error);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3069
                trx->stmt = transaction = NULL;
3070 3071
            }
        }
3072
    }
3073
    TOKUDB_DBUG_RETURN(error);
3074 3075 3076 3077 3078 3079 3080 3081 3082
}


/*
  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.
*/

3083
int ha_tokudb::start_stmt(THD * thd, thr_lock_type lock_type) {
3084
    TOKUDB_DBUG_ENTER("ha_tokudb::start_stmt");
3085 3086
    if (!(tokudb_init_flags & DB_INIT_TXN)) 
        TOKUDB_DBUG_RETURN(0);
3087
    int error = 0;
3088
    tokudb_trx_data *trx = (tokudb_trx_data *) thd_data_get(thd, tokudb_hton->slot);
3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100
    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) {
        DBUG_PRINT("trans", ("starting transaction stmt"));
        error = db_env->txn_begin(db_env, trx->sp_level, &trx->stmt, 0);
        trans_register_ha(thd, FALSE, tokudb_hton);
    }
    transaction = trx->stmt;
3101
    TOKUDB_DBUG_RETURN(error);
3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131
}

/*
  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.
*/

3132
THR_LOCK_DATA **ha_tokudb::store_lock(THD * thd, THR_LOCK_DATA ** to, enum thr_lock_type lock_type) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3133
    TOKUDB_DBUG_ENTER("ha_tokudb::store_lock, lock_type=%d", lock_type);
3134 3135
    if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK) {
        /* If we are not doing a LOCK TABLE, then allow multiple writers */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3136
        if ((lock_type >= TL_WRITE_CONCURRENT_INSERT && lock_type <= TL_WRITE) && !thd->in_lock_tables) {
3137
            lock_type = TL_WRITE_ALLOW_WRITE;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3138
        }
3139
        lock.type = lock_type;
3140
    }
3141
    *to++ = &lock;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3142
    DBUG_RETURN(to);
3143 3144 3145 3146
}


static int create_sub_table(const char *table_name, const char *sub_name, DBTYPE type, int flags) {
3147
    TOKUDB_DBUG_ENTER("create_sub_table");
3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164
    int error;
    DB *file;
    DBUG_PRINT("enter", ("sub_name: %s  flags: %d", sub_name, flags));

    if (!(error = db_create(&file, db_env, 0))) {
        file->set_flags(file, flags);
        error = (file->open(file, NULL, table_name, sub_name, type, DB_THREAD | DB_CREATE, my_umask));
        if (error) {
            DBUG_PRINT("error", ("Got error: %d when opening table '%s'", error, table_name));
            (void) file->remove(file, table_name, NULL, 0);
        } else
            (void) file->close(file, 0);
    } else {
        DBUG_PRINT("error", ("Got error: %d when creating table", error));
    }
    if (error)
        my_errno = error;
3165
    TOKUDB_DBUG_RETURN(error);
3166 3167
}

3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183
static int mkdirpath(char *name, mode_t mode) {
    int r = mkdir(name, mode);
    if (r == -1 && errno == ENOENT) {
        char parent[strlen(name)+1];
        strcpy(parent, name);
        char *cp = strrchr(parent, '/');
        if (cp) {
            *cp = 0;
            r = mkdir(parent, 0755);
            if (r == 0)
                r = mkdir(name, mode);
        }
    }
    return r;
}

3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197
#include <dirent.h>

static int rmall(const char *dname) {
    int error = 0;
    DIR *d = opendir(dname);
    if (d) {
        struct dirent *dirent;
        while ((dirent = readdir(d)) != 0) {
            if (0 == strcmp(dirent->d_name, ".") || 0 == strcmp(dirent->d_name, ".."))
                continue;
            char fname[strlen(dname) + 1 + strlen(dirent->d_name) + 1];
            sprintf(fname, "%s/%s", dname, dirent->d_name);
            if (dirent->d_type == DT_DIR) {
                error = rmall(fname);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234
            } 
            else {
                if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
                    TOKUDB_TRACE("removing:%s\n", fname);
                }
                //
                // if clause checks if the file is a .tokudb file
                //
                if (strlen(fname) >= strlen (ha_tokudb_ext) &&
                    strcmp(fname + (strlen(fname) - strlen(ha_tokudb_ext)), ha_tokudb_ext) == 0) 
                {
                    //
                    // if this fails under low memory conditions, gracefully exit and return error
                    // user will be notified that something went wrong, and he will
                    // have to deal with it
                    //
                    DB* db = NULL;
                    error = db_create(&db, db_env, 0);
                    if (error) {
                        break;
                    }
                    //
                    // it is ok to do db->remove on any .tokudb file, because any such
                    // file was created with db->open
                    //
                    db->remove(db, fname, NULL, 0);
                }
                else {
                    //
                    // in case we have some file that is not .tokudb, we just delete it
                    //
                    error = unlink(fname);
                    if (error != 0) {
                        error = errno;
                        break;
                    }
                }
3235 3236 3237 3238 3239 3240 3241 3242
            }
        }
        closedir(d);
        if (error == 0) {
            error = rmdir(dname);
            if (error != 0)
                error = errno;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3243 3244
    } 
    else {
3245
        error = errno;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3246
    }
3247 3248 3249
    return error;
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3250 3251 3252 3253 3254 3255 3256 3257 3258 3259
//
// 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
//
3260
int ha_tokudb::create(const char *name, TABLE * form, HA_CREATE_INFO * create_info) {
3261
    TOKUDB_DBUG_ENTER("ha_tokudb::create");
3262 3263
    char name_buff[FN_REFLEN];
    int error;
3264
    char dirname[get_name_length(name) + 32];
3265
    char newname[get_name_length(name) + 32];
3266

3267
    uint i;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3268 3269 3270 3271

    //
    // tracing information about what type of table we are creating
    //
3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289
    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);
            }
        }
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3290 3291
    //
    // check if auto increment is properly defined
3292 3293
    // tokudb only supports auto increment on the first field in the primary key
    // or the first field in the row
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3294
    //
3295 3296 3297
    int pk_found = 0;
    int ai_found = 0;
    for (i = 0; i < form->s->keys; i++) {
3298
        KEY *key = &form->s->key_info[i];
3299 3300
        int is_primary = (strcmp(key->name, "PRIMARY") == 0);
        if (is_primary) pk_found = 1;
3301
        uint p;
3302
        for (p = 0; p < key->key_parts; p++) {
3303 3304
            KEY_PART_INFO *key_part = &key->key_part[p];
            Field *field = key_part->field;
3305 3306 3307 3308
            if (field->flags & AUTO_INCREMENT_FLAG) {
                ai_found = 1;
                if (is_primary && p > 0) 
                    TOKUDB_DBUG_RETURN(HA_ERR_UNSUPPORTED);
3309
            }
3310
        }
3311 3312 3313 3314 3315 3316 3317
    }

    if (!pk_found && ai_found) {
        Field *field = form->s->field[0];
        if (!(field->flags & AUTO_INCREMENT_FLAG))
            TOKUDB_DBUG_RETURN(HA_ERR_UNSUPPORTED);
    }
3318

3319
    // a table is a directory of dictionaries
3320 3321
    make_name(dirname, name, 0);
    error = mkdirpath(dirname, 0777);
3322
    if (error != 0) {
3323
        TOKUDB_DBUG_RETURN(errno);
3324
    }
3325

3326 3327
    make_name(newname, name, "main");
    fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
3328 3329

    /* Create the main table that will hold the real rows */
3330
    error = create_sub_table(name_buff, NULL, DB_BTREE, 0);
Rich Prohaska's avatar
Rich Prohaska committed
3331
    if (tokudb_debug & TOKUDB_DEBUG_OPEN)
3332
        TOKUDB_TRACE("create:%s:error=%d\n", newname, error);
3333 3334 3335 3336
    if (error) {
        rmall(dirname);
        TOKUDB_DBUG_RETURN(error);
    }
3337 3338 3339 3340

    primary_key = form->s->primary_key;

    /* Create the keys */
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3341
    char part[MAX_ALIAS_NAME + 10];
3342 3343
    for (uint i = 0; i < form->s->keys; i++) {
        if (i != primary_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3344
            sprintf(part, "key-%s", form->s->key_info[i].name);
3345 3346
            make_name(newname, name, part);
            fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
3347
            error = create_sub_table(name_buff, NULL, DB_BTREE, (form->key_info[i].flags & HA_NOSAME) ? 0 : DB_DUP + DB_DUPSORT);
Rich Prohaska's avatar
Rich Prohaska committed
3348
            if (tokudb_debug & TOKUDB_DEBUG_OPEN)
3349
                TOKUDB_TRACE("create:%s:flags=%ld:error=%d\n", newname, form->key_info[i].flags, error);
3350 3351 3352 3353
            if (error) {
                rmall(dirname);
                TOKUDB_DBUG_RETURN(error);
            }
3354 3355
        }
    }
3356

3357
    /* Create the status block to save information from last status command */
3358 3359
    /* QQQ Is DB_BTREE the best option here ? (QUEUE can't be used in sub tables) */
    // QQQ what is the status DB used for?
3360 3361 3362

    DB *status_block;
    if (!(error = (db_create(&status_block, db_env, 0)))) {
3363 3364
        make_name(newname, name, "status");
        fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
3365

3366
        if (!(error = (status_block->open(status_block, NULL, name_buff, NULL, DB_BTREE, DB_CREATE, 0)))) {
3367 3368 3369 3370 3371 3372
            char rec_buff[4 + MAX_KEY * 4];
            uint length = 4 + form->s->keys * 4;
            bzero(rec_buff, length);
            error = write_status(status_block, rec_buff, length);
            status_block->close(status_block, 0);
        }
Rich Prohaska's avatar
Rich Prohaska committed
3373
        if (tokudb_debug & TOKUDB_DEBUG_OPEN)
3374
            TOKUDB_TRACE("create:%s:error=%d\n", newname, error);
Rich Prohaska's avatar
Rich Prohaska committed
3375

3376
    }
3377 3378 3379

    if (error)
        rmall(dirname);
3380
    TOKUDB_DBUG_RETURN(error);
3381 3382
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3383 3384 3385 3386 3387 3388 3389 3390
//
// Drops table
// Parameters:
//      [in]    name - name of table to be deleted
// Returns:
//      0 on success
//      error otherwise
//
3391
int ha_tokudb::delete_table(const char *name) {
3392
    TOKUDB_DBUG_ENTER("ha_tokudb::delete_table");
3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414
    int error;
#if 0 // QQQ single file per table
    char name_buff[FN_REFLEN];
    char newname[strlen(name) + 32];

    sprintf(newname, "%s/main", name);
    fn_format(name_buff, newname, "", ha_tokudb_ext, MY_UNPACK_FILENAME | MY_APPEND_EXT);
    error = db_create(&file, db_env, 0);
    if (error != 0)
        goto exit;
    error = file->remove(file, name_buff, NULL, 0);

    sprintf(newname, "%s/status", name);
    fn_format(name_buff, newname, "", ha_tokudb_ext, MY_UNPACK_FILENAME | MY_APPEND_EXT);
    error = db_create(&file, db_env, 0);
    if (error != 0)
        goto exit;
    error = file->remove(file, name_buff, NULL, 0);

  exit:
    file = 0;                   // Safety
    my_errno = error;
3415
#else
3416
    // remove all of the dictionaries in the table directory 
3417 3418
    char newname[(tokudb_data_dir ? strlen(tokudb_data_dir) : 0) + strlen(name) + 32];
    make_name(newname, name, 0);
3419 3420
    error = rmall(newname);
    my_errno = error;
3421
#endif
3422
    TOKUDB_DBUG_RETURN(error);
3423 3424 3425
}


Zardosht Kasheff's avatar
Zardosht Kasheff committed
3426 3427 3428 3429 3430 3431 3432 3433 3434
//
// 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
//
3435
int ha_tokudb::rename_table(const char *from, const char *to) {
3436
    TOKUDB_DBUG_ENTER("%s %s %s", __FUNCTION__, from, to);
3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448
    int error;
#if 0 // QQQ single file per table
    char from_buff[FN_REFLEN];
    char to_buff[FN_REFLEN];

    if ((error = db_create(&file, db_env, 0)))
        my_errno = error;
    else {
        /* On should not do a file->close() after rename returns */
        error = file->rename(file,
                             fn_format(from_buff, from, "", ha_tokudb_ext, MY_UNPACK_FILENAME | MY_APPEND_EXT), NULL, fn_format(to_buff, to, "", ha_tokudb_ext, MY_UNPACK_FILENAME | MY_APPEND_EXT), 0);
    }
3449
#else
3450 3451 3452 3453 3454 3455
    int n = get_name_length(from) + 32;
    char newfrom[n];
    make_name(newfrom, from, 0);
    n = get_name_length(to) + 32;
    char newto[n];
    make_name(newto, to, 0);
3456 3457 3458
    error = rename(newfrom, newto);
    if (error != 0)
        error = my_errno = errno;
3459
#endif
3460
    TOKUDB_DBUG_RETURN(error);
3461 3462 3463 3464
}


/*
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3465
  Returns estimate on number of seeks it will take to read through the table
3466 3467 3468
  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.
*/
3469
/// QQQ why divide by 3
3470
double ha_tokudb::scan_time() {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3471 3472 3473
    TOKUDB_DBUG_ENTER("ha_tokudb::scan_time");
    double ret_val = stats.records / 3;
    DBUG_RETURN(ret_val);
3474 3475
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516
//
// 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(
	uint	index,
	uint	ranges,
	ha_rows rows
	)
{
    double total_scan;
    double ret_val; 

    if (index != primary_key) {
        ret_val = handler::read_time(index, ranges, rows);
        goto cleanup;
    }

    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
3517
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528
// 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
3529
//
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3530
ha_rows ha_tokudb::records_in_range(uint keynr, key_range* start_key, key_range* end_key) {
3531
    TOKUDB_DBUG_ENTER("ha_tokudb::records_in_range");
3532
    DBT key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3533
    ha_rows ret_val = HA_TOKUDB_RANGE_COUNT;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3534
    DB *kfile = share->key_file[keynr];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3535 3536 3537
    u_int64_t less, equal, greater;
    u_int64_t start_rows, end_rows, rows;
    int is_exact;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3538
    int error;
3539

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3540
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3541
    // get start_rows and end_rows values so that we can estimate range
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3542 3543
    //
    if (start_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3544
        error = kfile->key_range64(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3545 3546 3547
            kfile, 
            transaction, 
            pack_key(&key, keynr, key_buff, start_key->key, start_key->length), 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3548 3549 3550 3551
            &less,
            &equal,
            &greater,
            &is_exact
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3552 3553 3554 3555 3556 3557
            );
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
        if (start_key->flag == HA_READ_KEY_EXACT) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3558
            start_rows= less;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3559 3560
        }
        else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3561
            start_rows = less + equal;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3562
        }
3563
    }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3564
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3565
        start_rows= 0;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3566 3567 3568
    }

    if (end_key) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3569
        error = kfile->key_range64(
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3570 3571 3572
            kfile, 
            transaction, 
            pack_key(&key, keynr, key_buff, end_key->key, end_key->length), 
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3573 3574 3575 3576
            &less,
            &equal,
            &greater,
            &is_exact
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3577 3578 3579 3580 3581 3582
            );
        if (error) {
            ret_val = HA_TOKUDB_RANGE_COUNT;
            goto cleanup;
        }
        if (end_key->flag == HA_READ_BEFORE_KEY) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3583
            end_rows= less;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3584 3585
        }
        else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3586
            end_rows= less + equal;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3587 3588 3589
        }
    }
    else {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3590
        end_rows = stats.records;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3591 3592
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3593 3594
    rows = end_rows - start_rows;

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3595
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3596 3597
    // 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
3598
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3599
    ret_val = (ha_rows) (rows <= 1 ? 1 : rows);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3600 3601
cleanup:
    TOKUDB_DBUG_RETURN(ret_val);
3602 3603
}

3604
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
3605
    TOKUDB_DBUG_ENTER("ha_tokudb::get_auto_increment");
3606 3607
    ulonglong nr;

3608
    pthread_mutex_lock(&share->mutex);
3609 3610 3611 3612 3613
    if (!(share->status & STATUS_AUTO_INCREMENT_INIT)) {
        share->status |= STATUS_AUTO_INCREMENT_INIT;
        int error = read_last();
        if (error == 0) {
            share->last_auto_increment = table->next_number_field->val_int_offset(table->s->rec_buff_length);
3614
            if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT) 
3615
                TOKUDB_TRACE("init auto increment:%lld\n", share->last_auto_increment);
3616 3617
        }
    }
3618 3619
    nr = share->last_auto_increment + increment;
    share->last_auto_increment = nr + nb_desired_values - 1;
3620
    pthread_mutex_unlock(&share->mutex);
3621

3622
    if (tokudb_debug & TOKUDB_DEBUG_AUTO_INCREMENT)
3623 3624
        TOKUDB_TRACE("get_auto_increment(%lld,%lld,%lld):got:%lld:%lld\n",
                     offset, increment, nb_desired_values, nr, nb_desired_values);
3625
    *first_value = nr;
3626
    *nb_reserved_values = nb_desired_values;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3627
    DBUG_VOID_RETURN;
3628 3629
}

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652
//
// 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");
    char name_buff[FN_REFLEN];
    int error;
    char newname[share->table_name_length + 32];
    uint curr_index = 0;
    DBC* tmp_cursor = NULL;
    int cursor_ret_val = 0;
    DBT current_primary_key;
    DBT row;
    DB_TXN* txn = NULL;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3653
    uchar tmp_key_buff[2*table_arg->s->rec_buff_length];
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3654
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3655 3656 3657 3658
    // 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
3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670
    // these variables are for error handling
    //
    uint num_files_created = 0;
    uint num_DB_opened = 0;
    
    //
    // in unpack_row, MySQL passes a buffer that is this long,
    // so this length should be good enough for us as well
    //
    uchar tmp_record[table_arg->s->rec_buff_length];
    bzero((void *) &current_primary_key, sizeof(current_primary_key));
    bzero((void *) &row, sizeof(row));
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687


    //
    // 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
3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707
    //
    // first create all the DB's files
    //
    char part[MAX_ALIAS_NAME + 10];
    for (uint i = 0; i < num_of_keys; i++) {
        sprintf(part, "key-%s", key_info[i].name);
        make_name(newname, share->table_name, part);
        fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
        error = create_sub_table(name_buff, NULL, DB_BTREE, (key_info[i].flags & HA_NOSAME) ? 0 : DB_DUP + DB_DUPSORT);
        if (tokudb_debug & TOKUDB_DEBUG_OPEN) {
            TOKUDB_TRACE("create:%s:flags=%ld:error=%d\n", newname, key_info[i].flags, error);
        }
        if (error) { goto cleanup; }
        num_files_created++;
    }

    //
    // 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
3708
    curr_index = curr_num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727
    for (uint i = 0; i < num_of_keys; i++, curr_index++) {
        error = open_secondary_table(
            &share->key_file[curr_index], 
            &key_info[i],
            share->table_name,
            0,
            &share->key_type[curr_index]
            );
        if (error) { goto cleanup; }
        num_DB_opened++;
    }
    

    //
    // scan primary table, create each secondary key, add to each DB
    //
    
    error = db_env->txn_begin(db_env, 0, &txn, 0);
    assert(error == 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748

    //
    // 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
    //
    error = share->file->pre_acquire_read_lock(
        share->file, 
        txn, 
        share->file->dbt_neg_infty(), 
        NULL, 
        share->file->dbt_pos_infty(), 
        NULL
        );
    if (error) { txn->commit(txn, 0); goto cleanup; }

    //
    // now grab a table write lock for secondary tables we
    // are creating
    //
    for (uint i = 0; i < num_of_keys; i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3749
        uint curr_index = i + curr_num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3750 3751 3752 3753 3754 3755 3756
        error = share->key_file[curr_index]->pre_acquire_table_lock(
            share->key_file[curr_index],
            txn
            );
        if (error) { txn->commit(txn, 0); goto cleanup; }
    }

Zardosht Kasheff's avatar
Zardosht Kasheff committed
3757 3758 3759 3760 3761 3762 3763 3764 3765
    if ((error = share->file->cursor(share->file, txn, &tmp_cursor, 0))) {
        tmp_cursor = NULL;             // Safety
        goto cleanup;
    }

    //
    // for each element in the primary table, insert the proper key value pair in each secondary table
    // that is created
    //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3766
    cursor_ret_val = tmp_cursor->c_get(tmp_cursor, &current_primary_key, &row, DB_NEXT | DB_PRELOCKED);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3767 3768 3769 3770 3771
    while (cursor_ret_val != DB_NOTFOUND) {
        if (cursor_ret_val) {
            error = cursor_ret_val;
            goto cleanup;
        }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3772
        unpack_row(tmp_record, &row, &current_primary_key);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3773 3774
        for (uint i = 0; i < num_of_keys; i++) {
            DBT secondary_key;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3775
            create_dbt_key_from_key(&secondary_key,&key_info[i], tmp_key_buff, tmp_record);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3776
            uint curr_index = i + curr_num_DBs;
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3777 3778
            u_int32_t put_flags = share->key_type[curr_index];
            
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3779
            error = share->key_file[curr_index]->put(share->key_file[curr_index], txn, &secondary_key, &current_primary_key, put_flags);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3780 3781 3782 3783 3784 3785 3786 3787
            if (error) {
                //
                // in the case of any error anywhere, we can just nuke all the files created, so we dont need
                // to be tricky and try to roll back changes. That is why we commit the transaction,
                // which should be fast. The DB is going to go away anyway, so no pt in trying to keep
                // it in a good state.
                //
                txn->commit(txn, 0);
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3788 3789 3790 3791 3792 3793 3794 3795
                //
                // found a duplicate in a no_dup DB
                //
                if ( (error == DB_KEYEXIST) && (key_info[i].flags & HA_NOSAME)) {
                    error = HA_ERR_FOUND_DUPP_KEY;
                    last_dup_key = i;
                    memcpy(table_arg->record[0], tmp_record, table_arg->s->rec_buff_length);
                }
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811
                goto cleanup;
            }
        }
        cursor_ret_val = tmp_cursor->c_get(tmp_cursor, &current_primary_key, &row, DB_NEXT);
    }
    error = txn->commit(txn, 0);
    assert(error == 0);
    tmp_cursor->c_close(tmp_cursor);
    
    error = 0;
cleanup:
    if (error) {
        //
        // We need to delete all the files that may have been created
        // The DB's must be closed and removed
        //
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3812
        for (uint i = curr_num_DBs; i < curr_num_DBs + num_DB_opened; i++) {
Zardosht Kasheff's avatar
Zardosht Kasheff committed
3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902
            share->key_file[i]->close(share->key_file[i], 0);
            share->key_file[i] = NULL;
        }
        for (uint i = 0; i < num_files_created; i++) {
            DB* tmp;
            sprintf(part, "key-%s", key_info[i].name);
            make_name(newname, share->table_name, part);
            fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);
            if (!(db_create(&tmp, db_env, 0))) {
                tmp->remove(tmp, name_buff, NULL, 0);
            }
        }
    }
    TOKUDB_DBUG_RETURN(error);
}

//
// 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;
    char name_buff[FN_REFLEN];
    char newname[share->table_name_length + 32];
    char part[MAX_ALIAS_NAME + 10];
    DB** dbs_to_remove = NULL;

    //
    // we allocate an array of DB's here to get ready for removal
    // We do this so that all potential memory allocation errors that may occur
    // will do so BEFORE we go about dropping any indexes. This way, we
    // can fail gracefully without losing integrity of data in such cases. If on
    // on the other hand, we started removing DB's, and in the middle, 
    // one failed, it is not immedietely obvious how one would rollback
    //
    dbs_to_remove = (DB **)my_malloc(sizeof(*dbs_to_remove)*num_of_keys, MYF(MY_ZEROFILL));
    if (dbs_to_remove == NULL) {
        error = ENOMEM; 
        goto cleanup;
    }
    for (uint i = 0; i < num_of_keys; i++) {
        error = db_create(&dbs_to_remove[i], db_env, 0);
        if (error) {
            goto cleanup;
        }
    }    
    
    for (uint i = 0; i < num_of_keys; i++) {
        uint curr_index = key_num[i];
        share->key_file[curr_index]->close(share->key_file[curr_index],0);
        share->key_file[curr_index] = NULL;
        
        sprintf(part, "key-%s", table_arg->key_info[curr_index].name);
        make_name(newname, share->table_name, part);
        fn_format(name_buff, newname, "", 0, MY_UNPACK_FILENAME);

        dbs_to_remove[i]->remove(dbs_to_remove[i], name_buff, NULL, 0);
    }
cleanup:
    my_free(dbs_to_remove, MYF(MY_ALLOW_ZERO_PTR));
    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);
}

3903
void ha_tokudb::print_error(int error, myf errflag) {
3904
    if (error == DB_LOCK_DEADLOCK)
3905
        error = HA_ERR_LOCK_DEADLOCK;
3906
    if (error == DB_LOCK_NOTGRANTED)
3907
        error = HA_ERR_LOCK_WAIT_TIMEOUT;
3908 3909 3910
    handler::print_error(error, errflag);
}

3911
#if 0 // QQQ use default
3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942
int ha_tokudb::analyze(THD * thd, HA_CHECK_OPT * check_opt) {
    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->rows = stat->bt_ndata;
    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
3943
    if (stat)
3944 3945 3946 3947
        free(stat);
    return ((share->status & STATUS_TOKUDB_ANALYZE) ? HA_ADMIN_FAILED : HA_ADMIN_OK);

  err:
3948
    if (stat)
3949 3950
        free(stat);
    return HA_ADMIN_FAILED;
3951
}
3952
#endif
3953

3954
#if 0 // QQQ use default
3955 3956
int ha_tokudb::optimize(THD * thd, HA_CHECK_OPT * check_opt) {
    return ha_tokudb::analyze(thd, check_opt);
3957
}
3958
#endif
3959

3960
#if 0 // QQQ use default
3961
int ha_tokudb::check(THD * thd, HA_CHECK_OPT * check_opt) {
3962
    TOKUDB_DBUG_ENTER("ha_tokudb::check");
3963
    TOKUDB_DBUG_RETURN(HA_ADMIN_NOT_IMPLEMENTED);
3964
    // look in old_ha_tokudb.cc for a start of an implementation
3965
}
3966
#endif
3967

3968 3969 3970 3971 3972 3973
ulong ha_tokudb::field_offset(Field *field) {
    if (table->record[0] <= field->ptr && field->ptr < table->record[1])
        return field->offset(table->record[0]);
    assert(0);
    return 0;
}
3974

3975
struct st_mysql_storage_engine storage_engine_structure = { MYSQL_HANDLERTON_INTERFACE_VERSION };
3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986

// options flags
//   PLUGIN_VAR_THDLOCAL  Variable is per-connection
//   PLUGIN_VAR_READONLY  Server variable is read only
//   PLUGIN_VAR_NOSYSVAR  Not a server variable
//   PLUGIN_VAR_NOCMDOPT  Not a command line option
//   PLUGIN_VAR_NOCMDARG  No argument for cmd line
//   PLUGIN_VAR_RQCMDARG  Argument required for cmd line
//   PLUGIN_VAR_OPCMDARG  Argument optional for cmd line
//   PLUGIN_VAR_MEMALLOC  String needs memory allocated

3987 3988 3989

// system variables

Rich Prohaska's avatar
Rich Prohaska committed
3990 3991
static MYSQL_SYSVAR_ULONGLONG(cache_size, tokudb_cache_size, PLUGIN_VAR_READONLY, "TokuDB cache table size", NULL, NULL, 0, 0, ~0LL, 0);

3992 3993
static MYSQL_SYSVAR_UINT(cache_memory_percent, tokudb_cache_memory_percent, PLUGIN_VAR_READONLY, "Default percent of physical memory in the TokuDB cache table", NULL, NULL, tokudb_cache_memory_percent, 0, 100, 0);

Rich Prohaska's avatar
Rich Prohaska committed
3994 3995
static MYSQL_SYSVAR_ULONG(max_lock, tokudb_max_lock, PLUGIN_VAR_READONLY, "TokuDB Max Locks", NULL, NULL, 8 * 1024, 0, ~0L, 0);

3996
static MYSQL_SYSVAR_ULONG(debug, tokudb_debug, PLUGIN_VAR_READONLY, "TokuDB Debug", NULL, NULL, 0, 0, ~0L, 0);
3997

3998
static MYSQL_SYSVAR_STR(log_dir, tokudb_log_dir, PLUGIN_VAR_READONLY, "TokuDB Log Directory", NULL, NULL, NULL);
3999

4000
static MYSQL_SYSVAR_STR(data_dir, tokudb_data_dir, PLUGIN_VAR_READONLY, "TokuDB Data Directory", NULL, NULL, NULL);
4001

4002 4003
static MYSQL_SYSVAR_STR(version, tokudb_version, PLUGIN_VAR_READONLY, "TokuDB Version", NULL, NULL, NULL);

4004 4005
static MYSQL_SYSVAR_UINT(init_flags, tokudb_init_flags, PLUGIN_VAR_READONLY, "Sets TokuDB DB_ENV->open flags", NULL, NULL, tokudb_init_flags, 0, ~0, 0);

4006 4007
#if 0

4008
static MYSQL_SYSVAR_ULONG(cache_parts, tokudb_cache_parts, PLUGIN_VAR_READONLY, "Sets TokuDB set_cache_parts", NULL, NULL, 0, 0, ~0L, 0);
4009 4010 4011

// this is really a u_int32_t
// ? use MYSQL_SYSVAR_SET
4012
static MYSQL_SYSVAR_UINT(env_flags, tokudb_env_flags, PLUGIN_VAR_READONLY, "Sets TokuDB env_flags", NULL, NULL, DB_LOG_AUTOREMOVE, 0, ~0, 0);
4013

4014
static MYSQL_SYSVAR_STR(home, tokudb_home, PLUGIN_VAR_READONLY, "Sets TokuDB env->open home", NULL, NULL, NULL);
4015 4016 4017 4018 4019

// this is really a u_int32_t
//? use MYSQL_SYSVAR_SET

// this looks to be unused
4020
static MYSQL_SYSVAR_LONG(lock_scan_time, tokudb_lock_scan_time, PLUGIN_VAR_READONLY, "Tokudb Lock Scan Time (UNUSED)", NULL, NULL, 0, 0, ~0L, 0);
4021 4022 4023

// this is really a u_int32_t
//? use MYSQL_SYSVAR_ENUM
4024 4025 4026 4027 4028 4029 4030 4031 4032
static MYSQL_SYSVAR_UINT(lock_type, tokudb_lock_type, PLUGIN_VAR_READONLY, "Sets set_lk_detect", NULL, NULL, DB_LOCK_DEFAULT, 0, ~0, 0);

static MYSQL_SYSVAR_ULONG(log_buffer_size, tokudb_log_buffer_size, PLUGIN_VAR_READONLY, "Tokudb Log Buffer Size", NULL, NULL, 0, 0, ~0L, 0);

static MYSQL_SYSVAR_ULONG(region_size, tokudb_region_size, PLUGIN_VAR_READONLY, "Tokudb Region Size", NULL, NULL, 128 * 1024, 0, ~0L, 0);

static MYSQL_SYSVAR_BOOL(shared_data, tokudb_shared_data, PLUGIN_VAR_READONLY, "Tokudb Shared Data", NULL, NULL, FALSE);

static MYSQL_SYSVAR_STR(tmpdir, tokudb_tmpdir, PLUGIN_VAR_READONLY, "Tokudb Tmp Dir", NULL, NULL, NULL);
4033
#endif
4034

4035 4036
static struct st_mysql_sys_var *tokudb_system_variables[] = {
    MYSQL_SYSVAR(cache_size),
4037
    MYSQL_SYSVAR(cache_memory_percent),
4038
    MYSQL_SYSVAR(max_lock),
4039 4040
    MYSQL_SYSVAR(data_dir),
    MYSQL_SYSVAR(log_dir),
4041
    MYSQL_SYSVAR(debug),
4042
    MYSQL_SYSVAR(commit_sync),
4043
    MYSQL_SYSVAR(version),
4044
    MYSQL_SYSVAR(init_flags),
4045
#if 0
4046
    MYSQL_SYSVAR(cache_parts),
4047 4048 4049 4050 4051 4052 4053 4054
    MYSQL_SYSVAR(env_flags),
    MYSQL_SYSVAR(home),
    MYSQL_SYSVAR(lock_scan_time),
    MYSQL_SYSVAR(lock_type),
    MYSQL_SYSVAR(log_buffer_size),
    MYSQL_SYSVAR(region_size),
    MYSQL_SYSVAR(shared_data),
    MYSQL_SYSVAR(tmpdir),
4055
#endif
4056
    NULL
4057 4058
};

4059
mysql_declare_plugin(tokudb) {
4060 4061 4062 4063
    MYSQL_STORAGE_ENGINE_PLUGIN, 
    &storage_engine_structure, 
    "TokuDB", 
    "Tokutek Inc", 
4064
    "Fractal trees, transactions, row level locks",
4065
    PLUGIN_LICENSE_PROPRIETARY,        /* QQQ license? */
4066
    tokudb_init_func,          /* plugin init */
4067
    tokudb_done_func,          /* plugin deinit */
4068 4069
    0x0200,                    /* QQQ 2.0 */
    NULL,                      /* status variables */
4070
    tokudb_system_variables,   /* system variables */
4071
    NULL                       /* config options */
4072 4073
}
mysql_declare_plugin_end;