ha_partition.cc 170 KB
Newer Older
1 2 3 4
/* Copyright (C) 2005 MySQL AB

  This program is free software; you can redistribute it and/or modify
  it under the terms of the GNU General Public License as published by
5
  the Free Software Foundation; version 2 of the License.
6 7 8 9 10 11 12 13 14 15 16

  This program is distributed in the hope that it will be useful,
  but WITHOUT ANY WARRANTY; without even the implied warranty of
  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  GNU General Public License for more details.

  You should have received a copy of the GNU General Public License
  along with this program; if not, write to the Free Software
  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */

/*
patg@krsna.patg.net's avatar
patg@krsna.patg.net committed
17
  This handler was developed by Mikael Ronstrom for version 5.1 of MySQL.
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
  It is an abstraction layer on top of other handlers such as MyISAM,
  InnoDB, Federated, Berkeley DB and so forth. Partitioned tables can also
  be handled by a storage engine. The current example of this is NDB
  Cluster that has internally handled partitioning. This have benefits in
  that many loops needed in the partition handler can be avoided.

  Partitioning has an inherent feature which in some cases is positive and
  in some cases is negative. It splits the data into chunks. This makes
  the data more manageable, queries can easily be parallelised towards the
  parts and indexes are split such that there are less levels in the
  index trees. The inherent disadvantage is that to use a split index
  one has to scan all index parts which is ok for large queries but for
  small queries it can be a disadvantage.

  Partitioning lays the foundation for more manageable databases that are
  extremely large. It does also lay the foundation for more parallelism
  in the execution of queries. This functionality will grow with later
  versions of MySQL.

  You can enable it in your buld by doing the following during your build
  process:
  ./configure --with-partition

  The partition is setup to use table locks. It implements an partition "SHARE"
  that is inserted into a hash by table name. You can use this to store
  information of state that any partition handler object will be able to see
  if it is using the same table.

  Please read the object definition in ha_partition.h before reading the rest
  if this file.
*/

#ifdef __GNUC__
#pragma implementation				// gcc: Class implementation
#endif

kent@mysql.com's avatar
kent@mysql.com committed
54
#include "mysql_priv.h"
55

acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
56
#ifdef WITH_PARTITION_STORAGE_ENGINE
57 58
#include "ha_partition.h"

acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
59 60
#include <mysql/plugin.h>

61 62 63 64 65 66 67 68 69 70
static const char *ha_par_ext= ".par";
#ifdef NOT_USED
static int free_share(PARTITION_SHARE * share);
static PARTITION_SHARE *get_share(const char *table_name, TABLE * table);
#endif

/****************************************************************************
                MODULE create/delete handler object
****************************************************************************/

71 72
static handler *partition_create_handler(handlerton *hton,
                                         TABLE_SHARE *share,
73
                                         MEM_ROOT *mem_root);
74 75
static uint partition_flags();
static uint alter_table_flags(uint flags);
76

77

78
static int partition_initialize(void *p)
79
{
80

81
  handlerton *partition_hton;
82 83 84 85 86 87 88 89 90
  partition_hton= (handlerton *)p;

  partition_hton->state= SHOW_OPTION_YES;
  partition_hton->db_type= DB_TYPE_PARTITION_DB;
  partition_hton->create= partition_create_handler;
  partition_hton->partition_flags= partition_flags;
  partition_hton->alter_table_flags= alter_table_flags;
  partition_hton->flags= HTON_NOT_USER_SELECTABLE | HTON_HIDDEN;

91 92
  return 0;
}
tulin@dl145b.mysql.com's avatar
merge  
tulin@dl145b.mysql.com committed
93

94 95 96 97 98 99 100 101 102 103 104
/*
  Create new partition handler

  SYNOPSIS
    partition_create_handler()
    table                       Table object

  RETURN VALUE
    New partition object
*/

105 106
static handler *partition_create_handler(handlerton *hton, 
                                         TABLE_SHARE *share,
107
                                         MEM_ROOT *mem_root)
108
{
109
  ha_partition *file= new (mem_root) ha_partition(hton, share);
110 111 112 113 114 115
  if (file && file->initialise_partition(mem_root))
  {
    delete file;
    file= 0;
  }
  return file;
116 117
}

118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147
/*
  HA_CAN_PARTITION:
  Used by storage engines that can handle partitioning without this
  partition handler
  (Partition, NDB)

  HA_CAN_UPDATE_PARTITION_KEY:
  Set if the handler can update fields that are part of the partition
  function.

  HA_CAN_PARTITION_UNIQUE:
  Set if the handler can handle unique indexes where the fields of the
  unique key are not part of the fields of the partition function. Thus
  a unique key can be set on all fields.

  HA_USE_AUTO_PARTITION
  Set if the handler sets all tables to be partitioned by default.
*/

static uint partition_flags()
{
  return HA_CAN_PARTITION;
}

static uint alter_table_flags(uint flags __attribute__((unused)))
{
  return (HA_PARTITION_FUNCTION_SUPPORTED |
          HA_FAST_CHANGE_PARTITION);
}

148 149
const uint ha_partition::NO_CURRENT_PART_ID= 0xFFFFFFFF;

150 151 152 153 154 155 156 157 158 159
/*
  Constructor method

  SYNOPSIS
    ha_partition()
    table                       Table object

  RETURN VALUE
    NONE
*/
160

161 162
ha_partition::ha_partition(handlerton *hton, TABLE_SHARE *share)
  :handler(hton, share), m_part_info(NULL), m_create_handler(FALSE),
163
   m_is_sub_partitioned(0), is_clone(FALSE)
164 165 166 167 168 169 170
{
  DBUG_ENTER("ha_partition::ha_partition(table)");
  init_handler_variables();
  DBUG_VOID_RETURN;
}


171 172 173 174 175 176 177 178 179 180 181
/*
  Constructor method

  SYNOPSIS
    ha_partition()
    part_info                       Partition info

  RETURN VALUE
    NONE
*/

182 183
ha_partition::ha_partition(handlerton *hton, partition_info *part_info)
  :handler(hton, NULL), m_part_info(part_info),
184
   m_create_handler(TRUE),
185
   m_is_sub_partitioned(m_part_info->is_sub_partitioned()), is_clone(FALSE)
186 187 188 189 190 191 192 193
{
  DBUG_ENTER("ha_partition::ha_partition(part_info)");
  init_handler_variables();
  DBUG_ASSERT(m_part_info);
  DBUG_VOID_RETURN;
}


194 195 196 197 198 199 200 201 202 203
/*
  Initialise handler object

  SYNOPSIS
    init_handler_variables()

  RETURN VALUE
    NONE
*/

204 205 206
void ha_partition::init_handler_variables()
{
  active_index= MAX_KEY;
207 208
  m_mode= 0;
  m_open_test_lock= 0;
209 210 211 212
  m_file_buffer= NULL;
  m_name_buffer_ptr= NULL;
  m_engine_array= NULL;
  m_file= NULL;
213
  m_file_tot_parts= 0;
214
  m_reorged_file= NULL;
215
  m_new_file= NULL;
216 217
  m_reorged_parts= 0;
  m_added_file= NULL;
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
  m_tot_parts= 0;
  m_pkey_is_clustered= 0;
  m_lock_type= F_UNLCK;
  m_part_spec.start_part= NO_CURRENT_PART_ID;
  m_scan_value= 2;
  m_ref_length= 0;
  m_part_spec.end_part= NO_CURRENT_PART_ID;
  m_index_scan_type= partition_no_index_scan;
  m_start_key.key= NULL;
  m_start_key.length= 0;
  m_myisam= FALSE;
  m_innodb= FALSE;
  m_extra_cache= FALSE;
  m_extra_cache_size= 0;
  m_table_flags= HA_FILE_BASED | HA_REC_NOT_IN_SEQ;
  m_low_byte_first= 1;
  m_part_field_array= NULL;
  m_ordered_rec_buffer= NULL;
  m_top_entry= NO_CURRENT_PART_ID;
  m_rec_length= 0;
  m_last_part= 0;
  m_rec0= 0;
  m_curr_key_info= 0;
patg@krsna.patg.net's avatar
patg@krsna.patg.net committed
241 242 243 244
  /*
    this allows blackhole to work properly
  */
  m_no_locks= 0;
245 246 247 248 249 250 251 252

#ifdef DONT_HAVE_TO_BE_INITALIZED
  m_start_key.flag= 0;
  m_ordered= TRUE;
#endif
}


253 254 255 256 257 258 259
const char *ha_partition::table_type() const
{ 
  // we can do this since we only support a single engine type
  return m_file[0]->table_type(); 
}


260 261 262 263 264 265 266 267 268 269
/*
  Destructor method

  SYNOPSIS
    ~ha_partition()

  RETURN VALUE
    NONE
*/

270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
ha_partition::~ha_partition()
{
  DBUG_ENTER("ha_partition::~ha_partition()");
  if (m_file != NULL)
  {
    uint i;
    for (i= 0; i < m_tot_parts; i++)
      delete m_file[i];
  }
  my_free((char*) m_ordered_rec_buffer, MYF(MY_ALLOW_ZERO_PTR));

  clear_handler_file();
  DBUG_VOID_RETURN;
}


/*
287 288 289
  Initialise partition handler object

  SYNOPSIS
290 291
    initialise_partition()
    mem_root			Allocate memory through this
292 293 294 295 296 297 298

  RETURN VALUE
    1                         Error
    0                         Success

  DESCRIPTION

299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327
  The partition handler is only a layer on top of other engines. Thus it
  can't really perform anything without the underlying handlers. Thus we
  add this method as part of the allocation of a handler object.

  1) Allocation of underlying handlers
     If we have access to the partition info we will allocate one handler
     instance for each partition.
  2) Allocation without partition info
     The cases where we don't have access to this information is when called
     in preparation for delete_table and rename_table and in that case we
     only need to set HA_FILE_BASED. In that case we will use the .par file
     that contains information about the partitions and their engines and
     the names of each partition.
  3) Table flags initialisation
     We need also to set table flags for the partition handler. This is not
     static since it depends on what storage engines are used as underlying
     handlers.
     The table flags is set in this routine to simulate the behaviour of a
     normal storage engine
     The flag HA_FILE_BASED will be set independent of the underlying handlers
  4) Index flags initialisation
     When knowledge exists on the indexes it is also possible to initialise the
     index flags. Again the index flags must be initialised by using the under-
     lying handlers since this is storage engine dependent.
     The flag HA_READ_ORDER will be reset for the time being to indicate no
     ordered output is available from partition handler indexes. Later a merge
     sort will be performed using the underlying handlers.
  5) primary_key_is_clustered, has_transactions and low_byte_first is
     calculated here.
328

329 330
*/

331
bool ha_partition::initialise_partition(MEM_ROOT *mem_root)
332 333
{
  handler **file_array, *file;
334
  DBUG_ENTER("ha_partition::initialise_partition");
335

336
  if (m_create_handler)
337
  {
338
    m_tot_parts= m_part_info->get_tot_partitions();
339
    DBUG_ASSERT(m_tot_parts > 0);
340
    if (new_handlers_from_part_info(mem_root))
341
      DBUG_RETURN(1);
342 343 344
  }
  else if (!table_share || !table_share->normalized_path.str)
  {
345
    /*
346 347 348
      Called with dummy table share (delete, rename and alter table)
      Don't need to set-up table flags other than
      HA_FILE_BASED here
349
    */
350 351 352
    m_table_flags|= HA_FILE_BASED | HA_REC_NOT_IN_SEQ;
    DBUG_RETURN(0);
  }
353
  else if (get_from_handler_file(table_share->normalized_path.str, mem_root))
354
  {
355
    mem_alloc_error(2);
356
    DBUG_RETURN(1);
357
  }
358 359 360 361 362 363 364 365 366
  /*
    We create all underlying table handlers here. We do it in this special
    method to be able to report allocation errors.

    Set up table_flags, low_byte_first, primary_key_is_clustered and
    has_transactions since they are called often in all kinds of places,
    other parameters are calculated on demand.
    HA_FILE_BASED is always set for partition handler since we use a
    special file for handling names of partitions, engine types.
367
    HA_CAN_GEOMETRY, HA_CAN_FULLTEXT, HA_CAN_SQL_HANDLER, HA_DUPLICATE_POS,
368 369
    HA_CAN_INSERT_DELAYED is disabled until further investigated.
  */
370
  m_table_flags= (ulong)m_file[0]->ha_table_flags();
371 372 373 374 375 376 377 378 379 380 381 382 383 384
  m_low_byte_first= m_file[0]->low_byte_first();
  m_pkey_is_clustered= TRUE;
  file_array= m_file;
  do
  {
    file= *file_array;
    if (m_low_byte_first != file->low_byte_first())
    {
      // Cannot have handlers with different endian
      my_error(ER_MIX_HANDLER_ERROR, MYF(0));
      DBUG_RETURN(1);
    }
    if (!file->primary_key_is_clustered())
      m_pkey_is_clustered= FALSE;
385
    m_table_flags&= file->ha_table_flags();
386
  } while (*(++file_array));
387
  m_table_flags&= ~(HA_CAN_GEOMETRY | HA_CAN_FULLTEXT | HA_DUPLICATE_POS |
388 389
                    HA_CAN_SQL_HANDLER | HA_CAN_INSERT_DELAYED |
                    HA_PRIMARY_KEY_REQUIRED_FOR_POSITION);
390 391 392 393 394 395 396 397
  m_table_flags|= HA_FILE_BASED | HA_REC_NOT_IN_SEQ;
  DBUG_RETURN(0);
}

/****************************************************************************
                MODULE meta data changes
****************************************************************************/
/*
398
  Delete a table
399

400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
  SYNOPSIS
    delete_table()
    name                    Full path of table name

  RETURN VALUE
    >0                        Error
    0                         Success

  DESCRIPTION
    Used to delete a table. By the time delete_table() has been called all
    opened references to this table will have been closed (and your globally
    shared references released. The variable name will just be the name of
    the table. You will need to remove any files you have created at this
    point.

    If you do not implement this, the default delete_table() is called from
    handler.cc and it will delete all files with the file extentions returned
    by bas_ext().

    Called from handler.cc by delete_table and  ha_create_table(). Only used
    during create if the table_flag HA_DROP_BEFORE_CREATE was specified for
    the storage engine.
422 423 424 425 426 427
*/

int ha_partition::delete_table(const char *name)
{
  int error;
  DBUG_ENTER("ha_partition::delete_table");
428

429 430 431 432 433 434 435
  if ((error= del_ren_cre_table(name, NULL, NULL, NULL)))
    DBUG_RETURN(error);
  DBUG_RETURN(handler::delete_table(name));
}


/*
436 437 438 439 440 441 442 443 444 445
  Rename a table

  SYNOPSIS
    rename_table()
    from                      Full path of old table name
    to                        Full path of new table name

  RETURN VALUE
    >0                        Error
    0                         Success
446

447 448
  DESCRIPTION
    Renames a table from one name to another from alter table call.
449

450 451 452 453 454
    If you do not implement this, the default rename_table() is called from
    handler.cc and it will rename all files with the file extentions returned
    by bas_ext().

    Called from sql_table.cc by mysql_rename_table().
455 456 457 458 459 460
*/

int ha_partition::rename_table(const char *from, const char *to)
{
  int error;
  DBUG_ENTER("ha_partition::rename_table");
461

462 463 464 465 466 467 468
  if ((error= del_ren_cre_table(from, to, NULL, NULL)))
    DBUG_RETURN(error);
  DBUG_RETURN(handler::rename_table(from, to));
}


/*
469 470 471 472 473
  Create the handler file (.par-file)

  SYNOPSIS
    create_handler_files()
    name                              Full path of table name
474
    create_info                       Create info generated for CREATE TABLE
475 476 477 478 479 480 481 482 483 484 485

  RETURN VALUE
    >0                        Error
    0                         Success

  DESCRIPTION
    create_handler_files is called to create any handler specific files
    before opening the file with openfrm to later call ::create on the
    file object.
    In the partition handler this is used to store the names of partitions
    and types of engines in the partitions.
486 487
*/

488 489
int ha_partition::create_handler_files(const char *path,
                                       const char *old_path,
490
                                       int action_flag,
491
                                       HA_CREATE_INFO *create_info)
492 493
{
  DBUG_ENTER("ha_partition::create_handler_files()");
494 495 496 497 498

  /*
    We need to update total number of parts since we might write the handler
    file as part of a partition management command
  */
499 500
  if (action_flag == CHF_DELETE_FLAG ||
      action_flag == CHF_RENAME_FLAG)
501
  {
502 503 504
    char name[FN_REFLEN];
    char old_name[FN_REFLEN];

505 506
    strxmov(name, path, ha_par_ext, NullS);
    strxmov(old_name, old_path, ha_par_ext, NullS);
507 508 509 510
    if ((action_flag == CHF_DELETE_FLAG &&
         my_delete(name, MYF(MY_WME))) ||
        (action_flag == CHF_RENAME_FLAG &&
         my_rename(old_name, name, MYF(MY_WME))))
511 512 513 514
    {
      DBUG_RETURN(TRUE);
    }
  }
515
  else if (action_flag == CHF_CREATE_FLAG)
516 517 518 519 520 521
  {
    if (create_handler_file(path))
    {
      my_error(ER_CANT_CREATE_HANDLER_FILE, MYF(0));
      DBUG_RETURN(1);
    }
522 523 524 525 526 527
  }
  DBUG_RETURN(0);
}


/*
528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591
  Create a partitioned table

  SYNOPSIS
    create()
    name                              Full path of table name
    table_arg                         Table object
    create_info                       Create info generated for CREATE TABLE

  RETURN VALUE
    >0                        Error
    0                         Success

  DESCRIPTION
    create() is called to create a table. The variable name will have the name
    of the table. When create() is called you do not need to worry about
    opening the table. Also, the FRM file will have already been created so
    adjusting create_info will not do you any good. You can overwrite the frm
    file at this point if you wish to change the table definition, but there
    are no methods currently provided for doing that.

    Called from handler.cc by ha_create_table().
*/

int ha_partition::create(const char *name, TABLE *table_arg,
			 HA_CREATE_INFO *create_info)
{
  char t_name[FN_REFLEN];
  DBUG_ENTER("ha_partition::create");

  strmov(t_name, name);
  DBUG_ASSERT(*fn_rext((char*)name) == '\0');
  if (del_ren_cre_table(t_name, NULL, table_arg, create_info))
  {
    handler::delete_table(t_name);
    DBUG_RETURN(1);
  }
  DBUG_RETURN(0);
}


/*
  Drop partitions as part of ALTER TABLE of partitions

  SYNOPSIS
    drop_partitions()
    path                        Complete path of db and table name

  RETURN VALUE
    >0                          Failure
    0                           Success

  DESCRIPTION
    Use part_info object on handler object to deduce which partitions to
    drop (each partition has a state attached to it)
*/

int ha_partition::drop_partitions(const char *path)
{
  List_iterator<partition_element> part_it(m_part_info->partitions);
  char part_name_buff[FN_REFLEN];
  uint no_parts= m_part_info->partitions.elements;
  uint no_subparts= m_part_info->no_subparts;
  uint i= 0;
  uint name_variant;
592
  int  ret_error;
593
  int  error= 0;
594 595
  DBUG_ENTER("ha_partition::drop_partitions");

596 597 598 599 600 601
  /*
    Assert that it works without HA_FILE_BASED and lower_case_table_name = 2.
    We use m_file[0] as long as all partitions have the same storage engine.
  */
  DBUG_ASSERT(!strcmp(path, get_canonical_filename(m_file[0], path,
                                                   part_name_buff)));
602 603
  do
  {
604 605
    partition_element *part_elem= part_it++;
    if (part_elem->part_state == PART_TO_BE_DROPPED)
606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622
    {
      handler *file;
      /*
        This part is to be dropped, meaning the part or all its subparts.
      */
      name_variant= NORMAL_PART_NAME;
      if (m_is_sub_partitioned)
      {
        List_iterator<partition_element> sub_it(part_elem->subpartitions);
        uint j= 0, part;
        do
        {
          partition_element *sub_elem= sub_it++;
          part= i * no_subparts + j;
          create_subpartition_name(part_name_buff, path,
                                   part_elem->partition_name,
                                   sub_elem->partition_name, name_variant);
623
          file= m_file[part];
624
          DBUG_PRINT("info", ("Drop subpartition %s", part_name_buff));
625
          if ((ret_error= file->ha_delete_table(part_name_buff)))
626
            error= ret_error;
627 628
          if (deactivate_ddl_log_entry(sub_elem->log_entry->entry_pos))
            error= 1;
629 630 631 632 633 634 635
        } while (++j < no_subparts);
      }
      else
      {
        create_partition_name(part_name_buff, path,
                              part_elem->partition_name, name_variant,
                              TRUE);
636
        file= m_file[i];
637
        DBUG_PRINT("info", ("Drop partition %s", part_name_buff));
638
        if ((ret_error= file->ha_delete_table(part_name_buff)))
639
          error= ret_error;
640 641
        if (deactivate_ddl_log_entry(part_elem->log_entry->entry_pos))
          error= 1;
642 643 644 645 646 647 648
      }
      if (part_elem->part_state == PART_IS_CHANGED)
        part_elem->part_state= PART_NORMAL;
      else
        part_elem->part_state= PART_IS_DROPPED;
    }
  } while (++i < no_parts);
649
  VOID(sync_ddl_log());
650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
  DBUG_RETURN(error);
}


/*
  Rename partitions as part of ALTER TABLE of partitions

  SYNOPSIS
    rename_partitions()
    path                        Complete path of db and table name

  RETURN VALUE
    TRUE                        Failure
    FALSE                       Success

  DESCRIPTION
    When reorganising partitions, adding hash partitions and coalescing
    partitions it can be necessary to rename partitions while holding
    an exclusive lock on the table.
    Which partitions to rename is given by state of partitions found by the
    partition info struct referenced from the handler object
*/

int ha_partition::rename_partitions(const char *path)
{
  List_iterator<partition_element> part_it(m_part_info->partitions);
  List_iterator<partition_element> temp_it(m_part_info->temp_partitions);
  char part_name_buff[FN_REFLEN];
  char norm_name_buff[FN_REFLEN];
  uint no_parts= m_part_info->partitions.elements;
  uint part_count= 0;
  uint no_subparts= m_part_info->no_subparts;
  uint i= 0;
  uint j= 0;
684
  int error= 0;
685
  int ret_error;
686 687 688 689 690
  uint temp_partitions= m_part_info->temp_partitions.elements;
  handler *file;
  partition_element *part_elem, *sub_elem;
  DBUG_ENTER("ha_partition::rename_partitions");

691 692 693 694 695 696 697
  /*
    Assert that it works without HA_FILE_BASED and lower_case_table_name = 2.
    We use m_file[0] as long as all partitions have the same storage engine.
  */
  DBUG_ASSERT(!strcmp(path, get_canonical_filename(m_file[0], path,
                                                   norm_name_buff)));

698 699
  if (temp_partitions)
  {
700 701 702 703 704 705 706 707
    /*
      These are the reorganised partitions that have already been copied.
      We delete the partitions and log the delete by inactivating the
      delete log entry in the table log. We only need to synchronise
      these writes before moving to the next loop since there is no
      interaction among reorganised partitions, they cannot have the
      same name.
    */
708 709 710 711 712 713 714 715 716 717 718 719 720 721
    do
    {
      part_elem= temp_it++;
      if (m_is_sub_partitioned)
      {
        List_iterator<partition_element> sub_it(part_elem->subpartitions);
        do
        {
          sub_elem= sub_it++;
          file= m_reorged_file[part_count++];
          create_subpartition_name(norm_name_buff, path,
                                   part_elem->partition_name,
                                   sub_elem->partition_name,
                                   NORMAL_PART_NAME);
722
          DBUG_PRINT("info", ("Delete subpartition %s", norm_name_buff));
723
          if ((ret_error= file->ha_delete_table(norm_name_buff)))
724
            error= ret_error;
725
          else if (deactivate_ddl_log_entry(sub_elem->log_entry->entry_pos))
726 727 728
            error= 1;
          else
            sub_elem->log_entry= NULL; /* Indicate success */
729 730 731 732 733 734 735 736
        } while (++j < no_subparts);
      }
      else
      {
        file= m_reorged_file[part_count++];
        create_partition_name(norm_name_buff, path,
                              part_elem->partition_name, NORMAL_PART_NAME,
                              TRUE);
737
        DBUG_PRINT("info", ("Delete partition %s", norm_name_buff));
738
        if ((ret_error= file->ha_delete_table(norm_name_buff)))
739
          error= ret_error;
740
        else if (deactivate_ddl_log_entry(part_elem->log_entry->entry_pos))
741 742
          error= 1;
        else
743
          part_elem->log_entry= NULL; /* Indicate success */
744 745
      }
    } while (++i < temp_partitions);
746
    VOID(sync_ddl_log());
747 748 749 750
  }
  i= 0;
  do
  {
751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770
    /*
       When state is PART_IS_CHANGED it means that we have created a new
       TEMP partition that is to be renamed to normal partition name and
       we are to delete the old partition with currently the normal name.
       
       We perform this operation by
       1) Delete old partition with normal partition name
       2) Signal this in table log entry
       3) Synch table log to ensure we have consistency in crashes
       4) Rename temporary partition name to normal partition name
       5) Signal this to table log entry
       It is not necessary to synch the last state since a new rename
       should not corrupt things if there was no temporary partition.

       The only other parts we need to cater for are new parts that
       replace reorganised parts. The reorganised parts were deleted
       by the code above that goes through the temp_partitions list.
       Thus the synch above makes it safe to simply perform step 4 and 5
       for those entries.
    */
771 772
    part_elem= part_it++;
    if (part_elem->part_state == PART_IS_CHANGED ||
773
        part_elem->part_state == PART_TO_BE_DROPPED ||
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792
        (part_elem->part_state == PART_IS_ADDED && temp_partitions))
    {
      if (m_is_sub_partitioned)
      {
        List_iterator<partition_element> sub_it(part_elem->subpartitions);
        uint part;

        j= 0;
        do
        {
          sub_elem= sub_it++;
          part= i * no_subparts + j;
          create_subpartition_name(norm_name_buff, path,
                                   part_elem->partition_name,
                                   sub_elem->partition_name,
                                   NORMAL_PART_NAME);
          if (part_elem->part_state == PART_IS_CHANGED)
          {
            file= m_reorged_file[part_count++];
793
            DBUG_PRINT("info", ("Delete subpartition %s", norm_name_buff));
794
            if ((ret_error= file->ha_delete_table(norm_name_buff)))
795
              error= ret_error;
796
            else if (deactivate_ddl_log_entry(sub_elem->log_entry->entry_pos))
797
              error= 1;
798
            VOID(sync_ddl_log());
799 800 801 802 803 804 805 806
          }
          file= m_new_file[part];
          create_subpartition_name(part_name_buff, path,
                                   part_elem->partition_name,
                                   sub_elem->partition_name,
                                   TEMP_PART_NAME);
          DBUG_PRINT("info", ("Rename subpartition from %s to %s",
                     part_name_buff, norm_name_buff));
807 808
          if ((ret_error= file->ha_rename_table(part_name_buff,
                                                norm_name_buff)))
809
            error= ret_error;
810
          else if (deactivate_ddl_log_entry(sub_elem->log_entry->entry_pos))
811 812 813
            error= 1;
          else
            sub_elem->log_entry= NULL;
814 815 816 817 818 819 820 821 822 823
        } while (++j < no_subparts);
      }
      else
      {
        create_partition_name(norm_name_buff, path,
                              part_elem->partition_name, NORMAL_PART_NAME,
                              TRUE);
        if (part_elem->part_state == PART_IS_CHANGED)
        {
          file= m_reorged_file[part_count++];
824
          DBUG_PRINT("info", ("Delete partition %s", norm_name_buff));
825
          if ((ret_error= file->ha_delete_table(norm_name_buff)))
826
            error= ret_error;
827
          else if (deactivate_ddl_log_entry(part_elem->log_entry->entry_pos))
828
            error= 1;
829
          VOID(sync_ddl_log());
830 831 832 833 834 835 836
        }
        file= m_new_file[i];
        create_partition_name(part_name_buff, path,
                              part_elem->partition_name, TEMP_PART_NAME,
                              TRUE);
        DBUG_PRINT("info", ("Rename partition from %s to %s",
                   part_name_buff, norm_name_buff));
837 838
        if ((ret_error= file->ha_rename_table(part_name_buff,
                                              norm_name_buff)))
839
          error= ret_error;
840
        else if (deactivate_ddl_log_entry(part_elem->log_entry->entry_pos))
841 842 843
          error= 1;
        else
          part_elem->log_entry= NULL;
844 845 846
      }
    }
  } while (++i < no_parts);
847
  VOID(sync_ddl_log());
848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043
  DBUG_RETURN(error);
}


#define OPTIMIZE_PARTS 1
#define ANALYZE_PARTS 2
#define CHECK_PARTS   3
#define REPAIR_PARTS 4

/*
  Optimize table

  SYNOPSIS
    optimize()
    thd               Thread object
    check_opt         Check/analyze/repair/optimize options

  RETURN VALUES
    >0                Error
    0                 Success
*/

int ha_partition::optimize(THD *thd, HA_CHECK_OPT *check_opt)
{
  DBUG_ENTER("ha_partition::optimize");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    OPTIMIZE_PARTS, TRUE));
}


/*
  Analyze table

  SYNOPSIS
    analyze()
    thd               Thread object
    check_opt         Check/analyze/repair/optimize options

  RETURN VALUES
    >0                Error
    0                 Success
*/

int ha_partition::analyze(THD *thd, HA_CHECK_OPT *check_opt)
{
  DBUG_ENTER("ha_partition::analyze");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    ANALYZE_PARTS, TRUE));
}


/*
  Check table

  SYNOPSIS
    check()
    thd               Thread object
    check_opt         Check/analyze/repair/optimize options

  RETURN VALUES
    >0                Error
    0                 Success
*/

int ha_partition::check(THD *thd, HA_CHECK_OPT *check_opt)
{
  DBUG_ENTER("ha_partition::check");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    CHECK_PARTS, TRUE));
}


/*
  Repair table

  SYNOPSIS
    repair()
    thd               Thread object
    check_opt         Check/analyze/repair/optimize options

  RETURN VALUES
    >0                Error
    0                 Success
*/

int ha_partition::repair(THD *thd, HA_CHECK_OPT *check_opt)
{
  DBUG_ENTER("ha_partition::repair");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    REPAIR_PARTS, TRUE));
}

/*
  Optimize partitions

  SYNOPSIS
    optimize_partitions()
    thd                   Thread object
  RETURN VALUE
    >0                        Failure
    0                         Success
  DESCRIPTION
    Call optimize on each partition marked with partition state PART_CHANGED
*/

int ha_partition::optimize_partitions(THD *thd)
{
  DBUG_ENTER("ha_partition::optimize_partitions");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    OPTIMIZE_PARTS, FALSE));
}

/*
  Analyze partitions

  SYNOPSIS
    analyze_partitions()
    thd                   Thread object
  RETURN VALUE
    >0                        Failure
    0                         Success
  DESCRIPTION
    Call analyze on each partition marked with partition state PART_CHANGED
*/

int ha_partition::analyze_partitions(THD *thd)
{
  DBUG_ENTER("ha_partition::analyze_partitions");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    ANALYZE_PARTS, FALSE));
}

/*
  Check partitions

  SYNOPSIS
    check_partitions()
    thd                   Thread object
  RETURN VALUE
    >0                        Failure
    0                         Success
  DESCRIPTION
    Call check on each partition marked with partition state PART_CHANGED
*/

int ha_partition::check_partitions(THD *thd)
{
  DBUG_ENTER("ha_partition::check_partitions");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    CHECK_PARTS, FALSE));
}

/*
  Repair partitions

  SYNOPSIS
    repair_partitions()
    thd                   Thread object
  RETURN VALUE
    >0                        Failure
    0                         Success
  DESCRIPTION
    Call repair on each partition marked with partition state PART_CHANGED
*/

int ha_partition::repair_partitions(THD *thd)
{
  DBUG_ENTER("ha_partition::repair_partitions");

  DBUG_RETURN(handle_opt_partitions(thd, &thd->lex->check_opt, 
                                    REPAIR_PARTS, FALSE));
}


/*
  Handle optimize/analyze/check/repair of one partition

  SYNOPSIS
    handle_opt_part()
    thd                      Thread object
    check_opt                Options
    file                     Handler object of partition
    flag                     Optimize/Analyze/Check/Repair flag

  RETURN VALUE
    >0                        Failure
    0                         Success
*/

mattiasj@witty's avatar
mattiasj@witty committed
1044
#ifdef WL4176_IS_DONE
1045 1046 1047 1048 1049 1050 1051
static int handle_opt_part(THD *thd, HA_CHECK_OPT *check_opt,
                           handler *file, uint flag)
{
  int error;
  DBUG_ENTER("handle_opt_part");
  DBUG_PRINT("enter", ("flag = %u", flag));

1052 1053 1054 1055 1056 1057
  /*
    TODO:
    Rewrite the code for ANALYZE/CHECK/OPTIMIZE/REPAIR PARTITION WL4176
  */
  DBUG_RETURN(HA_ADMIN_NOT_IMPLEMENTED);

1058
  if (flag == OPTIMIZE_PARTS)
1059
    error= file->ha_optimize(thd, check_opt);
1060
  else if (flag == ANALYZE_PARTS)
1061
    error= file->ha_analyze(thd, check_opt);
1062
  else if (flag == CHECK_PARTS)
1063
    error= file->ha_check(thd, check_opt);
1064
  else if (flag == REPAIR_PARTS)
1065
    error= file->ha_repair(thd, check_opt);
1066 1067 1068 1069 1070 1071 1072 1073 1074
  else
  {
    DBUG_ASSERT(FALSE);
    error= 1;
  }
  if (error == HA_ADMIN_ALREADY_DONE)
    error= 0;
  DBUG_RETURN(error);
}
mattiasj@witty's avatar
mattiasj@witty committed
1075
#endif
1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095


/*
  Handle optimize/analyze/check/repair of partitions

  SYNOPSIS
    handle_opt_partitions()
    thd                      Thread object
    check_opt                Options
    flag                     Optimize/Analyze/Check/Repair flag
    all_parts                All partitions or only a subset

  RETURN VALUE
    >0                        Failure
    0                         Success
*/

int ha_partition::handle_opt_partitions(THD *thd, HA_CHECK_OPT *check_opt,
                                        uint flag, bool all_parts)
{
mattiasj@witty's avatar
mattiasj@witty committed
1096
#ifdef WL4176_IS_DONE
1097 1098 1099 1100 1101
  List_iterator<partition_element> part_it(m_part_info->partitions);
  uint no_parts= m_part_info->no_parts;
  uint no_subparts= m_part_info->no_subparts;
  uint i= 0;
  int error;
mattiasj@witty's avatar
mattiasj@witty committed
1102
#endif
1103 1104 1105
  DBUG_ENTER("ha_partition::handle_opt_partitions");
  DBUG_PRINT("enter", ("all_parts %u, flag= %u", all_parts, flag));

1106 1107 1108 1109 1110 1111
  /*
    TODO:
    Rewrite the code for ANALYZE/CHECK/OPTIMIZE/REPAIR PARTITION WL4176
  */
  DBUG_RETURN(HA_ADMIN_NOT_IMPLEMENTED);
#ifdef WL4176_IS_DONE
1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126
  do
  {
    partition_element *part_elem= part_it++;
    if (all_parts || part_elem->part_state == PART_CHANGED)
    {
      if (m_is_sub_partitioned)
      {
        uint j= 0, part;
        do
        {
          part= i * no_subparts + j;
          DBUG_PRINT("info", ("Optimize subpartition %u",
                     part));
          if ((error= handle_opt_part(thd, check_opt, m_file[part], flag)))
          {
1127
            DBUG_RETURN(error);
1128 1129 1130 1131 1132 1133 1134 1135
          }
        } while (++j < no_subparts);
      }
      else
      {
        DBUG_PRINT("info", ("Optimize partition %u", i));
        if ((error= handle_opt_part(thd, check_opt, m_file[i], flag)))
        {
1136
          DBUG_RETURN(error);
1137 1138 1139 1140 1141
        }
      }
    }
  } while (++i < no_parts);
  DBUG_RETURN(FALSE);
1142
#endif
1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159
}

/*
  Prepare by creating a new partition

  SYNOPSIS
    prepare_new_partition()
    table                      Table object
    create_info                Create info from CREATE TABLE
    file                       Handler object of new partition
    part_name                  partition name

  RETURN VALUE
    >0                         Error
    0                          Success
*/

1160
int ha_partition::prepare_new_partition(TABLE *tbl,
1161
                                        HA_CREATE_INFO *create_info,
1162 1163
                                        handler *file, const char *part_name,
                                        partition_element *p_elem)
1164 1165 1166 1167 1168
{
  int error;
  bool create_flag= FALSE;
  DBUG_ENTER("prepare_new_partition");

1169
  if ((error= set_up_table_before_create(tbl, part_name, create_info,
1170 1171
                                         0, p_elem)))
    goto error;
1172
  if ((error= file->ha_create(part_name, tbl, create_info)))
1173 1174
    goto error;
  create_flag= TRUE;
1175
  if ((error= file->ha_open(tbl, part_name, m_mode, m_open_test_lock)))
1176
    goto error;
1177 1178 1179 1180 1181 1182
  /*
    Note: if you plan to add another call that may return failure,
    better to do it before external_lock() as cleanup_new_partition()
    assumes that external_lock() is last call that may fail here.
    Otherwise see description for cleanup_new_partition().
  */
1183
  if ((error= file->ha_external_lock(current_thd, m_lock_type)))
1184 1185 1186 1187 1188
    goto error;

  DBUG_RETURN(0);
error:
  if (create_flag)
1189
    VOID(file->ha_delete_table(part_name));
1190 1191 1192 1193 1194 1195
  DBUG_RETURN(error);
}


/*
  Cleanup by removing all created partitions after error
1196

1197 1198 1199 1200 1201 1202 1203 1204
  SYNOPSIS
    cleanup_new_partition()
    part_count             Number of partitions to remove

  RETURN VALUE
    NONE

  DESCRIPTION
1205 1206 1207 1208 1209 1210 1211 1212
    This function is called immediately after prepare_new_partition() in
    case the latter fails.

    In prepare_new_partition() last call that may return failure is
    external_lock(). That means if prepare_new_partition() fails,
    partition does not have external lock. Thus no need to call
    external_lock(F_UNLCK) here.

1213 1214 1215 1216 1217 1218
  TODO:
    We must ensure that in the case that we get an error during the process
    that we call external_lock with F_UNLCK, close the table and delete the
    table in the case where we have been successful with prepare_handler.
    We solve this by keeping an array of successful calls to prepare_handler
    which can then be used to undo the call.
1219 1220
*/

1221
void ha_partition::cleanup_new_partition(uint part_count)
1222
{
1223 1224
  handler **save_m_file= m_file;
  DBUG_ENTER("ha_partition::cleanup_new_partition");
1225

1226
  if (m_added_file && m_added_file[0])
1227
  {
1228 1229 1230 1231 1232 1233 1234 1235
    m_file= m_added_file;
    m_added_file= NULL;

    /* delete_table also needed, a bit more complex */
    close();

    m_added_file= m_file;
    m_file= save_m_file;
1236
  }
1237
  DBUG_VOID_RETURN;
1238 1239
}

1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272
/*
  Implement the partition changes defined by ALTER TABLE of partitions

  SYNOPSIS
    change_partitions()
    create_info                 HA_CREATE_INFO object describing all
                                fields and indexes in table
    path                        Complete path of db and table name
    out: copied                 Output parameter where number of copied
                                records are added
    out: deleted                Output parameter where number of deleted
                                records are added
    pack_frm_data               Reference to packed frm file
    pack_frm_len                Length of packed frm file

  RETURN VALUE
    >0                        Failure
    0                         Success

  DESCRIPTION
    Add and copy if needed a number of partitions, during this operation
    no other operation is ongoing in the server. This is used by
    ADD PARTITION all types as well as by REORGANIZE PARTITION. For
    one-phased implementations it is used also by DROP and COALESCE
    PARTITIONs.
    One-phased implementation needs the new frm file, other handlers will
    get zero length and a NULL reference here.
*/

int ha_partition::change_partitions(HA_CREATE_INFO *create_info,
                                    const char *path,
                                    ulonglong *copied,
                                    ulonglong *deleted,
1273
                                    const uchar *pack_frm_data
1274
                                    __attribute__((unused)),
1275
                                    size_t pack_frm_len
1276
                                    __attribute__((unused)))
1277 1278
{
  List_iterator<partition_element> part_it(m_part_info->partitions);
1279
  List_iterator <partition_element> t_it(m_part_info->temp_partitions);
1280
  char part_name_buff[FN_REFLEN];
1281 1282 1283
  uint no_parts= m_part_info->partitions.elements;
  uint no_subparts= m_part_info->no_subparts;
  uint i= 0;
1284
  uint no_remain_partitions, part_count, orig_count;
1285
  handler **new_file_array;
1286
  int error= 1;
1287 1288 1289 1290 1291
  bool first;
  uint temp_partitions= m_part_info->temp_partitions.elements;
  THD *thd= current_thd;
  DBUG_ENTER("ha_partition::change_partitions");

1292 1293 1294 1295 1296 1297
  /*
    Assert that it works without HA_FILE_BASED and lower_case_table_name = 2.
    We use m_file[0] as long as all partitions have the same storage engine.
  */
  DBUG_ASSERT(!strcmp(path, get_canonical_filename(m_file[0], path,
                                                   part_name_buff)));
1298
  m_reorged_parts= 0;
1299
  if (!m_part_info->is_sub_partitioned())
1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323
    no_subparts= 1;

  /*
    Step 1:
      Calculate number of reorganised partitions and allocate space for
      their handler references.
  */
  if (temp_partitions)
  {
    m_reorged_parts= temp_partitions * no_subparts;
  }
  else
  {
    do
    {
      partition_element *part_elem= part_it++;
      if (part_elem->part_state == PART_CHANGED ||
          part_elem->part_state == PART_REORGED_DROPPED)
      {
        m_reorged_parts+= no_subparts;
      }
    } while (++i < no_parts);
  }
  if (m_reorged_parts &&
1324
      !(m_reorged_file= (handler**)sql_calloc(sizeof(handler*)*
1325 1326
                                              (m_reorged_parts + 1))))
  {
1327
    mem_alloc_error(sizeof(handler*)*(m_reorged_parts+1));
1328
    DBUG_RETURN(ER_OUTOFMEMORY);
1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359
  }

  /*
    Step 2:
      Calculate number of partitions after change and allocate space for
      their handler references.
  */
  no_remain_partitions= 0;
  if (temp_partitions)
  {
    no_remain_partitions= no_parts * no_subparts;
  }
  else
  {
    part_it.rewind();
    i= 0;
    do
    {
      partition_element *part_elem= part_it++;
      if (part_elem->part_state == PART_NORMAL ||
          part_elem->part_state == PART_TO_BE_ADDED ||
          part_elem->part_state == PART_CHANGED)
      {
        no_remain_partitions+= no_subparts;
      }
    } while (++i < no_parts);
  }
  if (!(new_file_array= (handler**)sql_calloc(sizeof(handler*)*
                                              (2*(no_remain_partitions + 1)))))
  {
    mem_alloc_error(sizeof(handler*)*2*(no_remain_partitions+1));
1360
    DBUG_RETURN(ER_OUTOFMEMORY);
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 1392 1393 1394 1395 1396 1397
  }
  m_added_file= &new_file_array[no_remain_partitions + 1];

  /*
    Step 3:
      Fill m_reorged_file with handler references and NULL at the end
  */
  if (m_reorged_parts)
  {
    i= 0;
    part_count= 0;
    first= TRUE;
    part_it.rewind();
    do
    {
      partition_element *part_elem= part_it++;
      if (part_elem->part_state == PART_CHANGED ||
          part_elem->part_state == PART_REORGED_DROPPED)
      {
        memcpy((void*)&m_reorged_file[part_count],
               (void*)&m_file[i*no_subparts],
               sizeof(handler*)*no_subparts);
        part_count+= no_subparts;
      }
      else if (first && temp_partitions &&
               part_elem->part_state == PART_TO_BE_ADDED)
      {
        /*
          When doing an ALTER TABLE REORGANIZE PARTITION a number of
          partitions is to be reorganised into a set of new partitions.
          The reorganised partitions are in this case in the temp_partitions
          list. We copy all of them in one batch and thus we only do this
          until we find the first partition with state PART_TO_BE_ADDED
          since this is where the new partitions go in and where the old
          ones used to be.
        */
        first= FALSE;
1398
        DBUG_ASSERT(((i*no_subparts) + m_reorged_parts) <= m_file_tot_parts);
1399
        memcpy((void*)m_reorged_file, &m_file[i*no_subparts],
1400
               sizeof(handler*)*m_reorged_parts);
1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411
      }
    } while (++i < no_parts);
  }

  /*
    Step 4:
      Fill new_array_file with handler references. Create the handlers if
      needed.
  */
  i= 0;
  part_count= 0;
1412
  orig_count= 0;
1413
  first= TRUE;
1414 1415 1416 1417 1418 1419
  part_it.rewind();
  do
  {
    partition_element *part_elem= part_it++;
    if (part_elem->part_state == PART_NORMAL)
    {
1420 1421
      DBUG_ASSERT(orig_count + no_subparts <= m_file_tot_parts);
      memcpy((void*)&new_file_array[part_count], (void*)&m_file[orig_count],
1422 1423
             sizeof(handler*)*no_subparts);
      part_count+= no_subparts;
1424
      orig_count+= no_subparts;
1425 1426 1427 1428 1429 1430 1431
    }
    else if (part_elem->part_state == PART_CHANGED ||
             part_elem->part_state == PART_TO_BE_ADDED)
    {
      uint j= 0;
      do
      {
1432 1433 1434 1435
        if (!(new_file_array[part_count++]=
              get_new_handler(table->s,
                              thd->mem_root,
                              part_elem->engine_type)))
1436 1437
        {
          mem_alloc_error(sizeof(handler));
1438
          DBUG_RETURN(ER_OUTOFMEMORY);
1439 1440
        }
      } while (++j < no_subparts);
1441 1442 1443 1444 1445 1446 1447
      if (part_elem->part_state == PART_CHANGED)
        orig_count+= no_subparts;
      else if (temp_partitions && first)
      {
        orig_count+= (no_subparts * temp_partitions);
        first= FALSE;
      }
1448 1449
    }
  } while (++i < no_parts);
1450
  first= FALSE;
1451 1452 1453 1454 1455 1456 1457 1458 1459
  /*
    Step 5:
      Create the new partitions and also open, lock and call external_lock
      on them to prepare them for copy phase and also for later close
      calls
  */
  i= 0;
  part_count= 0;
  part_it.rewind();
1460 1461 1462
  do
  {
    partition_element *part_elem= part_it++;
1463 1464
    if (part_elem->part_state == PART_TO_BE_ADDED ||
        part_elem->part_state == PART_CHANGED)
1465 1466
    {
      /*
1467 1468 1469
        A new partition needs to be created PART_TO_BE_ADDED means an
        entirely new partition and PART_CHANGED means a changed partition
        that will still exist with either more or less data in it.
1470
      */
1471 1472 1473 1474
      uint name_variant= NORMAL_PART_NAME;
      if (part_elem->part_state == PART_CHANGED ||
          (part_elem->part_state == PART_TO_BE_ADDED && temp_partitions))
        name_variant= TEMP_PART_NAME;
1475
      if (m_part_info->is_sub_partitioned())
1476 1477 1478 1479 1480 1481 1482 1483
      {
        List_iterator<partition_element> sub_it(part_elem->subpartitions);
        uint j= 0, part;
        do
        {
          partition_element *sub_elem= sub_it++;
          create_subpartition_name(part_name_buff, path,
                                   part_elem->partition_name,
1484 1485
                                   sub_elem->partition_name,
                                   name_variant);
1486
          part= i * no_subparts + j;
1487 1488 1489
          DBUG_PRINT("info", ("Add subpartition %s", part_name_buff));
          if ((error= prepare_new_partition(table, create_info,
                                            new_file_array[part],
1490 1491
                                            (const char *)part_name_buff,
                                            sub_elem)))
1492 1493
          {
            cleanup_new_partition(part_count);
1494
            DBUG_RETURN(error);
1495 1496
          }
          m_added_file[part_count++]= new_file_array[part];
1497 1498 1499 1500 1501
        } while (++j < no_subparts);
      }
      else
      {
        create_partition_name(part_name_buff, path,
1502 1503 1504 1505 1506
                              part_elem->partition_name, name_variant,
                              TRUE);
        DBUG_PRINT("info", ("Add partition %s", part_name_buff));
        if ((error= prepare_new_partition(table, create_info,
                                          new_file_array[i],
1507 1508
                                          (const char *)part_name_buff,
                                          part_elem)))
1509 1510
        {
          cleanup_new_partition(part_count);
1511
          DBUG_RETURN(error);
1512 1513
        }
        m_added_file[part_count++]= new_file_array[i];
1514 1515 1516
      }
    }
  } while (++i < no_parts);
1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 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 1560 1561 1562 1563 1564 1565 1566

  /*
    Step 6:
      State update to prepare for next write of the frm file.
  */
  i= 0;
  part_it.rewind();
  do
  {
    partition_element *part_elem= part_it++;
    if (part_elem->part_state == PART_TO_BE_ADDED)
      part_elem->part_state= PART_IS_ADDED;
    else if (part_elem->part_state == PART_CHANGED)
      part_elem->part_state= PART_IS_CHANGED;
    else if (part_elem->part_state == PART_REORGED_DROPPED)
      part_elem->part_state= PART_TO_BE_DROPPED;
  } while (++i < no_parts);
  for (i= 0; i < temp_partitions; i++)
  {
    partition_element *part_elem= t_it++;
    DBUG_ASSERT(part_elem->part_state == PART_TO_BE_REORGED);
    part_elem->part_state= PART_TO_BE_DROPPED;
  }
  m_new_file= new_file_array;
  DBUG_RETURN(copy_partitions(copied, deleted));
}


/*
  Copy partitions as part of ALTER TABLE of partitions

  SYNOPSIS
    copy_partitions()
    out:copied                 Number of records copied
    out:deleted                Number of records deleted

  RETURN VALUE
    >0                         Error code
    0                          Success

  DESCRIPTION
    change_partitions has done all the preparations, now it is time to
    actually copy the data from the reorganised partitions to the new
    partitions.
*/

int ha_partition::copy_partitions(ulonglong *copied, ulonglong *deleted)
{
  uint reorg_part= 0;
  int result= 0;
1567
  longlong func_value;
1568 1569
  DBUG_ENTER("ha_partition::copy_partitions");

1570 1571 1572 1573 1574 1575 1576 1577
  if (m_part_info->linear_hash_ind)
  {
    if (m_part_info->part_type == HASH_PARTITION)
      set_linear_hash_mask(m_part_info, m_part_info->no_parts);
    else
      set_linear_hash_mask(m_part_info, m_part_info->no_subparts);
  }

1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600
  while (reorg_part < m_reorged_parts)
  {
    handler *file= m_reorged_file[reorg_part];
    uint32 new_part;

    late_extra_cache(reorg_part);
    if ((result= file->ha_rnd_init(1)))
      goto error;
    while (TRUE)
    {
      if ((result= file->rnd_next(m_rec0)))
      {
        if (result == HA_ERR_RECORD_DELETED)
          continue;                              //Probably MyISAM
        if (result != HA_ERR_END_OF_FILE)
          goto error;
        /*
          End-of-file reached, break out to continue with next partition or
          end the copy process.
        */
        break;
      }
      /* Found record to insert into new handler */
1601 1602
      if (m_part_info->get_partition_id(m_part_info, &new_part,
                                        &func_value))
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612
      {
        /*
           This record is in the original table but will not be in the new
           table since it doesn't fit into any partition any longer due to
           changed partitioning ranges or list values.
        */
        deleted++;
      }
      else
      {
1613
        THD *thd= ha_thd();
1614 1615
        /* Copy record to new handler */
        copied++;
1616 1617 1618 1619
        tmp_disable_binlog(thd); /* Do not replicate the low-level changes. */
        result= m_new_file[new_part]->ha_write_row(m_rec0);
        reenable_binlog(thd);
        if (result)
1620 1621 1622 1623
          goto error;
      }
    }
    late_extra_no_cache(reorg_part);
1624
    file->ha_rnd_end();
1625 1626 1627 1628
    reorg_part++;
  }
  DBUG_RETURN(FALSE);
error:
1629
  DBUG_RETURN(result);
1630
}
1631

1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646

/*
  Update create info as part of ALTER TABLE

  SYNOPSIS
    update_create_info()
    create_info                   Create info from ALTER TABLE

  RETURN VALUE
    NONE

  DESCRIPTION
    Method empty so far
*/

1647 1648
void ha_partition::update_create_info(HA_CREATE_INFO *create_info)
{
1649 1650 1651 1652 1653
  info(HA_STATUS_AUTO);

  if (!(create_info->used_fields & HA_CREATE_USED_AUTO))
    create_info->auto_increment_value= stats.auto_increment_value;

1654
  create_info->data_file_name= create_info->index_file_name = NULL;
1655 1656 1657 1658
  return;
}


1659 1660 1661 1662 1663 1664 1665 1666 1667
void ha_partition::change_table_ptr(TABLE *table_arg, TABLE_SHARE *share)
{
  handler **file_array= m_file;
  table= table_arg;
  table_share= share;
  do
  {
    (*file_array)->change_table_ptr(table_arg, share);
  } while (*(++file_array));
1668 1669 1670 1671 1672 1673 1674 1675 1676
  if (m_added_file && m_added_file[0])
  {
    /* if in middle of a drop/rename etc */
    file_array= m_added_file;
    do
    {
      (*file_array)->change_table_ptr(table_arg, share);
    } while (*(++file_array));
  }
1677 1678
}

1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692
/*
  Change comments specific to handler

  SYNOPSIS
    update_table_comment()
    comment                       Original comment

  RETURN VALUE
    new comment 

  DESCRIPTION
    No comment changes so far
*/

1693 1694
char *ha_partition::update_table_comment(const char *comment)
{
1695
  return (char*) comment;                       /* Nothing to change */
1696 1697 1698 1699 1700
}



/*
1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720
  Handle delete, rename and create table

  SYNOPSIS
    del_ren_cre_table()
    from                    Full path of old table
    to                      Full path of new table
    table_arg               Table object
    create_info             Create info

  RETURN VALUE
    >0                      Error
    0                       Success

  DESCRIPTION
    Common routine to handle delete_table and rename_table.
    The routine uses the partition handler file to get the
    names of the partition instances. Both these routines
    are called after creating the handler without table
    object and thus the file is needed to discover the
    names of the partitions and the underlying storage engines.
1721 1722 1723 1724 1725 1726 1727
*/

uint ha_partition::del_ren_cre_table(const char *from,
				     const char *to,
				     TABLE *table_arg,
				     HA_CREATE_INFO *create_info)
{
1728 1729
  int save_error= 0;
  int error;
1730 1731
  char from_buff[FN_REFLEN], to_buff[FN_REFLEN], from_lc_buff[FN_REFLEN],
       to_lc_buff[FN_REFLEN];
1732
  char *name_buffer_ptr;
1733
  const char *from_path, *to_path;
1734
  uint i;
1735
  handler **file, **abort_file;
1736 1737
  DBUG_ENTER("del_ren_cre_table()");

1738
  if (get_from_handler_file(from, current_thd->mem_root))
1739 1740
    DBUG_RETURN(TRUE);
  DBUG_ASSERT(m_file_buffer);
1741
  DBUG_PRINT("enter", ("from: (%s) to: (%s)", from, to));
1742 1743
  name_buffer_ptr= m_name_buffer_ptr;
  file= m_file;
1744 1745 1746 1747 1748 1749 1750 1751 1752 1753
  /*
    Since ha_partition has HA_FILE_BASED, it must alter underlying table names
    if they do not have HA_FILE_BASED and lower_case_table_names == 2.
    See Bug#37402, for Mac OS X.
    The appended #P#<partname>[#SP#<subpartname>] will remain in current case.
    Using the first partitions handler, since mixing handlers is not allowed.
  */
  from_path= get_canonical_filename(*file, from, from_lc_buff);
  if (to != NULL)
    to_path= get_canonical_filename(*file, to, to_lc_buff);
1754 1755 1756
  i= 0;
  do
  {
1757 1758 1759
    create_partition_name(from_buff, from_path, name_buffer_ptr,
                          NORMAL_PART_NAME, FALSE);

1760 1761
    if (to != NULL)
    {						// Rename branch
1762 1763
      create_partition_name(to_buff, to_path, name_buffer_ptr,
                            NORMAL_PART_NAME, FALSE);
1764
      error= (*file)->ha_rename_table(from_buff, to_buff);
1765 1766
    }
    else if (table_arg == NULL)			// delete branch
1767
      error= (*file)->ha_delete_table(from_buff);
1768 1769
    else
    {
1770 1771
      if ((error= set_up_table_before_create(table_arg, from_buff,
                                             create_info, i, NULL)) ||
1772
          ((error= (*file)->ha_create(from_buff, table_arg, create_info))))
1773
        goto create_error;
1774 1775 1776 1777 1778 1779 1780
    }
    name_buffer_ptr= strend(name_buffer_ptr) + 1;
    if (error)
      save_error= error;
    i++;
  } while (*(++file));
  DBUG_RETURN(save_error);
1781 1782 1783 1784
create_error:
  name_buffer_ptr= m_name_buffer_ptr;
  for (abort_file= file, file= m_file; file < abort_file; file++)
  {
1785
    create_partition_name(from_buff, from_path, name_buffer_ptr, NORMAL_PART_NAME,
1786
                          FALSE);
1787
    VOID((*file)->ha_delete_table((const char*) from_buff));
1788 1789 1790
    name_buffer_ptr= strend(name_buffer_ptr) + 1;
  }
  DBUG_RETURN(error);
1791 1792
}

1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803
/*
  Find partition based on partition id

  SYNOPSIS
    find_partition_element()
    part_id                   Partition id of partition looked for

  RETURN VALUE
    >0                        Reference to partition_element
    0                         Partition not found
*/
1804 1805 1806 1807 1808

partition_element *ha_partition::find_partition_element(uint part_id)
{
  uint i;
  uint curr_part_id= 0;
1809
  List_iterator_fast <partition_element> part_it(m_part_info->partitions);
1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829

  for (i= 0; i < m_part_info->no_parts; i++)
  {
    partition_element *part_elem;
    part_elem= part_it++;
    if (m_is_sub_partitioned)
    {
      uint j;
      List_iterator_fast <partition_element> sub_it(part_elem->subpartitions);
      for (j= 0; j < m_part_info->no_subparts; j++)
      {
	part_elem= sub_it++;
	if (part_id == curr_part_id++)
	  return part_elem;
      }
    }
    else if (part_id == curr_part_id++)
      return part_elem;
  }
  DBUG_ASSERT(0);
1830
  my_error(ER_OUT_OF_RESOURCES, MYF(0));
1831 1832 1833 1834 1835
  current_thd->fatal_error();                   // Abort
  return NULL;
}


1836 1837 1838 1839 1840 1841 1842 1843 1844 1845
/*
   Set up table share object before calling create on underlying handler

   SYNOPSIS
     set_up_table_before_create()
     table                       Table object
     info                        Create info
     part_id                     Partition id of partition to set-up

   RETURN VALUE
1846 1847
     TRUE                        Error
     FALSE                       Success
1848 1849 1850 1851 1852 1853 1854 1855 1856

   DESCRIPTION
     Set up
     1) Comment on partition
     2) MAX_ROWS, MIN_ROWS on partition
     3) Index file name on partition
     4) Data file name on partition
*/

1857
int ha_partition::set_up_table_before_create(TABLE *tbl,
1858 1859 1860 1861
                    const char *partition_name_with_path, 
                    HA_CREATE_INFO *info,
                    uint part_id,
                    partition_element *part_elem)
1862
{
1863
  int error= 0;
1864 1865
  const char *partition_name;
  THD *thd= current_thd;
1866 1867
  DBUG_ENTER("set_up_table_before_create");

1868
  if (!part_elem)
1869 1870 1871
  {
    part_elem= find_partition_element(part_id);
    if (!part_elem)
1872
      DBUG_RETURN(1);                             // Fatal error
1873
  }
1874 1875
  tbl->s->max_rows= part_elem->part_max_rows;
  tbl->s->min_rows= part_elem->part_min_rows;
1876
  partition_name= strrchr(partition_name_with_path, FN_LIBCHAR);
1877
  if ((part_elem->index_file_name &&
1878
      (error= append_file_to_dir(thd,
1879 1880 1881
                                 (const char**)&part_elem->index_file_name,
                                 partition_name+1))) ||
      (part_elem->data_file_name &&
1882
      (error= append_file_to_dir(thd,
1883 1884 1885 1886 1887
                                 (const char**)&part_elem->data_file_name,
                                 partition_name+1))))
  {
    DBUG_RETURN(error);
  }
1888 1889
  info->index_file_name= part_elem->index_file_name;
  info->data_file_name= part_elem->data_file_name;
1890
  DBUG_RETURN(0);
1891 1892 1893 1894
}


/*
1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911
  Add two names together

  SYNOPSIS
    name_add()
    out:dest                          Destination string
    first_name                        First name
    sec_name                          Second name

  RETURN VALUE
    >0                                Error
    0                                 Success

  DESCRIPTION
    Routine used to add two names with '_' in between then. Service routine
    to create_handler_file
    Include the NULL in the count of characters since it is needed as separator
    between the partition names.
1912 1913 1914 1915
*/

static uint name_add(char *dest, const char *first_name, const char *sec_name)
{
1916
  return (uint) (strxmov(dest, first_name, "#SP#", sec_name, NullS) -dest) + 1;
1917 1918 1919 1920
}


/*
1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933
  Create the special .par file

  SYNOPSIS
    create_handler_file()
    name                      Full path of table name

  RETURN VALUE
    >0                        Error code
    0                         Success

  DESCRIPTION
    Method used to create handler file with names of partitions, their
    engine types and the number of partitions.
1934 1935 1936 1937 1938 1939
*/

bool ha_partition::create_handler_file(const char *name)
{
  partition_element *part_elem, *subpart_elem;
  uint i, j, part_name_len, subpart_name_len;
1940 1941
  uint tot_partition_words, tot_name_len, no_parts;
  uint tot_parts= 0;
1942 1943 1944 1945 1946
  uint tot_len_words, tot_len_byte, chksum, tot_name_words;
  char *name_buffer_ptr;
  uchar *file_buffer, *engine_array;
  bool result= TRUE;
  char file_name[FN_REFLEN];
1947 1948
  char part_name[FN_REFLEN];
  char subpart_name[FN_REFLEN];
1949
  File file;
1950
  List_iterator_fast <partition_element> part_it(m_part_info->partitions);
1951 1952
  DBUG_ENTER("create_handler_file");

1953 1954 1955
  no_parts= m_part_info->partitions.elements;
  DBUG_PRINT("info", ("table name = %s, no_parts = %u", name,
                      no_parts));
1956
  tot_name_len= 0;
1957
  for (i= 0; i < no_parts; i++)
1958 1959
  {
    part_elem= part_it++;
1960
    if (part_elem->part_state != PART_NORMAL &&
1961 1962
        part_elem->part_state != PART_TO_BE_ADDED &&
        part_elem->part_state != PART_CHANGED)
1963 1964 1965 1966
      continue;
    tablename_to_filename(part_elem->partition_name, part_name,
                          FN_REFLEN);
    part_name_len= strlen(part_name);
1967
    if (!m_is_sub_partitioned)
1968
    {
1969
      tot_name_len+= part_name_len + 1;
1970 1971
      tot_parts++;
    }
1972 1973
    else
    {
1974
      List_iterator_fast <partition_element> sub_it(part_elem->subpartitions);
1975 1976 1977
      for (j= 0; j < m_part_info->no_subparts; j++)
      {
	subpart_elem= sub_it++;
1978 1979 1980 1981 1982 1983
        tablename_to_filename(subpart_elem->partition_name,
                              subpart_name,
                              FN_REFLEN);
	subpart_name_len= strlen(subpart_name);
	tot_name_len+= part_name_len + subpart_name_len + 5;
        tot_parts++;
1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
      }
    }
  }
  /*
     File format:
     Length in words              4 byte
     Checksum                     4 byte
     Total number of partitions   4 byte
     Array of engine types        n * 4 bytes where
     n = (m_tot_parts + 3)/4
     Length of name part in bytes 4 bytes
     Name part                    m * 4 bytes where
     m = ((length_name_part + 3)/4)*4

     All padding bytes are zeroed
  */
2000
  tot_partition_words= (tot_parts + 3) / 4;
2001 2002 2003 2004 2005 2006 2007 2008
  tot_name_words= (tot_name_len + 3) / 4;
  tot_len_words= 4 + tot_partition_words + tot_name_words;
  tot_len_byte= 4 * tot_len_words;
  if (!(file_buffer= (uchar *) my_malloc(tot_len_byte, MYF(MY_ZEROFILL))))
    DBUG_RETURN(TRUE);
  engine_array= (file_buffer + 12);
  name_buffer_ptr= (char*) (file_buffer + ((4 + tot_partition_words) * 4));
  part_it.rewind();
2009
  for (i= 0; i < no_parts; i++)
2010 2011
  {
    part_elem= part_it++;
2012
    if (part_elem->part_state != PART_NORMAL &&
2013 2014
        part_elem->part_state != PART_TO_BE_ADDED &&
        part_elem->part_state != PART_CHANGED)
2015
      continue;
2016 2017
    if (!m_is_sub_partitioned)
    {
2018 2019
      tablename_to_filename(part_elem->partition_name, part_name, FN_REFLEN);
      name_buffer_ptr= strmov(name_buffer_ptr, part_name)+1;
2020
      *engine_array= (uchar) ha_legacy_type(part_elem->engine_type);
2021 2022 2023 2024 2025
      DBUG_PRINT("info", ("engine: %u", *engine_array));
      engine_array++;
    }
    else
    {
2026
      List_iterator_fast <partition_element> sub_it(part_elem->subpartitions);
2027 2028 2029
      for (j= 0; j < m_part_info->no_subparts; j++)
      {
	subpart_elem= sub_it++;
2030 2031 2032 2033
        tablename_to_filename(part_elem->partition_name, part_name,
                              FN_REFLEN);
        tablename_to_filename(subpart_elem->partition_name, subpart_name,
                              FN_REFLEN);
2034
	name_buffer_ptr+= name_add(name_buffer_ptr,
2035 2036
				   part_name,
				   subpart_name);
2037 2038
        *engine_array= (uchar) ha_legacy_type(subpart_elem->engine_type);
        DBUG_PRINT("info", ("engine: %u", *engine_array));
2039 2040 2041 2042 2043 2044
	engine_array++;
      }
    }
  }
  chksum= 0;
  int4store(file_buffer, tot_len_words);
2045
  int4store(file_buffer + 8, tot_parts);
2046 2047 2048 2049 2050 2051 2052 2053 2054
  int4store(file_buffer + 12 + (tot_partition_words * 4), tot_name_len);
  for (i= 0; i < tot_len_words; i++)
    chksum^= uint4korr(file_buffer + 4 * i);
  int4store(file_buffer + 4, chksum);
  /*
    Remove .frm extension and replace with .par
    Create and write and close file
    to be used at open, delete_table and rename_table
  */
2055
  fn_format(file_name, name, "", ha_par_ext, MY_APPEND_EXT);
2056 2057 2058
  if ((file= my_create(file_name, CREATE_MODE, O_RDWR | O_TRUNC,
		       MYF(MY_WME))) >= 0)
  {
2059
    result= my_write(file, (uchar *) file_buffer, tot_len_byte,
2060
                     MYF(MY_WME | MY_NABP)) != 0;
2061 2062 2063 2064 2065 2066 2067 2068
    VOID(my_close(file, MYF(0)));
  }
  else
    result= TRUE;
  my_free((char*) file_buffer, MYF(0));
  DBUG_RETURN(result);
}

2069 2070 2071 2072 2073 2074 2075 2076 2077
/*
  Clear handler variables and free some memory

  SYNOPSIS
    clear_handler_file()

  RETURN VALUE 
    NONE
*/
2078 2079 2080

void ha_partition::clear_handler_file()
{
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2081 2082
  if (m_engine_array)
    plugin_unlock_list(NULL, m_engine_array, m_tot_parts);
2083
  my_free((char*) m_file_buffer, MYF(MY_ALLOW_ZERO_PTR));
2084
  my_free((char*) m_engine_array, MYF(MY_ALLOW_ZERO_PTR));
2085 2086 2087 2088
  m_file_buffer= NULL;
  m_engine_array= NULL;
}

2089 2090 2091 2092 2093
/*
  Create underlying handler objects

  SYNOPSIS
    create_handlers()
2094
    mem_root		Allocate memory through this
2095 2096 2097 2098 2099

  RETURN VALUE
    TRUE                  Error
    FALSE                 Success
*/
2100

2101
bool ha_partition::create_handlers(MEM_ROOT *mem_root)
2102 2103 2104
{
  uint i;
  uint alloc_len= (m_tot_parts + 1) * sizeof(handler*);
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2105
  handlerton *hton0;
2106 2107
  DBUG_ENTER("create_handlers");

2108
  if (!(m_file= (handler **) alloc_root(mem_root, alloc_len)))
2109
    DBUG_RETURN(TRUE);
2110
  m_file_tot_parts= m_tot_parts;
2111
  bzero((char*) m_file, alloc_len);
2112 2113
  for (i= 0; i < m_tot_parts; i++)
  {
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2114
    handlerton *hton= plugin_data(m_engine_array[i], handlerton*);
2115
    if (!(m_file[i]= get_new_handler(table_share, mem_root,
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2116
                                     hton)))
2117
      DBUG_RETURN(TRUE);
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2118
    DBUG_PRINT("info", ("engine_type: %u", hton->db_type));
2119 2120
  }
  /* For the moment we only support partition over the same table engine */
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2121 2122
  hton0= plugin_data(m_engine_array[0], handlerton*);
  if (hton0 == myisam_hton)
2123 2124 2125 2126
  {
    DBUG_PRINT("info", ("MyISAM"));
    m_myisam= TRUE;
  }
2127
  /* INNODB may not be compiled in... */
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2128
  else if (ha_legacy_type(hton0) == DB_TYPE_INNODB)
2129 2130 2131 2132 2133 2134 2135
  {
    DBUG_PRINT("info", ("InnoDB"));
    m_innodb= TRUE;
  }
  DBUG_RETURN(FALSE);
}

2136 2137 2138 2139 2140
/*
  Create underlying handler objects from partition info

  SYNOPSIS
    new_handlers_from_part_info()
2141
    mem_root		Allocate memory through this
2142 2143 2144 2145 2146

  RETURN VALUE
    TRUE                  Error
    FALSE                 Success
*/
2147

2148
bool ha_partition::new_handlers_from_part_info(MEM_ROOT *mem_root)
2149
{
2150
  uint i, j, part_count;
2151 2152 2153 2154 2155
  partition_element *part_elem;
  uint alloc_len= (m_tot_parts + 1) * sizeof(handler*);
  List_iterator_fast <partition_element> part_it(m_part_info->partitions);
  DBUG_ENTER("ha_partition::new_handlers_from_part_info");

2156
  if (!(m_file= (handler **) alloc_root(mem_root, alloc_len)))
2157 2158 2159 2160
  {
    mem_alloc_error(alloc_len);
    goto error_end;
  }
2161
  m_file_tot_parts= m_tot_parts;
2162
  bzero((char*) m_file, alloc_len);
2163 2164 2165
  DBUG_ASSERT(m_part_info->no_parts > 0);

  i= 0;
2166
  part_count= 0;
2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177
  /*
    Don't know the size of the underlying storage engine, invent a number of
    bytes allocated for error message if allocation fails
  */
  do
  {
    part_elem= part_it++;
    if (m_is_sub_partitioned)
    {
      for (j= 0; j < m_part_info->no_subparts; j++)
      {
2178 2179
	if (!(m_file[part_count++]= get_new_handler(table_share, mem_root,
                                                    part_elem->engine_type)))
2180
          goto error;
2181 2182
	DBUG_PRINT("info", ("engine_type: %u",
                   (uint) ha_legacy_type(part_elem->engine_type)));
2183 2184
      }
    }
2185 2186
    else
    {
2187
      if (!(m_file[part_count++]= get_new_handler(table_share, mem_root,
2188 2189 2190 2191 2192
                                                  part_elem->engine_type)))
        goto error;
      DBUG_PRINT("info", ("engine_type: %u",
                 (uint) ha_legacy_type(part_elem->engine_type)));
    }
2193
  } while (++i < m_part_info->no_parts);
2194
  if (part_elem->engine_type == myisam_hton)
2195 2196 2197 2198 2199 2200
  {
    DBUG_PRINT("info", ("MyISAM"));
    m_myisam= TRUE;
  }
  DBUG_RETURN(FALSE);
error:
2201 2202
  mem_alloc_error(sizeof(handler));
error_end:
2203 2204 2205 2206 2207
  DBUG_RETURN(TRUE);
}


/*
2208 2209 2210 2211 2212
  Get info about partition engines and their names from the .par file

  SYNOPSIS
    get_from_handler_file()
    name                        Full path of table name
2213
    mem_root			Allocate memory through this
2214 2215 2216 2217 2218 2219 2220 2221

  RETURN VALUE
    TRUE                        Error
    FALSE                       Success

  DESCRIPTION
    Open handler file to get partition names, engine types and number of
    partitions.
2222 2223
*/

2224
bool ha_partition::get_from_handler_file(const char *name, MEM_ROOT *mem_root)
2225 2226 2227 2228
{
  char buff[FN_REFLEN], *address_tot_name_len;
  File file;
  char *file_buffer, *name_buffer_ptr;
2229
  handlerton **engine_array;
2230 2231 2232 2233 2234 2235
  uint i, len_bytes, len_words, tot_partition_words, tot_name_words, chksum;
  DBUG_ENTER("ha_partition::get_from_handler_file");
  DBUG_PRINT("enter", ("table name: '%s'", name));

  if (m_file_buffer)
    DBUG_RETURN(FALSE);
2236
  fn_format(buff, name, "", ha_par_ext, MY_APPEND_EXT);
2237 2238 2239 2240

  /* Following could be done with my_stat to read in whole file */
  if ((file= my_open(buff, O_RDONLY | O_SHARE, MYF(0))) < 0)
    DBUG_RETURN(TRUE);
2241
  if (my_read(file, (uchar *) & buff[0], 8, MYF(MY_NABP)))
2242 2243 2244
    goto err1;
  len_words= uint4korr(buff);
  len_bytes= 4 * len_words;
2245
  if (!(file_buffer= (char*) my_malloc(len_bytes, MYF(0))))
2246 2247
    goto err1;
  VOID(my_seek(file, 0, MY_SEEK_SET, MYF(0)));
2248
  if (my_read(file, (uchar *) file_buffer, len_bytes, MYF(MY_NABP)))
2249 2250 2251 2252 2253 2254 2255 2256
    goto err2;

  chksum= 0;
  for (i= 0; i < len_words; i++)
    chksum ^= uint4korr((file_buffer) + 4 * i);
  if (chksum)
    goto err2;
  m_tot_parts= uint4korr((file_buffer) + 8);
2257
  DBUG_PRINT("info", ("No of parts = %u", m_tot_parts));
2258
  tot_partition_words= (m_tot_parts + 3) / 4;
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2259
  engine_array= (handlerton **) my_alloca(m_tot_parts * sizeof(handlerton*));
2260
  for (i= 0; i < m_tot_parts; i++)
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2261
    engine_array[i]= ha_resolve_by_legacy_type(current_thd,
2262 2263
                                               (enum legacy_db_type)
                                               *(uchar *) ((file_buffer) + 12 + i));
2264 2265 2266
  address_tot_name_len= file_buffer + 12 + 4 * tot_partition_words;
  tot_name_words= (uint4korr(address_tot_name_len) + 3) / 4;
  if (len_words != (tot_partition_words + tot_name_words + 4))
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2267
    goto err3;
2268 2269 2270 2271
  name_buffer_ptr= file_buffer + 16 + 4 * tot_partition_words;
  VOID(my_close(file, MYF(0)));
  m_file_buffer= file_buffer;          // Will be freed in clear_handler_file()
  m_name_buffer_ptr= name_buffer_ptr;
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2272 2273 2274
  
  if (!(m_engine_array= (plugin_ref*)
                my_malloc(m_tot_parts * sizeof(plugin_ref), MYF(MY_WME))))
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2275
    goto err3;
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2276 2277 2278

  for (i= 0; i < m_tot_parts; i++)
    m_engine_array[i]= ha_lock_engine(NULL, engine_array[i]);
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2279 2280

  my_afree((gptr) engine_array);
antony@ppcg5.local's avatar
antony@ppcg5.local committed
2281
    
2282
  if (!m_file && create_handlers(mem_root))
2283 2284 2285 2286 2287 2288
  {
    clear_handler_file();
    DBUG_RETURN(TRUE);
  }
  DBUG_RETURN(FALSE);

antony@ppcg5.local's avatar
antony@ppcg5.local committed
2289 2290
err3:
  my_afree((gptr) engine_array);
2291 2292 2293 2294 2295 2296 2297
err2:
  my_free(file_buffer, MYF(0));
err1:
  VOID(my_close(file, MYF(0)));
  DBUG_RETURN(TRUE);
}

2298

2299 2300 2301 2302
/****************************************************************************
                MODULE open/close object
****************************************************************************/
/*
2303
  Open handler object
2304

2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322
  SYNOPSIS
    open()
    name                  Full path of table name
    mode                  Open mode flags
    test_if_locked        ?

  RETURN VALUE
    >0                    Error
    0                     Success

  DESCRIPTION
    Used for opening tables. The name will be the name of the file.
    A table is opened when it needs to be opened. For instance
    when a request comes in for a select on the table (tables are not
    open and closed for each request, they are cached).

    Called from handler.cc by handler::ha_open(). The server opens all tables
    by calling ha_open() which then calls the handler specific open().
2323 2324 2325 2326 2327
*/

int ha_partition::open(const char *name, int mode, uint test_if_locked)
{
  char *name_buffer_ptr= m_name_buffer_ptr;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2328
  int error;
2329
  uint alloc_len;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2330 2331
  handler **file;
  char name_buff[FN_REFLEN];
2332 2333 2334
  DBUG_ENTER("ha_partition::open");

  ref_length= 0;
2335 2336
  m_mode= mode;
  m_open_test_lock= test_if_locked;
2337
  m_part_field_array= m_part_info->full_part_field_array;
2338
  if (get_from_handler_file(name, &table->mem_root))
2339 2340 2341 2342
    DBUG_RETURN(1);
  m_start_key.length= 0;
  m_rec0= table->record[0];
  m_rec_length= table->s->reclength;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2343
  alloc_len= m_tot_parts * (m_rec_length + PARTITION_BYTES_IN_POS);
2344 2345 2346
  alloc_len+= table->s->max_key_length;
  if (!m_ordered_rec_buffer)
  {
2347
    if (!(m_ordered_rec_buffer= (uchar*)my_malloc(alloc_len, MYF(MY_WME))))
2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358
    {
      DBUG_RETURN(1);
    }
    {
      /*
        We set-up one record per partition and each record has 2 bytes in
        front where the partition id is written. This is used by ordered
        index_read.
        We also set-up a reference to the first record for temporary use in
        setting up the scan.
      */
2359
      char *ptr= (char*)m_ordered_rec_buffer;
2360 2361 2362 2363 2364 2365
      uint i= 0;
      do
      {
        int2store(ptr, i);
        ptr+= m_rec_length + PARTITION_BYTES_IN_POS;
      } while (++i < m_tot_parts);
2366
      m_start_key.key= (const uchar*)ptr;
2367 2368
    }
  }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2369 2370

  /* Initialise the bitmap we use to determine what partitions are used */
2371 2372 2373 2374 2375 2376
  if (!is_clone)
  {
    if (bitmap_init(&(m_part_info->used_partitions), NULL, m_tot_parts, TRUE))
      DBUG_RETURN(1);
    bitmap_set_all(&(m_part_info->used_partitions));
  }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2377

2378
  /* Recalculate table flags as they may change after open */
2379
  m_table_flags= m_file[0]->ha_table_flags();
2380 2381 2382
  file= m_file;
  do
  {
2383 2384
    create_partition_name(name_buff, name, name_buffer_ptr, NORMAL_PART_NAME,
                          FALSE);
2385
    if ((error= (*file)->ha_open(table, (const char*) name_buff, mode,
2386 2387
                                 test_if_locked)))
      goto err_handler;
patg@krsna.patg.net's avatar
patg@krsna.patg.net committed
2388
    m_no_locks+= (*file)->lock_count();
2389 2390
    name_buffer_ptr+= strlen(name_buffer_ptr) + 1;
    set_if_bigger(ref_length, ((*file)->ref_length));
2391
    m_table_flags&= (*file)->ha_table_flags();
2392
  } while (*(++file));
2393 2394 2395
  m_table_flags&= ~(HA_CAN_GEOMETRY | HA_CAN_FULLTEXT | HA_DUPLICATE_POS |
                    HA_CAN_SQL_HANDLER | HA_CAN_INSERT_DELAYED);
  m_table_flags|= HA_FILE_BASED | HA_REC_NOT_IN_SEQ;
2396 2397
  key_used_on_scan= m_file[0]->key_used_on_scan;
  implicit_emptied= m_file[0]->implicit_emptied;
2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411
  /*
    Add 2 bytes for partition id in position ref length.
    ref_length=max_in_all_partitions(ref_length) + PARTITION_BYTES_IN_POS
  */
  ref_length+= PARTITION_BYTES_IN_POS;
  m_ref_length= ref_length;
  /*
    Release buffer read from .par file. It will not be reused again after
    being opened once.
  */
  clear_handler_file();
  /*
    Initialise priority queue, initialised to reading forward.
  */
2412
  if ((error= init_queue(&m_queue, m_tot_parts, (uint) PARTITION_BYTES_IN_POS,
2413 2414
                         0, key_rec_cmp, (void*)this)))
    goto err_handler;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2415

2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427
  /*
    Some handlers update statistics as part of the open call. This will in
    some cases corrupt the statistics of the partition handler and thus
    to ensure we have correct statistics we call info from open after
    calling open on all individual handlers.
  */
  info(HA_STATUS_VARIABLE | HA_STATUS_CONST);
  DBUG_RETURN(0);

err_handler:
  while (file-- != m_file)
    (*file)->close();
2428

2429 2430 2431
  DBUG_RETURN(error);
}

2432 2433
handler *ha_partition::clone(MEM_ROOT *mem_root)
{
serg@janus.mylan's avatar
serg@janus.mylan committed
2434 2435
  handler *new_handler= get_new_handler(table->s, mem_root,
                                        table->s->db_type());
2436 2437 2438 2439 2440 2441 2442 2443 2444 2445
  ((ha_partition*)new_handler)->m_part_info= m_part_info;
  ((ha_partition*)new_handler)->is_clone= TRUE;
  if (new_handler && !new_handler->ha_open(table,
                                           table->s->normalized_path.str,
                                           table->db_stat,
                                           HA_OPEN_IGNORE_IF_LOCKED))
    return new_handler;
  return NULL;
}

2446

2447
/*
2448
  Close handler object
2449

2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462
  SYNOPSIS
    close()

  RETURN VALUE
    >0                   Error code
    0                    Success

  DESCRIPTION
    Called from sql_base.cc, sql_select.cc, and table.cc.
    In sql_select.cc it is only used to close up temporary tables or during
    the process where a temporary table is converted over to being a
    myisam table.
    For sql_base.cc look at close_data_tables().
2463 2464 2465 2466
*/

int ha_partition::close(void)
{
2467
  bool first= TRUE;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2468
  handler **file;
2469
  DBUG_ENTER("ha_partition::close");
2470

2471
  delete_queue(&m_queue);
2472 2473
  if (!is_clone)
    bitmap_free(&(m_part_info->used_partitions));
2474
  file= m_file;
2475 2476

repeat:
2477 2478 2479 2480
  do
  {
    (*file)->close();
  } while (*(++file));
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2481

2482 2483 2484 2485 2486 2487
  if (first && m_added_file && m_added_file[0])
  {
    file= m_added_file;
    first= FALSE;
    goto repeat;
  }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2488

2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501
  DBUG_RETURN(0);
}

/****************************************************************************
                MODULE start/end statement
****************************************************************************/
/*
  A number of methods to define various constants for the handler. In
  the case of the partition handler we need to use some max and min
  of the underlying handlers in most cases.
*/

/*
2502
  Set external locks on table
2503

2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529
  SYNOPSIS
    external_lock()
    thd                    Thread object
    lock_type              Type of external lock

  RETURN VALUE
    >0                   Error code
    0                    Success

  DESCRIPTION
    First you should go read the section "locking functions for mysql" in
    lock.cc to understand this.
    This create a lock on the table. If you are implementing a storage engine
    that can handle transactions look at ha_berkeley.cc to see how you will
    want to go about doing this. Otherwise you should consider calling
    flock() here.
    Originally this method was used to set locks on file level to enable
    several MySQL Servers to work on the same data. For transactional
    engines it has been "abused" to also mean start and end of statements
    to enable proper rollback of statements and transactions. When LOCK
    TABLES has been issued the start_stmt method takes over the role of
    indicating start of statement but in this case there is no end of
    statement indicator(?).

    Called from lock.cc by lock_external() and unlock_external(). Also called
    from sql_table.cc by copy_data_between_tables().
2530 2531 2532 2533
*/

int ha_partition::external_lock(THD *thd, int lock_type)
{
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2534
  bool first= TRUE;
2535 2536 2537
  uint error;
  handler **file;
  DBUG_ENTER("ha_partition::external_lock");
2538

2539
  file= m_file;
2540 2541 2542
  m_lock_type= lock_type;

repeat:
2543 2544
  do
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2545
    DBUG_PRINT("info", ("external_lock(thd, %d) iteration %d",
2546
                        lock_type, (int) (file - m_file)));
2547
    if ((error= (*file)->ha_external_lock(thd, lock_type)))
2548
    {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2549 2550
      if (F_UNLCK != lock_type)
        goto err_handler;
2551 2552
    }
  } while (*(++file));
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2553

2554 2555 2556 2557 2558 2559 2560
  if (first && m_added_file && m_added_file[0])
  {
    DBUG_ASSERT(lock_type == F_UNLCK);
    file= m_added_file;
    first= FALSE;
    goto repeat;
  }
2561 2562 2563 2564
  DBUG_RETURN(0);

err_handler:
  while (file-- != m_file)
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2565
  {
2566
    (*file)->ha_external_lock(thd, F_UNLCK);
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2567
  }
2568 2569 2570 2571 2572
  DBUG_RETURN(error);
}


/*
2573
  Get the lock(s) for the table and perform conversion of locks if needed
2574

2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615
  SYNOPSIS
    store_lock()
    thd                   Thread object
    to                    Lock object array
    lock_type             Table lock type

  RETURN VALUE
    >0                   Error code
    0                    Success

  DESCRIPTION
    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).

    Berkeley DB for partition  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() is also called. In this case one
    usually doesn't have to do anything.

    store_lock is called when holding a global mutex to ensure that only
    one thread at a time changes the locking information of tables.

    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 as last
    time).  In the future we will probably try to remove this.

    Called from lock.cc by get_lock_data().
2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626
*/

THR_LOCK_DATA **ha_partition::store_lock(THD *thd,
					 THR_LOCK_DATA **to,
					 enum thr_lock_type lock_type)
{
  handler **file;
  DBUG_ENTER("ha_partition::store_lock");
  file= m_file;
  do
  {
2627
    DBUG_PRINT("info", ("store lock %d iteration", (int) (file - m_file)));
2628 2629 2630 2631 2632
    to= (*file)->store_lock(thd, to, lock_type);
  } while (*(++file));
  DBUG_RETURN(to);
}

2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648
/*
  Start a statement when table is locked

  SYNOPSIS
    start_stmt()
    thd                  Thread object
    lock_type            Type of external lock

  RETURN VALUE
    >0                   Error code
    0                    Success

  DESCRIPTION
    This method is called instead of external lock when the table is locked
    before the statement is executed.
*/
2649

tomas@poseidon.ndb.mysql.com's avatar
merge  
tomas@poseidon.ndb.mysql.com committed
2650
int ha_partition::start_stmt(THD *thd, thr_lock_type lock_type)
2651 2652 2653 2654
{
  int error= 0;
  handler **file;
  DBUG_ENTER("ha_partition::start_stmt");
2655

2656 2657 2658
  file= m_file;
  do
  {
tomas@poseidon.ndb.mysql.com's avatar
merge  
tomas@poseidon.ndb.mysql.com committed
2659
    if ((error= (*file)->start_stmt(thd, lock_type)))
2660 2661 2662 2663 2664 2665 2666
      break;
  } while (*(++file));
  DBUG_RETURN(error);
}


/*
2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679
  Get number of lock objects returned in store_lock

  SYNOPSIS
    lock_count()

  RETURN VALUE
    Number of locks returned in call to store_lock

  DESCRIPTION
    Returns the number of store locks needed in call to store lock.
    We return number of partitions since we call store_lock on each
    underlying handler. Assists the above functions in allocating
    sufficient space for lock structures.
2680 2681 2682 2683 2684
*/

uint ha_partition::lock_count() const
{
  DBUG_ENTER("ha_partition::lock_count");
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
2685
  DBUG_PRINT("info", ("m_no_locks %d", m_no_locks));
patg@krsna.patg.net's avatar
patg@krsna.patg.net committed
2686
  DBUG_RETURN(m_no_locks);
2687 2688 2689 2690
}


/*
2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701
  Unlock last accessed row

  SYNOPSIS
    unlock_row()

  RETURN VALUE
    NONE

  DESCRIPTION
    Record currently processed was not in the result set of the statement
    and is thus unlocked. Used for UPDATE and DELETE queries.
2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715
*/

void ha_partition::unlock_row()
{
  m_file[m_last_part]->unlock_row();
  return;
}


/****************************************************************************
                MODULE change record
****************************************************************************/

/*
2716
  Insert a row to the table
2717

2718 2719 2720
  SYNOPSIS
    write_row()
    buf                        The row in MySQL Row Format
2721

2722 2723 2724 2725 2726 2727 2728
  RETURN VALUE
    >0                         Error code
    0                          Success

  DESCRIPTION
    write_row() inserts a row. buf() is a byte array of data, normally
    record[0].
2729

2730 2731
    You can use the field information to extract the data from the native byte
    array type.
2732

2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744
    Example of this would be:
    for (Field **field=table->field ; *field ; field++)
    {
      ...
    }

    See ha_tina.cc for a variant of extracting all of the data as strings.
    ha_berkeley.cc has a variant of how to store it intact by "packing" it
    for ha_berkeley's own native storage type.

    Called from item_sum.cc, item_sum.cc, sql_acl.cc, sql_insert.cc,
    sql_insert.cc, sql_select.cc, sql_table.cc, sql_udf.cc, and sql_update.cc.
2745

2746
    ADDITIONAL INFO:
2747

2748 2749
    We have to set timestamp fields and auto_increment fields, because those
    may be used in determining which partition the row should be written to.
2750 2751
*/

2752
int ha_partition::write_row(uchar * buf)
2753 2754 2755
{
  uint32 part_id;
  int error;
2756
  longlong func_value;
2757 2758
  bool autoincrement_lock= FALSE;
  my_bitmap_map *old_map;
2759
  THD *thd= ha_thd();
2760
#ifdef NOT_NEEDED
2761
  uchar *rec0= m_rec0;
2762 2763 2764 2765
#endif
  DBUG_ENTER("ha_partition::write_row");
  DBUG_ASSERT(buf == m_rec0);

2766 2767 2768 2769 2770 2771 2772 2773 2774
  /* If we have a timestamp column, update it to the current time */
  if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT)
    table->timestamp_field->set_time();

  /*
    If we have an auto_increment column and we are writing a changed row
    or a new row, then update the auto_increment value in the record.
  */
  if (table->next_number_field && buf == table->record[0])
2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785
  {
    /*
      Some engines (InnoDB for example) can change autoincrement
      counter only after 'table->write_row' operation.
      So if another thread gets inside the ha_partition::write_row
      before it is complete, it gets same auto_increment value,
      which means DUP_KEY error (bug #27405)
      Here we separate the access using table_share->mutex, and
      use autoincrement_lock variable to avoid unnecessary locks.
      Probably not an ideal solution.
    */
2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796
    if (table_share->tmp_table == NO_TMP_TABLE)
    {
      /*
        Bug#30878 crash when alter table from non partitioned table
        to partitioned.
        Checking if tmp table then there is no need to lock,
        and the table_share->mutex may not be initialised.
      */
      autoincrement_lock= TRUE;
      pthread_mutex_lock(&table_share->mutex);
    }
antony@pcg5ppc.xiphis.org's avatar
antony@pcg5ppc.xiphis.org committed
2797 2798 2799 2800 2801 2802 2803 2804
    error= update_auto_increment();

    /*
      If we have failed to set the auto-increment value for this row,
      it is highly likely that we will not be able to insert it into
      the correct partition. We must check and fail if neccessary.
    */
    if (error)
2805
      goto exit;
2806
  }
2807

2808
  old_map= dbug_tmp_use_all_columns(table, table->read_set);
2809 2810 2811
#ifdef NOT_NEEDED
  if (likely(buf == rec0))
#endif
2812 2813
    error= m_part_info->get_partition_id(m_part_info, &part_id,
                                         &func_value);
2814 2815 2816 2817
#ifdef NOT_NEEDED
  else
  {
    set_field_ptr(m_part_field_array, buf, rec0);
2818 2819
    error= m_part_info->get_partition_id(m_part_info, &part_id,
                                         &func_value);
2820 2821 2822
    set_field_ptr(m_part_field_array, rec0, buf);
  }
#endif
2823
  dbug_tmp_restore_column_map(table->read_set, old_map);
2824
  if (unlikely(error))
2825 2826
  {
    m_part_info->err_value= func_value;
2827
    goto exit;
2828
  }
2829 2830
  m_last_part= part_id;
  DBUG_PRINT("info", ("Insert in partition %d", part_id));
2831 2832 2833
  tmp_disable_binlog(thd); /* Do not replicate the low-level changes. */
  error= m_file[part_id]->ha_write_row(buf);
  reenable_binlog(thd);
2834 2835 2836 2837
exit:
  if (autoincrement_lock)
    pthread_mutex_unlock(&table_share->mutex);
  DBUG_RETURN(error);
2838 2839 2840 2841
}


/*
2842
  Update an existing row
2843

2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869
  SYNOPSIS
    update_row()
    old_data                 Old record in MySQL Row Format
    new_data                 New record in MySQL Row Format

  RETURN VALUE
    >0                         Error code
    0                          Success

  DESCRIPTION
    Yes, update_row() does what you expect, it updates a row. old_data will
    have the previous row record in it, while new_data will have the newest
    data in it.
    Keep in mind that the server can do updates based on ordering if an
    ORDER BY clause was used. Consecutive ordering is not guarenteed.

    Currently new_data will not have an updated auto_increament record, or
    and updated timestamp field. You can do these for partition by doing these:
    if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_UPDATE)
      table->timestamp_field->set_time();
    if (table->next_number_field && record == table->record[0])
      update_auto_increment();

    Called from sql_select.cc, sql_acl.cc, sql_update.cc, and sql_insert.cc.
    new_data is always record[0]
    old_data is normally record[1] but may be anything
2870 2871
*/

2872
int ha_partition::update_row(const uchar *old_data, uchar *new_data)
2873
{
2874
  THD *thd= ha_thd();
2875
  uint32 new_part_id, old_part_id;
2876
  int error= 0;
2877
  longlong func_value;
2878
  timestamp_auto_set_type orig_timestamp_type= table->timestamp_field_type;
2879 2880
  DBUG_ENTER("ha_partition::update_row");

2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891
  /*
    We need to set timestamp field once before we calculate
    the partition. Then we disable timestamp calculations
    inside m_file[*]->update_row() methods
  */
  if (orig_timestamp_type & TIMESTAMP_AUTO_SET_ON_UPDATE)
  {
    table->timestamp_field->set_time();
    table->timestamp_field_type= TIMESTAMP_NO_AUTO_SET;
  }

2892
  if ((error= get_parts_for_update(old_data, new_data, table->record[0],
2893 2894
                                   m_part_info, &old_part_id, &new_part_id,
                                   &func_value)))
2895
  {
2896
    m_part_info->err_value= func_value;
2897
    goto exit;
2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908
  }

  /*
    TODO:
      set_internal_auto_increment=
        max(set_internal_auto_increment, new_data->auto_increment)
  */
  m_last_part= new_part_id;
  if (new_part_id == old_part_id)
  {
    DBUG_PRINT("info", ("Update in partition %d", new_part_id));
2909 2910 2911
    tmp_disable_binlog(thd); /* Do not replicate the low-level changes. */
    error= m_file[new_part_id]->ha_update_row(old_data, new_data);
    reenable_binlog(thd);
2912
    goto exit;
2913 2914 2915 2916 2917
  }
  else
  {
    DBUG_PRINT("info", ("Update from partition %d to partition %d",
			old_part_id, new_part_id));
2918 2919 2920 2921
    tmp_disable_binlog(thd); /* Do not replicate the low-level changes. */
    error= m_file[new_part_id]->ha_write_row(new_data);
    reenable_binlog(thd);
    if (error)
2922
      goto exit;
2923 2924 2925 2926 2927

    tmp_disable_binlog(thd); /* Do not replicate the low-level changes. */
    error= m_file[old_part_id]->ha_delete_row(old_data);
    reenable_binlog(thd);
    if (error)
2928 2929 2930 2931
    {
#ifdef IN_THE_FUTURE
      (void) m_file[new_part_id]->delete_last_inserted_row(new_data);
#endif
2932
      goto exit;
2933 2934
    }
  }
2935 2936 2937 2938

exit:
  table->timestamp_field_type= orig_timestamp_type;
  DBUG_RETURN(error);
2939 2940 2941 2942
}


/*
2943
  Remove an existing row
2944

2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967
  SYNOPSIS
    delete_row
    buf                      Deleted row in MySQL Row Format

  RETURN VALUE
    >0                       Error Code
    0                        Success

  DESCRIPTION
    This will delete a row. buf will contain a copy of the row to be deleted.
    The server will call this right after the current row has been read
    (from either a previous rnd_xxx() or index_xxx() call).
    If you keep a pointer to the last row or can access a primary key it will
    make doing the deletion quite a bit easier.
    Keep in mind that the server does no guarentee consecutive deletions.
    ORDER BY clauses can be used.

    Called in sql_acl.cc and sql_udf.cc to manage internal table information.
    Called in sql_delete.cc, sql_insert.cc, and sql_select.cc. In sql_select
    it is used for removing duplicates while in insert it is used for REPLACE
    calls.

    buf is either record[0] or record[1]
2968 2969
*/

2970
int ha_partition::delete_row(const uchar *buf)
2971 2972 2973
{
  uint32 part_id;
  int error;
2974
  THD *thd= ha_thd();
2975 2976 2977 2978 2979 2980 2981
  DBUG_ENTER("ha_partition::delete_row");

  if ((error= get_part_for_delete(buf, m_rec0, m_part_info, &part_id)))
  {
    DBUG_RETURN(error);
  }
  m_last_part= part_id;
2982 2983 2984 2985
  tmp_disable_binlog(thd);
  error= m_file[part_id]->ha_delete_row(buf);
  reenable_binlog(thd);
  DBUG_RETURN(error);
2986 2987 2988 2989
}


/*
2990
  Delete all rows in a table
2991

2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008
  SYNOPSIS
    delete_all_rows()

  RETURN VALUE
    >0                       Error Code
    0                        Success

  DESCRIPTION
    Used to delete all rows in a table. Both for cases of truncate and
    for cases where the optimizer realizes that all rows will be
    removed as a result of a SQL statement.

    Called from item_sum.cc by Item_func_group_concat::clear(),
    Item_sum_count_distinct::clear(), and Item_func_group_concat::clear().
    Called from sql_delete.cc by mysql_delete().
    Called from sql_select.cc by JOIN::reinit().
    Called from sql_union.cc by st_select_lex_unit::exec().
3009 3010 3011 3012 3013 3014 3015
*/

int ha_partition::delete_all_rows()
{
  int error;
  handler **file;
  DBUG_ENTER("ha_partition::delete_all_rows");
3016

3017 3018 3019
  file= m_file;
  do
  {
3020
    if ((error= (*file)->ha_delete_all_rows()))
3021 3022 3023 3024 3025
      DBUG_RETURN(error);
  } while (*(++file));
  DBUG_RETURN(0);
}

3026

3027
/*
3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038
  Start a large batch of insert rows

  SYNOPSIS
    start_bulk_insert()
    rows                  Number of rows to insert

  RETURN VALUE
    NONE

  DESCRIPTION
    rows == 0 means we will probably insert many rows
3039 3040 3041 3042 3043 3044
*/

void ha_partition::start_bulk_insert(ha_rows rows)
{
  handler **file;
  DBUG_ENTER("ha_partition::start_bulk_insert");
3045

3046
  rows= rows ? rows/m_tot_parts + 1 : 0;
3047 3048 3049
  file= m_file;
  do
  {
3050
    (*file)->ha_start_bulk_insert(rows);
3051 3052 3053 3054 3055
  } while (*(++file));
  DBUG_VOID_RETURN;
}


3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066
/*
  Finish a large batch of insert rows

  SYNOPSIS
    end_bulk_insert()

  RETURN VALUE
    >0                      Error code
    0                       Success
*/

3067 3068 3069 3070 3071 3072 3073 3074 3075 3076
int ha_partition::end_bulk_insert()
{
  int error= 0;
  handler **file;
  DBUG_ENTER("ha_partition::end_bulk_insert");

  file= m_file;
  do
  {
    int tmp;
3077
    if ((tmp= (*file)->ha_end_bulk_insert()))
3078 3079 3080 3081 3082
      error= tmp;
  } while (*(++file));
  DBUG_RETURN(error);
}

3083

3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094
/****************************************************************************
                MODULE full table scan
****************************************************************************/
/*
  Initialize engine for random reads

  SYNOPSIS
    ha_partition::rnd_init()
    scan	0  Initialize for random reads through rnd_pos()
		1  Initialize for random scan through rnd_next()

3095 3096 3097
  RETURN VALUE
    >0          Error code
    0           Success
3098

3099 3100 3101
  DESCRIPTION 
    rnd_init() is called when the server wants the storage engine to do a
    table scan or when the server wants to access data through rnd_pos.
3102

3103 3104 3105 3106 3107 3108 3109 3110
    When scan is used we will scan one handler partition at a time.
    When preparing for rnd_pos we will init all handler partitions.
    No extra cache handling is needed when scannning is not performed.

    Before initialising we will call rnd_end to ensure that we clean up from
    any previous incarnation of a table scan.
    Called from filesort.cc, records.cc, sql_handler.cc, sql_select.cc,
    sql_table.cc, and sql_update.cc.
3111 3112 3113 3114 3115
*/

int ha_partition::rnd_init(bool scan)
{
  int error;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3116 3117
  uint i= 0;
  uint32 part_id;
3118 3119
  DBUG_ENTER("ha_partition::rnd_init");

3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147
  /*
    For operations that may need to change data, we may need to extend
    read_set.
  */
  if (m_lock_type == F_WRLCK)
  {
    /*
      If write_set contains any of the fields used in partition and
      subpartition expression, we need to set all bits in read_set because
      the row may need to be inserted in a different [sub]partition. In
      other words update_row() can be converted into write_row(), which
      requires a complete record.
    */
    if (bitmap_is_overlapping(&m_part_info->full_part_field_set,
                              table->write_set))
      bitmap_set_all(table->read_set);
    else
    {
      /*
        Some handlers only read fields as specified by the bitmap for the
        read set. For partitioned handlers we always require that the
        fields of the partition functions are read such that we can
        calculate the partition id to place updated and deleted records.
      */
      bitmap_union(table->read_set, &m_part_info->full_part_field_set);
    }
  }

patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3148
  /* Now we see what the index of our first important partition is */
3149 3150
  DBUG_PRINT("info", ("m_part_info->used_partitions: 0x%lx",
                      (long) m_part_info->used_partitions.bitmap));
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3151 3152 3153 3154
  part_id= bitmap_get_first_set(&(m_part_info->used_partitions));
  DBUG_PRINT("info", ("m_part_spec.start_part %d", part_id));

  if (MY_BIT_NONE == part_id)
3155 3156
  {
    error= 0;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3157
    goto err1;
3158
  }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3159 3160 3161 3162 3163 3164

  /*
    We have a partition and we are scanning with rnd_next
    so we bump our cache
  */
  DBUG_PRINT("info", ("rnd_init on partition %d", part_id));
3165 3166 3167 3168 3169 3170 3171
  if (scan)
  {
    /*
      rnd_end() is needed for partitioning to reset internal data if scan
      is already in use
    */
    rnd_end();
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3172 3173 3174 3175 3176 3177 3178
    late_extra_cache(part_id);
    if ((error= m_file[part_id]->ha_rnd_init(scan)))
      goto err;
  }
  else
  {
    for (i= part_id; i < m_tot_parts; i++)
3179
    {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3180 3181 3182 3183 3184
      if (bitmap_is_set(&(m_part_info->used_partitions), i))
      {
        if ((error= m_file[i]->ha_rnd_init(scan)))
          goto err;
      }
3185 3186
    }
  }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3187 3188 3189 3190
  m_scan_value= scan;
  m_part_spec.start_part= part_id;
  m_part_spec.end_part= m_tot_parts - 1;
  DBUG_PRINT("info", ("m_scan_value=%d", m_scan_value));
3191 3192 3193
  DBUG_RETURN(0);

err:
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3194 3195 3196
  while ((int)--i >= (int)part_id)
  {
    if (bitmap_is_set(&(m_part_info->used_partitions), i))
3197
      m_file[i]->ha_rnd_end();
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3198 3199 3200 3201
  }
err1:
  m_scan_value= 2;
  m_part_spec.start_part= NO_CURRENT_PART_ID;
3202 3203 3204 3205
  DBUG_RETURN(error);
}


3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216
/*
  End of a table scan

  SYNOPSIS
    rnd_end()

  RETURN VALUE
    >0          Error code
    0           Success
*/

3217 3218 3219 3220 3221 3222 3223
int ha_partition::rnd_end()
{
  handler **file;
  DBUG_ENTER("ha_partition::rnd_end");
  switch (m_scan_value) {
  case 2:                                       // Error
    break;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3224 3225
  case 1:
    if (NO_CURRENT_PART_ID != m_part_spec.start_part)         // Table scan
3226 3227 3228 3229 3230 3231 3232 3233 3234
    {
      late_extra_no_cache(m_part_spec.start_part);
      m_file[m_part_spec.start_part]->ha_rnd_end();
    }
    break;
  case 0:
    file= m_file;
    do
    {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3235 3236
      if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
        (*file)->ha_rnd_end();
3237 3238 3239 3240
    } while (*(++file));
    break;
  }
  m_scan_value= 2;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3241
  m_part_spec.start_part= NO_CURRENT_PART_ID;
3242 3243 3244 3245 3246 3247 3248 3249 3250 3251
  DBUG_RETURN(0);
}

/*
  read next row during full table scan (scan in random row order)

  SYNOPSIS
    rnd_next()
    buf		buffer that should be filled with data

3252 3253 3254
  RETURN VALUE
    >0          Error code
    0           Success
3255

3256 3257 3258 3259 3260 3261 3262 3263
  DESCRIPTION
    This is called for each row of the table scan. When you run out of records
    you should return HA_ERR_END_OF_FILE.
    The Field structure for the table is the key to getting data into buf
    in a manner that will allow the server to understand it.

    Called from filesort.cc, records.cc, sql_handler.cc, sql_select.cc,
    sql_table.cc, and sql_update.cc.
3264 3265
*/

3266
int ha_partition::rnd_next(uchar *buf)
3267
{
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3268
  handler *file;
3269
  int result= HA_ERR_END_OF_FILE;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3270
  uint part_id= m_part_spec.start_part;
3271 3272
  DBUG_ENTER("ha_partition::rnd_next");

patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3273
  if (NO_CURRENT_PART_ID == part_id)
3274 3275 3276 3277 3278 3279 3280
  {
    /*
      The original set of partitions to scan was empty and thus we report
      the result here.
    */
    goto end;
  }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3281 3282 3283 3284
  
  DBUG_ASSERT(m_scan_value == 1);
  file= m_file[part_id];
  
3285 3286
  while (TRUE)
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3287 3288
    int result= file->rnd_next(buf);
    if (!result)
3289 3290
    {
      m_last_part= part_id;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3291
      m_part_spec.start_part= part_id;
3292 3293 3294
      table->status= 0;
      DBUG_RETURN(0);
    }
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3295 3296 3297 3298 3299 3300 3301 3302

    /*
      if we get here, then the current partition rnd_next returned failure
    */
    if (result == HA_ERR_RECORD_DELETED)
      continue;                               // Probably MyISAM

    if (result != HA_ERR_END_OF_FILE)
3303
      goto end_dont_reset_start_part;         // Return error
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324

    /* End current partition */
    late_extra_no_cache(part_id);
    DBUG_PRINT("info", ("rnd_end on partition %d", part_id));
    if ((result= file->ha_rnd_end()))
      break;
    
    /* Shift to next partition */
    while (++part_id < m_tot_parts &&
           !bitmap_is_set(&(m_part_info->used_partitions), part_id))
      ;
    if (part_id >= m_tot_parts)
    {
      result= HA_ERR_END_OF_FILE;
      break;
    }
    file= m_file[part_id];
    DBUG_PRINT("info", ("rnd_init on partition %d", part_id));
    if ((result= file->ha_rnd_init(1)))
      break;
    late_extra_cache(part_id);
3325 3326 3327 3328
  }

end:
  m_part_spec.start_part= NO_CURRENT_PART_ID;
3329
end_dont_reset_start_part:
3330 3331 3332 3333 3334
  table->status= STATUS_NOT_FOUND;
  DBUG_RETURN(result);
}


3335 3336
/*
  Save position of current row
3337

3338 3339 3340
  SYNOPSIS
    position()
    record             Current record in MySQL Row Format
3341

3342 3343
  RETURN VALUE
    NONE
3344

3345 3346 3347 3348 3349
  DESCRIPTION
    position() is called after each call to rnd_next() if the data needs
    to be ordered. You can do something like the following to store
    the position:
    ha_store_ptr(ref, ref_length, current_position);
3350

3351 3352 3353 3354 3355 3356 3357
    The server uses ref to store data. ref_length in the above case is
    the size needed to store current_position. ref is just a byte array
    that the server will maintain. If you are using offsets to mark rows, then
    current_position should be the offset. If it is a primary key like in
    BDB, then it needs to be a primary key.

    Called from filesort.cc, sql_select.cc, sql_delete.cc and sql_update.cc.
3358 3359
*/

3360
void ha_partition::position(const uchar *record)
3361
{
3362
  handler *file= m_file[m_last_part];
3363
  DBUG_ENTER("ha_partition::position");
3364

3365
  file->position(record);
3366
  int2store(ref, m_last_part);
3367 3368 3369 3370 3371
  memcpy((ref + PARTITION_BYTES_IN_POS), file->ref,
	 (ref_length - PARTITION_BYTES_IN_POS));

#ifdef SUPPORTING_PARTITION_OVER_DIFFERENT_ENGINES
#ifdef HAVE_purify
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3372 3373
  bzero(ref + PARTITION_BYTES_IN_POS + ref_length,
        max_ref_length-ref_length);
3374 3375 3376 3377 3378
#endif /* HAVE_purify */
#endif
  DBUG_VOID_RETURN;
}

3379 3380 3381 3382 3383 3384 3385 3386

void ha_partition::column_bitmaps_signal()
{
    handler::column_bitmaps_signal();
    bitmap_union(table->read_set, &m_part_info->full_part_field_set);
}
 

3387
/*
3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405
  Read row using position

  SYNOPSIS
    rnd_pos()
    out:buf                     Row read in MySQL Row Format
    position                    Position of read row

  RETURN VALUE
    >0                          Error code
    0                           Success

  DESCRIPTION
    This is like rnd_next, but you are given a position to use
    to determine the row. The position will be of the type that you stored in
    ref. You can use ha_get_ptr(pos,ref_length) to retrieve whatever key
    or position you saved when position() was called.
    Called from filesort.cc records.cc sql_insert.cc sql_select.cc
    sql_update.cc.
3406 3407
*/

3408
int ha_partition::rnd_pos(uchar * buf, uchar *pos)
3409 3410 3411 3412 3413
{
  uint part_id;
  handler *file;
  DBUG_ENTER("ha_partition::rnd_pos");

3414
  part_id= uint2korr((const uchar *) pos);
3415 3416 3417 3418 3419 3420 3421
  DBUG_ASSERT(part_id < m_tot_parts);
  file= m_file[part_id];
  m_last_part= part_id;
  DBUG_RETURN(file->rnd_pos(buf, (pos + PARTITION_BYTES_IN_POS)));
}


3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451
/*
  Read row using position using given record to find

  SYNOPSIS
    rnd_pos_by_record()
    record             Current record in MySQL Row Format

  RETURN VALUE
    >0                 Error code
    0                  Success

  DESCRIPTION
    this works as position()+rnd_pos() functions, but does some extra work,
    calculating m_last_part - the partition to where the 'record'
    should go.

    called from replication (log_event.cc)
*/

int ha_partition::rnd_pos_by_record(uchar *record)
{
  DBUG_ENTER("ha_partition::rnd_pos_by_record");

  if (unlikely(get_part_for_delete(record, m_rec0, m_part_info, &m_last_part)))
    DBUG_RETURN(1);

  DBUG_RETURN(handler::rnd_pos_by_record(record));
}


3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471
/****************************************************************************
                MODULE index scan
****************************************************************************/
/*
  Positions an index cursor to the index specified in the handle. Fetches the
  row if available. If the key value is null, begin at the first key of the
  index.

  There are loads of optimisations possible here for the partition handler.
  The same optimisations can also be checked for full table scan although
  only through conditions and not from index ranges.
  Phase one optimisations:
    Check if the fields of the partition function are bound. If so only use
    the single partition it becomes bound to.
  Phase two optimisations:
    If it can be deducted through range or list partitioning that only a
    subset of the partitions are used, then only use those partitions.
*/

/*
3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485
  Initialise handler before start of index scan

  SYNOPSIS
    index_init()
    inx                Index number
    sorted             Is rows to be returned in sorted order

  RETURN VALUE
    >0                 Error code
    0                  Success

  DESCRIPTION
    index_init is always called before starting index scans (except when
    starting through index_read_idx and using read_range variants).
3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498
*/

int ha_partition::index_init(uint inx, bool sorted)
{
  int error= 0;
  handler **file;
  DBUG_ENTER("ha_partition::index_init");

  active_index= inx;
  m_part_spec.start_part= NO_CURRENT_PART_ID;
  m_start_key.length= 0;
  m_ordered= sorted;
  m_curr_key_info= table->key_info+inx;
3499 3500 3501 3502 3503 3504 3505 3506 3507
  /*
    Some handlers only read fields as specified by the bitmap for the
    read set. For partitioned handlers we always require that the
    fields of the partition functions are read such that we can
    calculate the partition id to place updated and deleted records.
    But this is required for operations that may need to change data only.
  */
  if (m_lock_type == F_WRLCK)
    bitmap_union(table->read_set, &m_part_info->full_part_field_set);
3508
  else if (sorted)
3509 3510
  {
    /*
3511 3512 3513 3514 3515 3516 3517 3518
      An ordered scan is requested. We must make sure all fields of the 
      used index are in the read set, as partitioning requires them for
      sorting (see ha_partition::handle_ordered_index_scan).

      The SQL layer may request an ordered index scan without having index
      fields in the read set when
       - it needs to do an ordered scan over an index prefix.
       - it evaluates ORDER BY with SELECT COUNT(*) FROM t1.
3519 3520 3521 3522 3523 3524 3525 3526

      TODO: handle COUNT(*) queries via unordered scan.
    */
    uint i;
    for (i= 0; i < m_curr_key_info->key_parts; i++)
      bitmap_set_bit(table->read_set,
                     m_curr_key_info->key_part[i].field->field_index);
  }
3527 3528 3529 3530
  file= m_file;
  do
  {
    /* TODO RONM: Change to index_init() when code is stable */
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3531 3532 3533 3534 3535 3536
    if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
      if ((error= (*file)->ha_index_init(inx, sorted)))
      {
        DBUG_ASSERT(0);                           // Should never happen
        break;
      }
3537 3538 3539 3540 3541 3542
  } while (*(++file));
  DBUG_RETURN(error);
}


/*
3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554
  End of index scan

  SYNOPSIS
    index_end()

  RETURN VALUE
    >0                 Error code
    0                  Success

  DESCRIPTION
    index_end is called at the end of an index scan to clean up any
    things needed to clean up.
3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569
*/

int ha_partition::index_end()
{
  int error= 0;
  handler **file;
  DBUG_ENTER("ha_partition::index_end");

  active_index= MAX_KEY;
  m_part_spec.start_part= NO_CURRENT_PART_ID;
  file= m_file;
  do
  {
    int tmp;
    /* TODO RONM: Change to index_end() when code is stable */
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3570 3571 3572
    if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
      if ((tmp= (*file)->ha_index_end()))
        error= tmp;
3573 3574 3575 3576 3577 3578
  } while (*(++file));
  DBUG_RETURN(error);
}


/*
3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600
  Read one record in an index scan and start an index scan

  SYNOPSIS
    index_read()
    buf                    Read row in MySQL Row Format
    key                    Key parts in consecutive order
    key_len                Total length of key parts
    find_flag              What type of key condition is used

  RETURN VALUE
    >0                 Error code
    0                  Success

  DESCRIPTION
    index_read starts a new index scan using a start key. The MySQL Server
    will check the end key on its own. Thus to function properly the
    partitioned handler need to ensure that it delivers records in the sort
    order of the MySQL Server.
    index_read can be restarted without calling index_end on the previous
    index scan and without calling index_init. In this case the index_read
    is on the same index as the previous index_scan. This is particularly
    used in conjuntion with multi read ranges.
3601 3602
*/

3603 3604 3605
int ha_partition::index_read_map(uchar *buf, const uchar *key,
                                 key_part_map keypart_map,
                                 enum ha_rkey_function find_flag)
3606
{
3607
  DBUG_ENTER("ha_partition::index_read_map");
3608

3609
  end_range= 0;
3610
  m_index_scan_type= partition_index_read;
3611
  DBUG_RETURN(common_index_read(buf, key, keypart_map, find_flag));
3612 3613 3614
}


3615 3616 3617 3618 3619 3620 3621 3622 3623
/*
  Common routine for a number of index_read variants

  SYNOPSIS
    common_index_read
  
  see index_read for rest
*/

3624
int ha_partition::common_index_read(uchar *buf, const uchar *key,
3625
                                    key_part_map keypart_map,
3626 3627 3628
				    enum ha_rkey_function find_flag)
{
  int error;
3629
  bool reverse_order= FALSE;
3630
  uint key_len= calculate_key_len(table, active_index, key, keypart_map);
3631 3632 3633
  DBUG_ENTER("ha_partition::common_index_read");

  memcpy((void*)m_start_key.key, key, key_len);
3634
  m_start_key.keypart_map= keypart_map;
3635 3636 3637 3638 3639 3640 3641
  m_start_key.length= key_len;
  m_start_key.flag= find_flag;

  if ((error= partition_scan_set_up(buf, TRUE)))
  {
    DBUG_RETURN(error);
  }
3642 3643 3644 3645 3646 3647 3648
  if (find_flag == HA_READ_PREFIX_LAST ||
      find_flag == HA_READ_PREFIX_LAST_OR_PREV ||
      find_flag == HA_READ_BEFORE_KEY)
  {
    reverse_order= TRUE;
    m_ordered_scan_ongoing= TRUE;
  }
3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673
  if (!m_ordered_scan_ongoing ||
      (find_flag == HA_READ_KEY_EXACT &&
       (key_len >= m_curr_key_info->key_length ||
	key_len == 0)))
  {
    /*
      We use unordered index scan either when read_range is used and flag
      is set to not use ordered or when an exact key is used and in this
      case all records will be sorted equal and thus the sort order of the
      resulting records doesn't matter.
      We also use an unordered index scan when the number of partitions to
      scan is only one.
      The unordered index scan will use the partition set created.
      Need to set unordered scan ongoing since we can come here even when
      it isn't set.
    */
    m_ordered_scan_ongoing= FALSE;
    error= handle_unordered_scan_next_partition(buf);
  }
  else
  {
    /*
      In all other cases we will use the ordered index scan. This will use
      the partition set created by the get_partition_set method.
    */
3674
    error= handle_ordered_index_scan(buf, reverse_order);
3675 3676 3677 3678 3679 3680
  }
  DBUG_RETURN(error);
}


/*
3681 3682 3683 3684 3685
  Start an index scan from leftmost record and return first record

  SYNOPSIS
    index_first()
    buf                 Read row in MySQL Row Format
3686

3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698
  RETURN VALUE
    >0                  Error code
    0                   Success

  DESCRIPTION
    index_first() asks for the first key in the index.
    This is similar to index_read except that there is no start key since
    the scan starts from the leftmost entry and proceeds forward with
    index_next.

    Called from opt_range.cc, opt_sum.cc, sql_handler.cc,
    and sql_select.cc.
3699 3700
*/

3701
int ha_partition::index_first(uchar * buf)
3702 3703
{
  DBUG_ENTER("ha_partition::index_first");
3704

3705 3706 3707 3708 3709 3710 3711
  end_range= 0;
  m_index_scan_type= partition_index_first;
  DBUG_RETURN(common_first_last(buf));
}


/*
3712 3713 3714 3715 3716
  Start an index scan from rightmost record and return first record
  
  SYNOPSIS
    index_last()
    buf                 Read row in MySQL Row Format
3717

3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729
  RETURN VALUE
    >0                  Error code
    0                   Success

  DESCRIPTION
    index_last() asks for the last key in the index.
    This is similar to index_read except that there is no start key since
    the scan starts from the rightmost entry and proceeds forward with
    index_prev.

    Called from opt_range.cc, opt_sum.cc, sql_handler.cc,
    and sql_select.cc.
3730 3731
*/

3732
int ha_partition::index_last(uchar * buf)
3733 3734
{
  DBUG_ENTER("ha_partition::index_last");
3735

3736 3737 3738 3739
  m_index_scan_type= partition_index_last;
  DBUG_RETURN(common_first_last(buf));
}

3740 3741 3742 3743 3744 3745 3746 3747 3748
/*
  Common routine for index_first/index_last

  SYNOPSIS
    common_index_first_last
  
  see index_first for rest
*/

3749
int ha_partition::common_first_last(uchar *buf)
3750 3751
{
  int error;
3752

3753 3754
  if ((error= partition_scan_set_up(buf, FALSE)))
    return error;
3755 3756
  if (!m_ordered_scan_ongoing &&
      m_index_scan_type != partition_index_last)
3757
    return handle_unordered_scan_next_partition(buf);
3758
  return handle_ordered_index_scan(buf, FALSE);
3759 3760
}

3761

3762
/*
3763 3764 3765 3766 3767 3768
  Read last using key

  SYNOPSIS
    index_read_last()
    buf                   Read row in MySQL Row Format
    key                   Key
3769
    keypart_map           Which part of key is used
3770 3771 3772 3773 3774 3775 3776 3777

  RETURN VALUE
    >0                    Error code
    0                     Success

  DESCRIPTION
    This is used in join_read_last_key to optimise away an ORDER BY.
    Can only be used on indexes supporting HA_READ_ORDER
3778 3779
*/

3780 3781
int ha_partition::index_read_last_map(uchar *buf, const uchar *key,
                                      key_part_map keypart_map)
3782 3783
{
  DBUG_ENTER("ha_partition::index_read_last");
3784

3785
  m_ordered= TRUE;				// Safety measure
3786 3787
  end_range= 0;
  m_index_scan_type= partition_index_read_last;
3788
  DBUG_RETURN(common_index_read(buf, key, keypart_map, HA_READ_PREFIX_LAST));
3789 3790 3791 3792
}


/*
3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804
  Read next record in a forward index scan

  SYNOPSIS
    index_next()
    buf                   Read row in MySQL Row Format

  RETURN VALUE
    >0                    Error code
    0                     Success

  DESCRIPTION
    Used to read forward through the index.
3805 3806
*/

3807
int ha_partition::index_next(uchar * buf)
3808 3809
{
  DBUG_ENTER("ha_partition::index_next");
3810

3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825
  /*
    TODO(low priority):
    If we want partition to work with the HANDLER commands, we
    must be able to do index_last() -> index_prev() -> index_next()
  */
  DBUG_ASSERT(m_index_scan_type != partition_index_last);
  if (!m_ordered_scan_ongoing)
  {
    DBUG_RETURN(handle_unordered_next(buf, FALSE));
  }
  DBUG_RETURN(handle_ordered_next(buf, FALSE));
}


/*
3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840
  Read next record special

  SYNOPSIS
    index_next_same()
    buf                   Read row in MySQL Row Format
    key                   Key
    keylen                Length of key

  RETURN VALUE
    >0                    Error code
    0                     Success

  DESCRIPTION
    This routine is used to read the next but only if the key is the same
    as supplied in the call.
3841 3842
*/

3843
int ha_partition::index_next_same(uchar *buf, const uchar *key, uint keylen)
3844 3845
{
  DBUG_ENTER("ha_partition::index_next_same");
3846

3847 3848 3849 3850 3851 3852 3853
  DBUG_ASSERT(keylen == m_start_key.length);
  DBUG_ASSERT(m_index_scan_type != partition_index_last);
  if (!m_ordered_scan_ongoing)
    DBUG_RETURN(handle_unordered_next(buf, TRUE));
  DBUG_RETURN(handle_ordered_next(buf, TRUE));
}

3854

3855
/*
3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867
  Read next record when performing index scan backwards

  SYNOPSIS
    index_prev()
    buf                   Read row in MySQL Row Format

  RETURN VALUE
    >0                    Error code
    0                     Success

  DESCRIPTION
    Used to read backwards through the index.
3868 3869
*/

3870
int ha_partition::index_prev(uchar * buf)
3871 3872
{
  DBUG_ENTER("ha_partition::index_prev");
3873

3874 3875 3876 3877 3878 3879 3880
  /* TODO: read comment in index_next */
  DBUG_ASSERT(m_index_scan_type != partition_index_first);
  DBUG_RETURN(handle_ordered_prev(buf));
}


/*
3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898
  Start a read of one range with start and end key

  SYNOPSIS
    read_range_first()
    start_key           Specification of start key
    end_key             Specification of end key
    eq_range_arg        Is it equal range
    sorted              Should records be returned in sorted order

  RETURN VALUE
    >0                    Error code
    0                     Success

  DESCRIPTION
    We reimplement read_range_first since we don't want the compare_key
    check at the end. This is already performed in the partition handler.
    read_range_next is very much different due to that we need to scan
    all underlying handlers.
3899 3900 3901 3902 3903 3904 3905 3906
*/

int ha_partition::read_range_first(const key_range *start_key,
				   const key_range *end_key,
				   bool eq_range_arg, bool sorted)
{
  int error;
  DBUG_ENTER("ha_partition::read_range_first");
3907

3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922
  m_ordered= sorted;
  eq_range= eq_range_arg;
  end_range= 0;
  if (end_key)
  {
    end_range= &save_end_range;
    save_end_range= *end_key;
    key_compare_result_on_equal=
      ((end_key->flag == HA_READ_BEFORE_KEY) ? 1 :
       (end_key->flag == HA_READ_AFTER_KEY) ? -1 : 0);
  }
  range_key_part= m_curr_key_info->key_part;

  if (!start_key)				// Read first record
  {
3923 3924 3925 3926
    if (m_ordered)
      m_index_scan_type= partition_index_first;
    else
      m_index_scan_type= partition_index_first_unordered;
3927 3928 3929 3930
    error= common_first_last(m_rec0);
  }
  else
  {
3931
    m_index_scan_type= partition_index_read;
3932 3933
    error= common_index_read(m_rec0,
			     start_key->key,
3934
                             start_key->keypart_map, start_key->flag);
3935 3936 3937 3938 3939
  }
  DBUG_RETURN(error);
}


3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950
/*
  Read next record in read of a range with start and end key

  SYNOPSIS
    read_range_next()

  RETURN VALUE
    >0                    Error code
    0                     Success
*/

3951 3952 3953
int ha_partition::read_range_next()
{
  DBUG_ENTER("ha_partition::read_range_next");
3954

3955 3956 3957 3958 3959 3960 3961 3962
  if (m_ordered)
  {
    DBUG_RETURN(handler::read_range_next());
  }
  DBUG_RETURN(handle_unordered_next(m_rec0, eq_range));
}


3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978
/*
  Common routine to set up scans

  SYNOPSIS
    buf                  Buffer to later return record in
    idx_read_flag        Is it index scan

  RETURN VALUE
    >0                    Error code
    0                     Success

  DESCRIPTION
    This is where we check which partitions to actually scan if not all
    of them
*/

3979
int ha_partition::partition_scan_set_up(uchar * buf, bool idx_read_flag)
3980 3981 3982 3983 3984 3985
{
  DBUG_ENTER("ha_partition::partition_scan_set_up");

  if (idx_read_flag)
    get_partition_set(table,buf,active_index,&m_start_key,&m_part_spec);
  else
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
3986 3987 3988 3989
  {
    m_part_spec.start_part= 0;
    m_part_spec.end_part= m_tot_parts - 1;
  }
3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012
  if (m_part_spec.start_part > m_part_spec.end_part)
  {
    /*
      We discovered a partition set but the set was empty so we report
      key not found.
    */
    DBUG_PRINT("info", ("scan with no partition to scan"));
    DBUG_RETURN(HA_ERR_END_OF_FILE);
  }
  if (m_part_spec.start_part == m_part_spec.end_part)
  {
    /*
      We discovered a single partition to scan, this never needs to be
      performed using the ordered index scan.
    */
    DBUG_PRINT("info", ("index scan using the single partition %d",
			m_part_spec.start_part));
    m_ordered_scan_ongoing= FALSE;
  }
  else
  {
    /*
      Set m_ordered_scan_ongoing according how the scan should be done
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4013 4014 4015
      Only exact partitions are discovered atm by get_partition_set.
      Verify this, also bitmap must have at least one bit set otherwise
      the result from this table is the empty set.
4016
    */
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4017 4018 4019 4020 4021 4022 4023 4024 4025
    uint start_part= bitmap_get_first_set(&(m_part_info->used_partitions));
    if (start_part == MY_BIT_NONE)
    {
      DBUG_PRINT("info", ("scan with no partition to scan"));
      DBUG_RETURN(HA_ERR_END_OF_FILE);
    }
    if (start_part > m_part_spec.start_part)
      m_part_spec.start_part= start_part;
    DBUG_ASSERT(m_part_spec.start_part < m_tot_parts);
4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037
    m_ordered_scan_ongoing= m_ordered;
  }
  DBUG_ASSERT(m_part_spec.start_part < m_tot_parts &&
              m_part_spec.end_part < m_tot_parts);
  DBUG_RETURN(0);
}


/****************************************************************************
  Unordered Index Scan Routines
****************************************************************************/
/*
4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057
  Common routine to handle index_next with unordered results

  SYNOPSIS
    handle_unordered_next()
    out:buf                       Read row in MySQL Row Format
    next_same                     Called from index_next_same

  RETURN VALUE
    HA_ERR_END_OF_FILE            End of scan
    0                             Success
    other                         Error code

  DESCRIPTION
    These routines are used to scan partitions without considering order.
    This is performed in two situations.
    1) In read_multi_range this is the normal case
    2) When performing any type of index_read, index_first, index_last where
    all fields in the partition function is bound. In this case the index
    scan is performed on only one partition and thus it isn't necessary to
    perform any sort.
4058 4059
*/

4060
int ha_partition::handle_unordered_next(uchar *buf, bool is_next_same)
4061
{
4062
  handler *file= m_file[m_part_spec.start_part];
4063 4064 4065 4066 4067 4068 4069
  int error;
  DBUG_ENTER("ha_partition::handle_unordered_next");

  /*
    We should consider if this should be split into two functions as
    next_same is alwas a local constant
  */
4070
  if (is_next_same)
4071 4072 4073 4074 4075 4076 4077 4078 4079 4080
  {
    if (!(error= file->index_next_same(buf, m_start_key.key,
                                       m_start_key.length)))
    {
      m_last_part= m_part_spec.start_part;
      DBUG_RETURN(0);
    }
  }
  else if (!(error= file->index_next(buf)))
  {
4081
    if (!(file->index_flags(active_index, 0, 1) & HA_READ_ORDER) ||
4082
        compare_key(end_range) <= 0)
4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099
    {
      m_last_part= m_part_spec.start_part;
      DBUG_RETURN(0);                           // Row was in range
    }
    error= HA_ERR_END_OF_FILE;
  }

  if (error == HA_ERR_END_OF_FILE)
  {
    m_part_spec.start_part++;                    // Start using next part
    error= handle_unordered_scan_next_partition(buf);
  }
  DBUG_RETURN(error);
}


/*
4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113
  Handle index_next when changing to new partition

  SYNOPSIS
    handle_unordered_scan_next_partition()
    buf                       Read row in MySQL Row Format

  RETURN VALUE
    HA_ERR_END_OF_FILE            End of scan
    0                             Success
    other                         Error code

  DESCRIPTION
    This routine is used to start the index scan on the next partition.
    Both initial start and after completing scan on one partition.
4114 4115
*/

4116
int ha_partition::handle_unordered_scan_next_partition(uchar * buf)
4117 4118 4119 4120 4121 4122 4123
{
  uint i;
  DBUG_ENTER("ha_partition::handle_unordered_scan_next_partition");

  for (i= m_part_spec.start_part; i <= m_part_spec.end_part; i++)
  {
    int error;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4124
    handler *file;
4125

patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4126 4127 4128
    if (!(bitmap_is_set(&(m_part_info->used_partitions), i)))
      continue;
    file= m_file[i];
4129 4130 4131 4132
    m_part_spec.start_part= i;
    switch (m_index_scan_type) {
    case partition_index_read:
      DBUG_PRINT("info", ("index_read on partition %d", i));
4133 4134 4135
      error= file->index_read_map(buf, m_start_key.key,
                                  m_start_key.keypart_map,
                                  m_start_key.flag);
4136 4137 4138 4139 4140
      break;
    case partition_index_first:
      DBUG_PRINT("info", ("index_first on partition %d", i));
      error= file->index_first(buf);
      break;
4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152
    case partition_index_first_unordered:
      /*
        We perform a scan without sorting and this means that we
        should not use the index_first since not all handlers
        support it and it is also unnecessary to restrict sort
        order.
      */
      DBUG_PRINT("info", ("read_range_first on partition %d", i));
      table->record[0]= buf;
      error= file->read_range_first(0, end_range, eq_range, 0);
      table->record[0]= m_rec0;
      break;
4153 4154 4155 4156 4157 4158
    default:
      DBUG_ASSERT(FALSE);
      DBUG_RETURN(1);
    }
    if (!error)
    {
4159
      if (!(file->index_flags(active_index, 0, 1) & HA_READ_ORDER) ||
4160
          compare_key(end_range) <= 0)
4161 4162
      {
        m_last_part= i;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4163
        DBUG_RETURN(0);
4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176
      }
      error= HA_ERR_END_OF_FILE;
    }
    if ((error != HA_ERR_END_OF_FILE) && (error != HA_ERR_KEY_NOT_FOUND))
      DBUG_RETURN(error);
    DBUG_PRINT("info", ("HA_ERR_END_OF_FILE on partition %d", i));
  }
  m_part_spec.start_part= NO_CURRENT_PART_ID;
  DBUG_RETURN(HA_ERR_END_OF_FILE);
}


/*
4177
  Common routine to start index scan with ordered results
4178

4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201
  SYNOPSIS
    handle_ordered_index_scan()
    out:buf                       Read row in MySQL Row Format

  RETURN VALUE
    HA_ERR_END_OF_FILE            End of scan
    0                             Success
    other                         Error code

  DESCRIPTION
    This part contains the logic to handle index scans that require ordered
    output. This includes all except those started by read_range_first with
    the flag ordered set to FALSE. Thus most direct index_read and all
    index_first and index_last.

    We implement ordering by keeping one record plus a key buffer for each
    partition. Every time a new entry is requested we will fetch a new
    entry from the partition that is currently not filled with an entry.
    Then the entry is put into its proper sort position.

    Returning a record is done by getting the top record, copying the
    record to the request buffer and setting the partition as empty on
    entries.
4202 4203
*/

4204
int ha_partition::handle_ordered_index_scan(uchar *buf, bool reverse_order)
4205
{
4206 4207
  uint i;
  uint j= 0;
4208 4209 4210 4211
  bool found= FALSE;
  DBUG_ENTER("ha_partition::handle_ordered_index_scan");

  m_top_entry= NO_CURRENT_PART_ID;
4212
  queue_remove_all(&m_queue);
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4213 4214

  DBUG_PRINT("info", ("m_part_spec.start_part %d", m_part_spec.start_part));
4215 4216
  for (i= m_part_spec.start_part; i <= m_part_spec.end_part; i++)
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4217 4218
    if (!(bitmap_is_set(&(m_part_info->used_partitions), i)))
      continue;
4219
    uchar *rec_buf_ptr= rec_buf(i);
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4220
    int error;
4221 4222 4223 4224
    handler *file= m_file[i];

    switch (m_index_scan_type) {
    case partition_index_read:
4225 4226 4227 4228
      error= file->index_read_map(rec_buf_ptr,
                                  m_start_key.key,
                                  m_start_key.keypart_map,
                                  m_start_key.flag);
4229 4230 4231 4232 4233 4234 4235 4236 4237
      break;
    case partition_index_first:
      error= file->index_first(rec_buf_ptr);
      reverse_order= FALSE;
      break;
    case partition_index_last:
      error= file->index_last(rec_buf_ptr);
      reverse_order= TRUE;
      break;
4238
    case partition_index_read_last:
4239 4240 4241
      error= file->index_read_last_map(rec_buf_ptr,
                                       m_start_key.key,
                                       m_start_key.keypart_map);
4242 4243
      reverse_order= TRUE;
      break;
4244 4245 4246 4247 4248 4249 4250 4251 4252 4253
    default:
      DBUG_ASSERT(FALSE);
      DBUG_RETURN(HA_ERR_END_OF_FILE);
    }
    if (!error)
    {
      found= TRUE;
      /*
        Initialise queue without order first, simply insert
      */
4254
      queue_element(&m_queue, j++)= (uchar*)queue_buf(i);
4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266
    }
    else if (error != HA_ERR_KEY_NOT_FOUND && error != HA_ERR_END_OF_FILE)
    {
      DBUG_RETURN(error);
    }
  }
  if (found)
  {
    /*
      We found at least one partition with data, now sort all entries and
      after that read the first entry and copy it to the buffer to return in.
    */
4267 4268 4269 4270
    queue_set_max_at_top(&m_queue, reverse_order);
    queue_set_cmp_arg(&m_queue, (void*)m_curr_key_info);
    m_queue.elements= j;
    queue_fix(&m_queue);
4271
    return_top_record(buf);
4272
    table->status= 0;
4273 4274 4275 4276 4277 4278 4279
    DBUG_PRINT("info", ("Record returned from partition %d", m_top_entry));
    DBUG_RETURN(0);
  }
  DBUG_RETURN(HA_ERR_END_OF_FILE);
}


4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290
/*
  Return the top record in sort order

  SYNOPSIS
    return_top_record()
    out:buf                  Row returned in MySQL Row Format

  RETURN VALUE
    NONE
*/

4291
void ha_partition::return_top_record(uchar *buf)
4292 4293
{
  uint part_id;
4294 4295
  uchar *key_buffer= queue_top(&m_queue);
  uchar *rec_buffer= key_buffer + PARTITION_BYTES_IN_POS;
4296

4297 4298 4299 4300 4301 4302 4303
  part_id= uint2korr(key_buffer);
  memcpy(buf, rec_buffer, m_rec_length);
  m_last_part= part_id;
  m_top_entry= part_id;
}


4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317
/*
  Common routine to handle index_next with ordered results

  SYNOPSIS
    handle_ordered_next()
    out:buf                       Read row in MySQL Row Format
    next_same                     Called from index_next_same

  RETURN VALUE
    HA_ERR_END_OF_FILE            End of scan
    0                             Success
    other                         Error code
*/

4318
int ha_partition::handle_ordered_next(uchar *buf, bool is_next_same)
4319 4320 4321 4322 4323 4324
{
  int error;
  uint part_id= m_top_entry;
  handler *file= m_file[part_id];
  DBUG_ENTER("ha_partition::handle_ordered_next");

4325
  if (!is_next_same)
4326 4327 4328 4329 4330 4331 4332 4333 4334
    error= file->index_next(rec_buf(part_id));
  else
    error= file->index_next_same(rec_buf(part_id), m_start_key.key,
				 m_start_key.length);
  if (error)
  {
    if (error == HA_ERR_END_OF_FILE)
    {
      /* Return next buffered row */
4335 4336
      queue_remove(&m_queue, (uint) 0);
      if (m_queue.elements)
4337 4338 4339 4340
      {
         DBUG_PRINT("info", ("Record returned from partition %u (2)",
                     m_top_entry));
         return_top_record(buf);
4341
         table->status= 0;
4342 4343 4344 4345 4346
         error= 0;
      }
    }
    DBUG_RETURN(error);
  }
4347
  queue_replaced(&m_queue);
4348 4349 4350 4351 4352 4353
  return_top_record(buf);
  DBUG_PRINT("info", ("Record returned from partition %u", m_top_entry));
  DBUG_RETURN(0);
}


4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366
/*
  Common routine to handle index_prev with ordered results

  SYNOPSIS
    handle_ordered_prev()
    out:buf                       Read row in MySQL Row Format

  RETURN VALUE
    HA_ERR_END_OF_FILE            End of scan
    0                             Success
    other                         Error code
*/

4367
int ha_partition::handle_ordered_prev(uchar *buf)
4368 4369 4370 4371 4372
{
  int error;
  uint part_id= m_top_entry;
  handler *file= m_file[part_id];
  DBUG_ENTER("ha_partition::handle_ordered_prev");
4373

4374 4375 4376 4377
  if ((error= file->index_prev(rec_buf(part_id))))
  {
    if (error == HA_ERR_END_OF_FILE)
    {
4378 4379
      queue_remove(&m_queue, (uint) 0);
      if (m_queue.elements)
4380 4381 4382 4383 4384
      {
	return_top_record(buf);
	DBUG_PRINT("info", ("Record returned from partition %d (2)",
			    m_top_entry));
        error= 0;
4385
        table->status= 0;
4386 4387 4388 4389
      }
    }
    DBUG_RETURN(error);
  }
4390
  queue_replaced(&m_queue);
4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406
  return_top_record(buf);
  DBUG_PRINT("info", ("Record returned from partition %d", m_top_entry));
  DBUG_RETURN(0);
}


/****************************************************************************
                MODULE information calls
****************************************************************************/

/*
  These are all first approximations of the extra, info, scan_time
  and read_time calls
*/

/*
4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468
  General method to gather info from handler

  SYNOPSIS
    info()
    flag              Specifies what info is requested

  RETURN VALUE
    NONE

  DESCRIPTION
    ::info() is used to return information to the optimizer.
    Currently this table handler doesn't implement most of the fields
    really needed. SHOW also makes use of this data
    Another note, if your handler doesn't proved exact record count,
    you will probably want to have the following in your code:
    if (records < 2)
      records = 2;
    The reason is that the server will optimize for cases of only a single
    record. If in a table scan you don't know the number of records
    it will probably be better to set records to two so you can return
    as many records as you need.

    Along with records a few more variables you may wish to set are:
      records
      deleted
      data_file_length
      index_file_length
      delete_length
      check_time
    Take a look at the public variables in handler.h for more information.

    Called in:
      filesort.cc
      ha_heap.cc
      item_sum.cc
      opt_sum.cc
      sql_delete.cc
     sql_delete.cc
     sql_derived.cc
      sql_select.cc
      sql_select.cc
      sql_select.cc
      sql_select.cc
      sql_select.cc
      sql_show.cc
      sql_show.cc
      sql_show.cc
      sql_show.cc
      sql_table.cc
      sql_union.cc
      sql_update.cc

    Some flags that are not implemented
      HA_STATUS_POS:
        This parameter is never used from the MySQL Server. It is checked in a
        place in MyISAM so could potentially be used by MyISAM specific
        programs.
      HA_STATUS_NO_LOCK:
      This is declared and often used. It's only used by MyISAM.
      It means that MySQL doesn't need the absolute latest statistics
      information. This may save the handler from doing internal locks while
      retrieving statistics data.
4469 4470
*/

4471
int ha_partition::info(uint flag)
4472 4473 4474 4475 4476 4477
{
  handler *file, **file_array;
  DBUG_ENTER("ha_partition:info");

  if (flag & HA_STATUS_AUTO)
  {
4478
    ulonglong auto_increment_value= 0;
4479
    DBUG_PRINT("info", ("HA_STATUS_AUTO"));
4480 4481
    file_array= m_file;
    do
4482
    {
4483 4484 4485 4486 4487
      file= *file_array;
      file->info(HA_STATUS_AUTO);
      set_if_bigger(auto_increment_value, file->stats.auto_increment_value);
    } while (*(++file_array));
    stats.auto_increment_value= auto_increment_value;
4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503
  }
  if (flag & HA_STATUS_VARIABLE)
  {
    DBUG_PRINT("info", ("HA_STATUS_VARIABLE"));
    /*
      Calculates statistical variables
      records:           Estimate of number records in table
      We report sum (always at least 2)
      deleted:           Estimate of number holes in the table due to
      deletes
      We report sum
      data_file_length:  Length of data file, in principle bytes in table
      We report sum
      index_file_length: Length of index file, in principle bytes in
      indexes in the table
      We report sum
4504 4505
      delete_length: Length of free space easily used by new records in table
      We report sum
4506 4507 4508 4509 4510
      mean_record_length:Mean record length in the table
      We calculate this
      check_time:        Time of last check (only applicable to MyISAM)
      We report last time of all underlying handlers
    */
4511 4512 4513 4514 4515
    stats.records= 0;
    stats.deleted= 0;
    stats.data_file_length= 0;
    stats.index_file_length= 0;
    stats.check_time= 0;
4516
    stats.delete_length= 0;
4517 4518 4519
    file_array= m_file;
    do
    {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4520 4521 4522 4523
      if (bitmap_is_set(&(m_part_info->used_partitions), (file_array - m_file)))
      {
        file= *file_array;
        file->info(HA_STATUS_VARIABLE);
4524 4525 4526 4527
        stats.records+= file->stats.records;
        stats.deleted+= file->stats.deleted;
        stats.data_file_length+= file->stats.data_file_length;
        stats.index_file_length+= file->stats.index_file_length;
monty@mysql.com's avatar
monty@mysql.com committed
4528
        stats.delete_length+= file->stats.delete_length;
4529 4530
        if (file->stats.check_time > stats.check_time)
          stats.check_time= file->stats.check_time;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
4531
      }
4532
    } while (*(++file_array));
4533 4534 4535 4536 4537
    if (stats.records < 2 &&
        !(m_table_flags & HA_STATS_RECORDS_IS_EXACT))
      stats.records= 2;
    if (stats.records > 0)
      stats.mean_rec_length= (ulong) (stats.data_file_length / stats.records);
4538
    else
4539
      stats.mean_rec_length= 1; //? What should we set here 
4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594
  }
  if (flag & HA_STATUS_CONST)
  {
    DBUG_PRINT("info", ("HA_STATUS_CONST"));
    /*
      Recalculate loads of constant variables. MyISAM also sets things
      directly on the table share object.

      Check whether this should be fixed since handlers should not
      change things directly on the table object.

      Monty comment: This should NOT be changed!  It's the handlers
      responsibility to correct table->s->keys_xxxx information if keys
      have been disabled.

      The most important parameters set here is records per key on
      all indexes. block_size and primar key ref_length.

      For each index there is an array of rec_per_key.
      As an example if we have an index with three attributes a,b and c
      we will have an array of 3 rec_per_key.
      rec_per_key[0] is an estimate of number of records divided by
      number of unique values of the field a.
      rec_per_key[1] is an estimate of the number of records divided
      by the number of unique combinations of the fields a and b.
      rec_per_key[2] is an estimate of the number of records divided
      by the number of unique combinations of the fields a,b and c.

      Many handlers only set the value of rec_per_key when all fields
      are bound (rec_per_key[2] in the example above).

      If the handler doesn't support statistics, it should set all of the
      above to 0.

      We will allow the first handler to set the rec_per_key and use
      this as an estimate on the total table.

      max_data_file_length:     Maximum data file length
      We ignore it, is only used in
      SHOW TABLE STATUS
      max_index_file_length:    Maximum index file length
      We ignore it since it is never used
      block_size:               Block size used
      We set it to the value of the first handler
      ref_length:               We set this to the value calculated
      and stored in local object
      create_time:              Creation time of table
      Set by first handler

      So we calculate these constants by using the variables on the first
      handler.
    */

    file= m_file[0];
    file->info(HA_STATUS_CONST);
4595
    stats.create_time= file->stats.create_time;
4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618
    ref_length= m_ref_length;
  }
  if (flag & HA_STATUS_ERRKEY)
  {
    handler *file= m_file[m_last_part];
    DBUG_PRINT("info", ("info: HA_STATUS_ERRKEY"));
    /*
      This flag is used to get index number of the unique index that
      reported duplicate key
      We will report the errkey on the last handler used and ignore the rest
    */
    file->info(HA_STATUS_ERRKEY);
    if (file->errkey != (uint) -1)
      errkey= file->errkey;
  }
  if (flag & HA_STATUS_TIME)
  {
    DBUG_PRINT("info", ("info: HA_STATUS_TIME"));
    /*
      This flag is used to set the latest update time of the table.
      Used by SHOW commands
      We will report the maximum of these times
    */
4619
    stats.update_time= 0;
4620 4621 4622 4623 4624
    file_array= m_file;
    do
    {
      file= *file_array;
      file->info(HA_STATUS_TIME);
4625 4626
      if (file->stats.update_time > stats.update_time)
	stats.update_time= file->stats.update_time;
4627 4628
    } while (*(++file_array));
  }
4629
  DBUG_RETURN(0);
4630 4631 4632
}


4633 4634 4635 4636 4637 4638 4639
void ha_partition::get_dynamic_partition_info(PARTITION_INFO *stat_info,
                                              uint part_id)
{
  handler *file= m_file[part_id];
  file->info(HA_STATUS_CONST | HA_STATUS_TIME | HA_STATUS_VARIABLE |
             HA_STATUS_NO_LOCK);

4640 4641 4642 4643 4644 4645 4646 4647 4648
  stat_info->records=              file->stats.records;
  stat_info->mean_rec_length=      file->stats.mean_rec_length;
  stat_info->data_file_length=     file->stats.data_file_length;
  stat_info->max_data_file_length= file->stats.max_data_file_length;
  stat_info->index_file_length=    file->stats.index_file_length;
  stat_info->delete_length=        file->stats.delete_length;
  stat_info->create_time=          file->stats.create_time;
  stat_info->update_time=          file->stats.update_time;
  stat_info->check_time=           file->stats.check_time;
4649
  stat_info->check_sum= 0;
4650
  if (file->ha_table_flags() & HA_HAS_CHECKSUM)
4651 4652 4653 4654 4655
    stat_info->check_sum= file->checksum();
  return;
}


4656
/*
4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667
  General function to prepare handler for certain behavior

  SYNOPSIS
    extra()
    operation              Operation type for extra call

  RETURN VALUE
    >0                     Error code
    0                      Success

  DESCRIPTION
4668 4669 4670 4671 4672 4673 4674 4675 4676 4677
  extra() is called whenever the server wishes to send a hint to
  the storage engine. The MyISAM engine implements the most hints.

  We divide the parameters into the following categories:
  1) Parameters used by most handlers
  2) Parameters used by some non-MyISAM handlers
  3) Parameters used only by MyISAM
  4) Parameters only used by temporary tables for query processing
  5) Parameters only used by MyISAM internally
  6) Parameters not used at all
4678 4679
  7) Parameters only used by federated tables for query processing
  8) Parameters only used by NDB
4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696

  The partition handler need to handle category 1), 2) and 3).

  1) Parameters used by most handlers
  -----------------------------------
  HA_EXTRA_RESET:
    This option is used by most handlers and it resets the handler state
    to the same state as after an open call. This includes releasing
    any READ CACHE or WRITE CACHE or other internal buffer used.

    It is called from the reset method in the handler interface. There are
    three instances where this is called.
    1) After completing a INSERT ... SELECT ... query the handler for the
       table inserted into is reset
    2) It is called from close_thread_table which in turn is called from
       close_thread_tables except in the case where the tables are locked
       in which case ha_commit_stmt is called instead.
4697
       It is only called from here if refresh_version hasn't changed and the
4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717
       table is not an old table when calling close_thread_table.
       close_thread_tables is called from many places as a general clean up
       function after completing a query.
    3) It is called when deleting the QUICK_RANGE_SELECT object if the
       QUICK_RANGE_SELECT object had its own handler object. It is called
       immediatley before close of this local handler object.
  HA_EXTRA_KEYREAD:
  HA_EXTRA_NO_KEYREAD:
    These parameters are used to provide an optimisation hint to the handler.
    If HA_EXTRA_KEYREAD is set it is enough to read the index fields, for
    many handlers this means that the index-only scans can be used and it
    is not necessary to use the real records to satisfy this part of the
    query. Index-only scans is a very important optimisation for disk-based
    indexes. For main-memory indexes most indexes contain a reference to the
    record and thus KEYREAD only says that it is enough to read key fields.
    HA_EXTRA_NO_KEYREAD disables this for the handler, also HA_EXTRA_RESET
    will disable this option.
    The handler will set HA_KEYREAD_ONLY in its table flags to indicate this
    feature is supported.
  HA_EXTRA_FLUSH:
4718
    Indication to flush tables to disk, is supposed to be used to
4719
    ensure disk based tables are flushed at end of query execution.
4720
    Currently is never used.
4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840

  2) Parameters used by some non-MyISAM handlers
  ----------------------------------------------
  HA_EXTRA_KEYREAD_PRESERVE_FIELDS:
    This is a strictly InnoDB feature that is more or less undocumented.
    When it is activated InnoDB copies field by field from its fetch
    cache instead of all fields in one memcpy. Have no idea what the
    purpose of this is.
    Cut from include/my_base.h:
    When using HA_EXTRA_KEYREAD, overwrite only key member fields and keep
    other fields intact. When this is off (by default) InnoDB will use memcpy
    to overwrite entire row.
  HA_EXTRA_IGNORE_DUP_KEY:
  HA_EXTRA_NO_IGNORE_DUP_KEY:
    Informs the handler to we will not stop the transaction if we get an
    duplicate key errors during insert/upate.
    Always called in pair, triggered by INSERT IGNORE and other similar
    SQL constructs.
    Not used by MyISAM.

  3) Parameters used only by MyISAM
  ---------------------------------
  HA_EXTRA_NORMAL:
    Only used in MyISAM to reset quick mode, not implemented by any other
    handler. Quick mode is also reset in MyISAM by HA_EXTRA_RESET.

    It is called after completing a successful DELETE query if the QUICK
    option is set.

  HA_EXTRA_QUICK:
    When the user does DELETE QUICK FROM table where-clause; this extra
    option is called before the delete query is performed and
    HA_EXTRA_NORMAL is called after the delete query is completed.
    Temporary tables used internally in MySQL always set this option

    The meaning of quick mode is that when deleting in a B-tree no merging
    of leafs is performed. This is a common method and many large DBMS's
    actually only support this quick mode since it is very difficult to
    merge leaves in a tree used by many threads concurrently.

  HA_EXTRA_CACHE:
    This flag is usually set with extra_opt along with a cache size.
    The size of this buffer is set by the user variable
    record_buffer_size. The value of this cache size is the amount of
    data read from disk in each fetch when performing a table scan.
    This means that before scanning a table it is normal to call
    extra with HA_EXTRA_CACHE and when the scan is completed to call
    HA_EXTRA_NO_CACHE to release the cache memory.

    Some special care is taken when using this extra parameter since there
    could be a write ongoing on the table in the same statement. In this
    one has to take special care since there might be a WRITE CACHE as
    well. HA_EXTRA_CACHE specifies using a READ CACHE and using
    READ CACHE and WRITE CACHE at the same time is not possible.

    Only MyISAM currently use this option.

    It is set when doing full table scans using rr_sequential and
    reset when completing such a scan with end_read_record
    (resetting means calling extra with HA_EXTRA_NO_CACHE).

    It is set in filesort.cc for MyISAM internal tables and it is set in
    a multi-update where HA_EXTRA_CACHE is called on a temporary result
    table and after that ha_rnd_init(0) on table to be updated
    and immediately after that HA_EXTRA_NO_CACHE on table to be updated.

    Apart from that it is always used from init_read_record but not when
    used from UPDATE statements. It is not used from DELETE statements
    with ORDER BY and LIMIT but it is used in normal scan loop in DELETE
    statements. The reason here is that DELETE's in MyISAM doesn't move
    existings data rows.

    It is also set in copy_data_between_tables when scanning the old table
    to copy over to the new table.
    And it is set in join_init_read_record where quick objects are used
    to perform a scan on the table. In this case the full table scan can
    even be performed multiple times as part of the nested loop join.

    For purposes of the partition handler it is obviously necessary to have
    special treatment of this extra call. If we would simply pass this
    extra call down to each handler we would allocate
    cache size * no of partitions amount of memory and this is not
    necessary since we will only scan one partition at a time when doing
    full table scans.

    Thus we treat it by first checking whether we have MyISAM handlers in
    the table, if not we simply ignore the call and if we have we will
    record the call but will not call any underlying handler yet. Then
    when performing the sequential scan we will check this recorded value
    and call extra_opt whenever we start scanning a new partition.

    monty: Neads to be fixed so that it's passed to all handlers when we
    move to another partition during table scan.

  HA_EXTRA_NO_CACHE:
    When performing a UNION SELECT HA_EXTRA_NO_CACHE is called from the
    flush method in the select_union class.
    It is used to some extent when insert delayed inserts.
    See HA_EXTRA_RESET_STATE for use in conjunction with delete_all_rows().

    It should be ok to call HA_EXTRA_NO_CACHE on all underlying handlers
    if they are MyISAM handlers. Other handlers we can ignore the call
    for. If no cache is in use they will quickly return after finding
    this out. And we also ensure that all caches are disabled and no one
    is left by mistake.
    In the future this call will probably be deleted an we will instead call
    ::reset();

  HA_EXTRA_WRITE_CACHE:
    See above, called from various places. It is mostly used when we
    do INSERT ... SELECT
    No special handling to save cache space is developed currently.

  HA_EXTRA_PREPARE_FOR_UPDATE:
    This is called as part of a multi-table update. When the table to be
    updated is also scanned then this informs MyISAM handler to drop any
    caches if dynamic records are used (fixed size records do not care
    about this call). We pass this along to all underlying MyISAM handlers
    and ignore it for the rest.

4841
  HA_EXTRA_PREPARE_FOR_DROP:
4842 4843 4844 4845
    Only used by MyISAM, called in preparation for a DROP TABLE.
    It's used mostly by Windows that cannot handle dropping an open file.
    On other platforms it has the same effect as HA_EXTRA_FORCE_REOPEN.

4846 4847 4848
  HA_EXTRA_PREPARE_FOR_RENAME:
    Informs the handler we are about to attempt a rename of the table.

4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874
  HA_EXTRA_READCHECK:
  HA_EXTRA_NO_READCHECK:
    Only one call to HA_EXTRA_NO_READCHECK from ha_open where it says that
    this is not needed in SQL. The reason for this call is that MyISAM sets
    the READ_CHECK_USED in the open call so the call is needed for MyISAM
    to reset this feature.
    The idea with this parameter was to inform of doing/not doing a read
    check before applying an update. Since SQL always performs a read before
    applying the update No Read Check is needed in MyISAM as well.

    This is a cut from Docs/myisam.txt
     Sometimes you might want to force an update without checking whether
     another user has changed the record since you last read it. This is
     somewhat dangerous, so it should ideally not be used. That can be
     accomplished by wrapping the mi_update() call in two calls to mi_extra(),
     using these functions:
     HA_EXTRA_NO_READCHECK=5                 No readcheck on update
     HA_EXTRA_READCHECK=6                    Use readcheck (def)

  HA_EXTRA_FORCE_REOPEN:
    Only used by MyISAM, called when altering table, closing tables to
    enforce a reopen of the table files.

  4) Parameters only used by temporary tables for query processing
  ----------------------------------------------------------------
  HA_EXTRA_RESET_STATE:
4875
    Same as reset() except that buffers are not released. If there is
4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913
    a READ CACHE it is reinit'ed. A cache is reinit'ed to restart reading
    or to change type of cache between READ CACHE and WRITE CACHE.

    This extra function is always called immediately before calling
    delete_all_rows on the handler for temporary tables.
    There are cases however when HA_EXTRA_RESET_STATE isn't called in
    a similar case for a temporary table in sql_union.cc and in two other
    cases HA_EXTRA_NO_CACHE is called before and HA_EXTRA_WRITE_CACHE
    called afterwards.
    The case with HA_EXTRA_NO_CACHE and HA_EXTRA_WRITE_CACHE means
    disable caching, delete all rows and enable WRITE CACHE. This is
    used for temporary tables containing distinct sums and a
    functional group.

    The only case that delete_all_rows is called on non-temporary tables
    is in sql_delete.cc when DELETE FROM table; is called by a user.
    In this case no special extra calls are performed before or after this
    call.

    The partition handler should not need to bother about this one. It
    should never be called.

  HA_EXTRA_NO_ROWS:
    Don't insert rows indication to HEAP and MyISAM, only used by temporary
    tables used in query processing.
    Not handled by partition handler.

  5) Parameters only used by MyISAM internally
  --------------------------------------------
  HA_EXTRA_REINIT_CACHE:
    This call reinitialises the READ CACHE described above if there is one
    and otherwise the call is ignored.

    We can thus safely call it on all underlying handlers if they are
    MyISAM handlers. It is however never called so we don't handle it at all.
  HA_EXTRA_FLUSH_CACHE:
    Flush WRITE CACHE in MyISAM. It is only from one place in the code.
    This is in sql_insert.cc where it is called if the table_flags doesn't
4914 4915 4916
    contain HA_DUPLICATE_POS. The only handler having the HA_DUPLICATE_POS
    set is the MyISAM handler and so the only handler not receiving this
    call is MyISAM.
4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942
    Thus in effect this call is called but never used. Could be removed
    from sql_insert.cc
  HA_EXTRA_NO_USER_CHANGE:
    Only used by MyISAM, never called.
    Simulates lock_type as locked.
  HA_EXTRA_WAIT_LOCK:
  HA_EXTRA_WAIT_NOLOCK:
    Only used by MyISAM, called from MyISAM handler but never from server
    code on top of the handler.
    Sets lock_wait on/off
  HA_EXTRA_NO_KEYS:
    Only used MyISAM, only used internally in MyISAM handler, never called
    from server level.
  HA_EXTRA_KEYREAD_CHANGE_POS:
  HA_EXTRA_REMEMBER_POS:
  HA_EXTRA_RESTORE_POS:
  HA_EXTRA_PRELOAD_BUFFER_SIZE:
  HA_EXTRA_CHANGE_KEY_TO_DUP:
  HA_EXTRA_CHANGE_KEY_TO_UNIQUE:
    Only used by MyISAM, never called.

  6) Parameters not used at all
  -----------------------------
  HA_EXTRA_KEY_CACHE:
  HA_EXTRA_NO_KEY_CACHE:
    This parameters are no longer used and could be removed.
istruewing@synthia.local's avatar
istruewing@synthia.local committed
4943 4944 4945 4946 4947 4948

  7) Parameters only used by federated tables for query processing
  ----------------------------------------------------------------
  HA_EXTRA_INSERT_WITH_UPDATE:
    Inform handler that an "INSERT...ON DUPLICATE KEY UPDATE" will be
    executed. This condition is unset by HA_EXTRA_NO_IGNORE_DUP_KEY.
4949 4950 4951 4952 4953 4954 4955 4956 4957

  8) Parameters only used by NDB
  ------------------------------
  HA_EXTRA_DELETE_CANNOT_BATCH:
  HA_EXTRA_UPDATE_CANNOT_BATCH:
    Inform handler that delete_row()/update_row() cannot batch deletes/updates
    and should perform them immediately. This may be needed when table has 
    AFTER DELETE/UPDATE triggers which access to subject table.
    These flags are reset by the handler::extra(HA_EXTRA_RESET) call.
4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982
*/

int ha_partition::extra(enum ha_extra_function operation)
{
  DBUG_ENTER("ha_partition:extra");
  DBUG_PRINT("info", ("operation: %d", (int) operation));

  switch (operation) {
    /* Category 1), used by most handlers */
  case HA_EXTRA_KEYREAD:
  case HA_EXTRA_NO_KEYREAD:
  case HA_EXTRA_FLUSH:
    DBUG_RETURN(loop_extra(operation));

    /* Category 2), used by non-MyISAM handlers */
  case HA_EXTRA_IGNORE_DUP_KEY:
  case HA_EXTRA_NO_IGNORE_DUP_KEY:
  case HA_EXTRA_KEYREAD_PRESERVE_FIELDS:
  {
    if (!m_myisam)
      DBUG_RETURN(loop_extra(operation));
    break;
  }

  /* Category 3), used by MyISAM handlers */
4983 4984
  case HA_EXTRA_PREPARE_FOR_RENAME:
    DBUG_RETURN(prepare_for_rename());
4985
    break;
4986 4987 4988 4989 4990
  case HA_EXTRA_NORMAL:
  case HA_EXTRA_QUICK:
  case HA_EXTRA_NO_READCHECK:
  case HA_EXTRA_PREPARE_FOR_UPDATE:
  case HA_EXTRA_FORCE_REOPEN:
4991
  case HA_EXTRA_PREPARE_FOR_DROP:
4992
  case HA_EXTRA_FLUSH_CACHE:
4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003
  {
    if (m_myisam)
      DBUG_RETURN(loop_extra(operation));
    break;
  }
  case HA_EXTRA_CACHE:
  {
    prepare_extra_cache(0);
    break;
  }
  case HA_EXTRA_NO_CACHE:
5004
  case HA_EXTRA_WRITE_CACHE:
5005 5006 5007 5008 5009
  {
    m_extra_cache= FALSE;
    m_extra_cache_size= 0;
    DBUG_RETURN(loop_extra(operation));
  }
5010 5011 5012 5013 5014 5015 5016 5017 5018
  case HA_EXTRA_IGNORE_NO_KEY:
  case HA_EXTRA_NO_IGNORE_NO_KEY:
  {
    /*
      Ignore as these are specific to NDB for handling
      idempotency
     */
    break;
  }
5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039
  case HA_EXTRA_WRITE_CAN_REPLACE:
  case HA_EXTRA_WRITE_CANNOT_REPLACE:
  {
    /*
      Informs handler that write_row() can replace rows which conflict
      with row being inserted by PK/unique key without reporting error
      to the SQL-layer.

      This optimization is not safe for partitioned table in general case
      since we may have to put new version of row into partition which is
      different from partition in which old version resides (for example
      when we partition by non-PK column or by some column which is not
      part of unique key which were violated).
      And since NDB which is the only engine at the moment that supports
      this optimization handles partitioning on its own we simple disable
      it here. (BTW for NDB this optimization is safe since it supports
      only KEY partitioning and won't use this optimization for tables
      which have additional unique constraints).
    */
    break;
  }
istruewing@synthia.local's avatar
istruewing@synthia.local committed
5040 5041 5042
    /* Category 7), used by federated handlers */
  case HA_EXTRA_INSERT_WITH_UPDATE:
    DBUG_RETURN(loop_extra(operation));
5043 5044 5045 5046 5047 5048 5049
    /* Category 8) Parameters only used by NDB */
  case HA_EXTRA_DELETE_CANNOT_BATCH:
  case HA_EXTRA_UPDATE_CANNOT_BATCH:
  {
    /* Currently only NDB use the *_CANNOT_BATCH */
    break;
  }
5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061
  default:
  {
    /* Temporary crash to discover what is wrong */
    DBUG_ASSERT(0);
    break;
  }
  }
  DBUG_RETURN(0);
}


/*
5062 5063 5064 5065 5066 5067 5068 5069 5070 5071
  Special extra call to reset extra parameters

  SYNOPSIS
    reset()

  RETURN VALUE
    >0                   Error code
    0                    Success

  DESCRIPTION
5072
    Called at end of each statement to reste buffers
5073 5074 5075 5076 5077 5078 5079
*/

int ha_partition::reset(void)
{
  int result= 0, tmp;
  handler **file;
  DBUG_ENTER("ha_partition::reset");
5080
  if (m_part_info)
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5081 5082
    bitmap_set_all(&m_part_info->used_partitions);
  file= m_file;
5083 5084
  do
  {
5085
    if ((tmp= (*file)->ha_reset()))
5086 5087 5088 5089 5090
      result= tmp;
  } while (*(++file));
  DBUG_RETURN(result);
}

5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103
/*
  Special extra method for HA_EXTRA_CACHE with cachesize as extra parameter

  SYNOPSIS
    extra_opt()
    operation                      Must be HA_EXTRA_CACHE
    cachesize                      Size of cache in full table scan

  RETURN VALUE
    >0                   Error code
    0                    Success
*/

5104 5105 5106
int ha_partition::extra_opt(enum ha_extra_function operation, ulong cachesize)
{
  DBUG_ENTER("ha_partition::extra_opt()");
5107

5108 5109 5110 5111 5112 5113
  DBUG_ASSERT(HA_EXTRA_CACHE == operation);
  prepare_extra_cache(cachesize);
  DBUG_RETURN(0);
}


5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124
/*
  Call extra on handler with HA_EXTRA_CACHE and cachesize

  SYNOPSIS
    prepare_extra_cache()
    cachesize                Size of cache for full table scan

  RETURN VALUE
    NONE
*/

5125 5126 5127 5128 5129 5130 5131 5132
void ha_partition::prepare_extra_cache(uint cachesize)
{
  DBUG_ENTER("ha_partition::prepare_extra_cache()");

  m_extra_cache= TRUE;
  m_extra_cache_size= cachesize;
  if (m_part_spec.start_part != NO_CURRENT_PART_ID)
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5133
    late_extra_cache(m_part_spec.start_part);
5134 5135 5136 5137 5138
  }
  DBUG_VOID_RETURN;
}


5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149
/*
  Prepares our new and reorged handlers for rename or delete

  SYNOPSIS
    prepare_for_delete()

  RETURN VALUE
    >0                    Error code
    0                     Success
*/

5150
int ha_partition::prepare_for_rename()
5151 5152 5153
{
  int result= 0, tmp;
  handler **file;
5154
  DBUG_ENTER("ha_partition::prepare_for_rename()");
5155 5156 5157 5158
  
  if (m_new_file != NULL)
  {
    for (file= m_new_file; *file; file++)
5159
      if ((tmp= (*file)->extra(HA_EXTRA_PREPARE_FOR_RENAME)))
5160 5161
        result= tmp;      
    for (file= m_reorged_file; *file; file++)
5162
      if ((tmp= (*file)->extra(HA_EXTRA_PREPARE_FOR_RENAME)))
reggie@big_geek's avatar
reggie@big_geek committed
5163 5164
        result= tmp;   
    DBUG_RETURN(result);   
5165
  }
reggie@big_geek's avatar
reggie@big_geek committed
5166
  
5167
  DBUG_RETURN(loop_extra(HA_EXTRA_PREPARE_FOR_RENAME));
5168 5169
}

5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181
/*
  Call extra on all partitions

  SYNOPSIS
    loop_extra()
    operation             extra operation type

  RETURN VALUE
    >0                    Error code
    0                     Success
*/

5182 5183 5184 5185 5186
int ha_partition::loop_extra(enum ha_extra_function operation)
{
  int result= 0, tmp;
  handler **file;
  DBUG_ENTER("ha_partition::loop_extra()");
5187
  
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5188 5189 5190 5191
  /* 
    TODO, 5.2: this is where you could possibly add optimisations to add the bitmap
    _if_ a SELECT.
  */
5192 5193 5194 5195 5196 5197 5198 5199 5200
  for (file= m_file; *file; file++)
  {
    if ((tmp= (*file)->extra(operation)))
      result= tmp;
  }
  DBUG_RETURN(result);
}


5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211
/*
  Call extra(HA_EXTRA_CACHE) on next partition_id

  SYNOPSIS
    late_extra_cache()
    partition_id               Partition id to call extra on

  RETURN VALUE
    NONE
*/

5212 5213 5214 5215
void ha_partition::late_extra_cache(uint partition_id)
{
  handler *file;
  DBUG_ENTER("ha_partition::late_extra_cache");
5216

5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227
  if (!m_extra_cache)
    DBUG_VOID_RETURN;
  file= m_file[partition_id];
  if (m_extra_cache_size == 0)
    VOID(file->extra(HA_EXTRA_CACHE));
  else
    VOID(file->extra_opt(HA_EXTRA_CACHE, m_extra_cache_size));
  DBUG_VOID_RETURN;
}


5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238
/*
  Call extra(HA_EXTRA_NO_CACHE) on next partition_id

  SYNOPSIS
    late_extra_no_cache()
    partition_id               Partition id to call extra on

  RETURN VALUE
    NONE
*/

5239 5240 5241 5242
void ha_partition::late_extra_no_cache(uint partition_id)
{
  handler *file;
  DBUG_ENTER("ha_partition::late_extra_no_cache");
5243

5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255
  if (!m_extra_cache)
    DBUG_VOID_RETURN;
  file= m_file[partition_id];
  VOID(file->extra(HA_EXTRA_NO_CACHE));
  DBUG_VOID_RETURN;
}


/****************************************************************************
                MODULE optimiser support
****************************************************************************/

5256 5257 5258 5259 5260 5261 5262 5263 5264 5265
/*
  Get keys to use for scanning

  SYNOPSIS
    keys_to_use_for_scanning()

  RETURN VALUE
    key_map of keys usable for scanning
*/

5266 5267 5268
const key_map *ha_partition::keys_to_use_for_scanning()
{
  DBUG_ENTER("ha_partition::keys_to_use_for_scanning");
5269

5270 5271 5272
  DBUG_RETURN(m_file[0]->keys_to_use_for_scanning());
}

5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283

/*
  Return time for a scan of the table

  SYNOPSIS
    scan_time()

  RETURN VALUE
    time for scan
*/

5284 5285 5286 5287 5288 5289 5290
double ha_partition::scan_time()
{
  double scan_time= 0;
  handler **file;
  DBUG_ENTER("ha_partition::scan_time");

  for (file= m_file; *file; file++)
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5291 5292
    if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
      scan_time+= (*file)->scan_time();
5293 5294 5295 5296 5297
  DBUG_RETURN(scan_time);
}


/*
5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313
  Get time to read

  SYNOPSIS
    read_time()
    index                Index number used
    ranges               Number of ranges
    rows                 Number of rows

  RETURN VALUE
    time for read

  DESCRIPTION
    This will be optimised later to include whether or not the index can
    be used with partitioning. To achieve we need to add another parameter
    that specifies how many of the index fields that are bound in the ranges.
    Possibly added as a new call to handlers.
5314 5315 5316 5317 5318
*/

double ha_partition::read_time(uint index, uint ranges, ha_rows rows)
{
  DBUG_ENTER("ha_partition::read_time");
5319

5320 5321 5322 5323
  DBUG_RETURN(m_file[0]->read_time(index, ranges, rows));
}

/*
5324 5325 5326 5327 5328 5329 5330 5331 5332 5333
  Find number of records in a range

  SYNOPSIS
    records_in_range()
    inx                  Index number
    min_key              Start of range
    max_key              End of range

  RETURN VALUE
    Number of rows in range
5334

5335 5336 5337 5338
  DESCRIPTION
    Given a starting key, and an ending key estimate the number of rows that
    will exist between the two. end_key may be empty which in case determine
    if start_key matches any rows.
5339

5340 5341 5342 5343 5344
    Called from opt_range.cc by check_quick_keys().

    monty: MUST be called for each range and added.
          Note that MySQL will assume that if this returns 0 there is no
          matching rows for the range!
5345 5346 5347 5348 5349 5350
*/

ha_rows ha_partition::records_in_range(uint inx, key_range *min_key,
				       key_range *max_key)
{
  handler **file;
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5351
  ha_rows in_range= 0;
5352 5353 5354 5355 5356
  DBUG_ENTER("ha_partition::records_in_range");

  file= m_file;
  do
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5357 5358 5359 5360 5361 5362 5363
    if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
    {
      ha_rows tmp_in_range= (*file)->records_in_range(inx, min_key, max_key);
      if (tmp_in_range == HA_POS_ERROR)
        DBUG_RETURN(tmp_in_range);
      in_range+= tmp_in_range;
    }
5364 5365 5366 5367 5368
  } while (*(++file));
  DBUG_RETURN(in_range);
}


5369 5370 5371 5372 5373 5374 5375 5376 5377 5378
/*
  Estimate upper bound of number of rows

  SYNOPSIS
    estimate_rows_upper_bound()

  RETURN VALUE
    Number of rows
*/

5379 5380 5381 5382 5383 5384 5385 5386 5387
ha_rows ha_partition::estimate_rows_upper_bound()
{
  ha_rows rows, tot_rows= 0;
  handler **file;
  DBUG_ENTER("ha_partition::estimate_rows_upper_bound");

  file= m_file;
  do
  {
patg@govinda.patg.net's avatar
patg@govinda.patg.net committed
5388 5389 5390 5391 5392 5393 5394
    if (bitmap_is_set(&(m_part_info->used_partitions), (file - m_file)))
    {
      rows= (*file)->estimate_rows_upper_bound();
      if (rows == HA_POS_ERROR)
        DBUG_RETURN(HA_POS_ERROR);
      tot_rows+= rows;
    }
5395 5396 5397 5398 5399
  } while (*(++file));
  DBUG_RETURN(tot_rows);
}


5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437
/*
  Is it ok to switch to a new engine for this table

  SYNOPSIS
    can_switch_engine()

  RETURN VALUE
    TRUE                  Ok
    FALSE                 Not ok

  DESCRIPTION
    Used to ensure that tables with foreign key constraints are not moved
    to engines without foreign key support.
*/

bool ha_partition::can_switch_engines()
{
  handler **file;
  DBUG_ENTER("ha_partition::can_switch_engines");
 
  file= m_file;
  do
  {
    if (!(*file)->can_switch_engines())
      DBUG_RETURN(FALSE);
  } while (*(++file));
  DBUG_RETURN(TRUE);
}


/*
  Is table cache supported

  SYNOPSIS
    table_cache_type()

*/

5438 5439 5440
uint8 ha_partition::table_cache_type()
{
  DBUG_ENTER("ha_partition::table_cache_type");
5441

5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452
  DBUG_RETURN(m_file[0]->table_cache_type());
}


/****************************************************************************
                MODULE print messages
****************************************************************************/

const char *ha_partition::index_type(uint inx)
{
  DBUG_ENTER("ha_partition::index_type");
5453

5454 5455 5456 5457
  DBUG_RETURN(m_file[0]->index_type(inx));
}


5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473
enum row_type ha_partition::get_row_type() const
{
  handler **file;
  enum row_type type= (*m_file)->get_row_type();

  for (file= m_file, file++; *file; file++)
  {
    enum row_type part_type= (*file)->get_row_type();
    if (part_type != type)
      return ROW_TYPE_NOT_USED;
  }

  return type;
}


5474 5475 5476
void ha_partition::print_error(int error, myf errflag)
{
  DBUG_ENTER("ha_partition::print_error");
5477

5478
  /* Should probably look for my own errors first */
5479
  DBUG_PRINT("enter", ("error: %d", error));
5480

5481
  if (error == HA_ERR_NO_PARTITION_FOUND)
5482
    m_part_info->print_no_partition_found(table);
5483
  else
5484
    m_file[m_last_part]->print_error(error, errflag);
5485 5486 5487 5488 5489 5490 5491
  DBUG_VOID_RETURN;
}


bool ha_partition::get_error_message(int error, String *buf)
{
  DBUG_ENTER("ha_partition::get_error_message");
5492

5493
  /* Should probably look for my own errors first */
5494
  DBUG_RETURN(m_file[m_last_part]->get_error_message(error, buf));
5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515
}


/****************************************************************************
                MODULE handler characteristics
****************************************************************************/
/*
  If frm_error() is called then we will use this to to find out what file
  extensions exist for the storage engine. This is also used by the default
  rename_table and delete_table method in handler.cc.
*/

static const char *ha_partition_ext[]=
{
  ha_par_ext, NullS
};

const char **ha_partition::bas_ext() const
{ return ha_partition_ext; }


5516 5517
uint ha_partition::min_of_the_max_uint(
                       uint (handler::*operator_func)(void) const) const
5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564
{
  handler **file;
  uint min_of_the_max= ((*m_file)->*operator_func)();

  for (file= m_file+1; *file; file++)
  {
    uint tmp= ((*file)->*operator_func)();
    set_if_smaller(min_of_the_max, tmp);
  }
  return min_of_the_max;
}


uint ha_partition::max_supported_key_parts() const
{
  return min_of_the_max_uint(&handler::max_supported_key_parts);
}


uint ha_partition::max_supported_key_length() const
{
  return min_of_the_max_uint(&handler::max_supported_key_length);
}


uint ha_partition::max_supported_key_part_length() const
{
  return min_of_the_max_uint(&handler::max_supported_key_part_length);
}


uint ha_partition::max_supported_record_length() const
{
  return min_of_the_max_uint(&handler::max_supported_record_length);
}


uint ha_partition::max_supported_keys() const
{
  return min_of_the_max_uint(&handler::max_supported_keys);
}


uint ha_partition::extra_rec_buf_length() const
{
  handler **file;
  uint max= (*m_file)->extra_rec_buf_length();
5565

5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576
  for (file= m_file, file++; *file; file++)
    if (max < (*file)->extra_rec_buf_length())
      max= (*file)->extra_rec_buf_length();
  return max;
}


uint ha_partition::min_record_length(uint options) const
{
  handler **file;
  uint max= (*m_file)->min_record_length(options);
5577

5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588
  for (file= m_file, file++; *file; file++)
    if (max < (*file)->min_record_length(options))
      max= (*file)->min_record_length(options);
  return max;
}


/****************************************************************************
                MODULE compare records
****************************************************************************/
/*
5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605
  Compare two positions

  SYNOPSIS
    cmp_ref()
    ref1                   First position
    ref2                   Second position

  RETURN VALUE
    <0                     ref1 < ref2
    0                      Equal
    >0                     ref1 > ref2

  DESCRIPTION
    We get two references and need to check if those records are the same.
    If they belong to different partitions we decide that they are not
    the same record. Otherwise we use the particular handler to decide if
    they are the same. Sort in partition id order if not equal.
5606 5607
*/

5608
int ha_partition::cmp_ref(const uchar *ref1, const uchar *ref2)
5609 5610 5611 5612 5613
{
  uint part_id;
  my_ptrdiff_t diff1, diff2;
  handler *file;
  DBUG_ENTER("ha_partition::cmp_ref");
5614

5615 5616
  if ((ref1[0] == ref2[0]) && (ref1[1] == ref2[1]))
  {
5617
    part_id= uint2korr(ref1);
5618 5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644
    file= m_file[part_id];
    DBUG_ASSERT(part_id < m_tot_parts);
    DBUG_RETURN(file->cmp_ref((ref1 + PARTITION_BYTES_IN_POS),
			      (ref2 + PARTITION_BYTES_IN_POS)));
  }
  diff1= ref2[1] - ref1[1];
  diff2= ref2[0] - ref1[0];
  if (diff1 > 0)
  {
    DBUG_RETURN(-1);
  }
  if (diff1 < 0)
  {
    DBUG_RETURN(+1);
  }
  if (diff2 > 0)
  {
    DBUG_RETURN(-1);
  }
  DBUG_RETURN(+1);
}


/****************************************************************************
                MODULE auto increment
****************************************************************************/

5645
void ha_partition::restore_auto_increment(ulonglong)
5646 5647
{
  DBUG_ENTER("ha_partition::restore_auto_increment");
5648

5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659
  DBUG_VOID_RETURN;
}


/*
  This method is called by update_auto_increment which in turn is called
  by the individual handlers as part of write_row. We will always let
  the first handler keep track of the auto increment value for all
  partitions.
*/

5660 5661 5662 5663
void ha_partition::get_auto_increment(ulonglong offset, ulonglong increment,
                                      ulonglong nb_desired_values,
                                      ulonglong *first_value,
                                      ulonglong *nb_reserved_values)
5664
{
5665
  ulonglong first_value_part, last_value_part, nb_reserved_values_part,
monty@mysql.com's avatar
monty@mysql.com committed
5666
    last_value= ~ (ulonglong) 0;
5667
  handler **pos, **end;
antony@pcg5ppc.xiphis.org's avatar
antony@pcg5ppc.xiphis.org committed
5668
  bool retry= TRUE;
5669
  DBUG_ENTER("ha_partition::get_auto_increment");
5670

antony@pcg5ppc.xiphis.org's avatar
antony@pcg5ppc.xiphis.org committed
5671
again:
5672
  for (pos=m_file, end= m_file+ m_tot_parts; pos != end ; pos++)
5673
  {
5674
    first_value_part= *first_value;
5675 5676
    (*pos)->get_auto_increment(offset, increment, nb_desired_values,
                               &first_value_part, &nb_reserved_values_part);
5677 5678 5679
    if (first_value_part == ~(ulonglong)(0)) // error in one partition
    {
      *first_value= first_value_part;
antony@pcg5ppc.xiphis.org's avatar
antony@pcg5ppc.xiphis.org committed
5680 5681
      sql_print_error("Partition failed to reserve auto_increment value");
      DBUG_VOID_RETURN;
5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693
    }
    /*
      Partition has reserved an interval. Intersect it with the intervals
      already reserved for the previous partitions.
    */
    last_value_part= (nb_reserved_values_part == ULONGLONG_MAX) ?
      ULONGLONG_MAX : (first_value_part + nb_reserved_values_part * increment);
    set_if_bigger(*first_value, first_value_part);
    set_if_smaller(last_value, last_value_part);
  }
  if (last_value < *first_value) /* empty intersection, error */
  {
antony@pcg5ppc.xiphis.org's avatar
antony@pcg5ppc.xiphis.org committed
5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712
    /*
      When we have an empty intersection, it means that one or more
      partitions may have a significantly different autoinc next value.
      We should not fail here - it just means that we should try to
      find a new reservation making use of the current *first_value
      wbich should now be compatible with all partitions.
    */
    if (retry)
    {
      retry= FALSE;
      last_value= ~ (ulonglong) 0;
      release_auto_increment();
      goto again;
    }
    /*
      We should not get here.
    */
    sql_print_error("Failed to calculate auto_increment value for partition");
    
5713
    *first_value= ~(ulonglong)(0);
5714
  }
monty@mysql.com's avatar
monty@mysql.com committed
5715 5716 5717
  if (increment)                                // If not check for values
    *nb_reserved_values= (last_value == ULONGLONG_MAX) ?
      ULONGLONG_MAX : ((last_value - *first_value) / increment);
5718
  DBUG_VOID_RETURN;
5719 5720
}

5721 5722 5723 5724 5725 5726
void ha_partition::release_auto_increment()
{
  DBUG_ENTER("ha_partition::release_auto_increment");

  for (uint i= 0; i < m_tot_parts; i++)
  {
5727
    m_file[i]->ha_release_auto_increment();
5728 5729 5730
  }
  DBUG_VOID_RETURN;
}
5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741

/****************************************************************************
                MODULE initialise handler for HANDLER call
****************************************************************************/

void ha_partition::init_table_handle_for_HANDLER()
{
  return;
}


5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762
/****************************************************************************
                MODULE enable/disable indexes
****************************************************************************/

/*
  Disable indexes for a while
  SYNOPSIS
    disable_indexes()
    mode                      Mode
  RETURN VALUES
    0                         Success
    != 0                      Error
*/

int ha_partition::disable_indexes(uint mode)
{
  handler **file;
  int error= 0;

  for (file= m_file; *file; file++)
  {
5763
    if ((error= (*file)->ha_disable_indexes(mode)))
5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786
      break;
  }
  return error;
}


/*
  Enable indexes again
  SYNOPSIS
    enable_indexes()
    mode                      Mode
  RETURN VALUES
    0                         Success
    != 0                      Error
*/

int ha_partition::enable_indexes(uint mode)
{
  handler **file;
  int error= 0;

  for (file= m_file; *file; file++)
  {
5787
    if ((error= (*file)->ha_enable_indexes(mode)))
5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817
      break;
  }
  return error;
}


/*
  Check if indexes are disabled
  SYNOPSIS
    indexes_are_disabled()

  RETURN VALUES
    0                      Indexes are enabled
    != 0                   Indexes are disabled
*/

int ha_partition::indexes_are_disabled(void)
{
  handler **file;
  int error= 0;

  for (file= m_file; *file; file++)
  {
    if ((error= (*file)->indexes_are_disabled()))
      break;
  }
  return error;
}


5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838
/****************************************************************************
                MODULE Partition Share
****************************************************************************/
/*
  Service routines for ... methods.
-------------------------------------------------------------------------
  Variables for partition share methods. A hash used to track open tables.
  A mutex for the hash table and an init variable to check if hash table
  is initialised.
  There is also a constant ending of the partition handler file name.
*/

#ifdef NOT_USED
static HASH partition_open_tables;
static pthread_mutex_t partition_mutex;
static int partition_init= 0;


/*
  Function we use in the creation of our hash to get key.
*/
5839

5840
static uchar *partition_get_key(PARTITION_SHARE *share, size_t *length,
5841 5842 5843
			       my_bool not_used __attribute__ ((unused)))
{
  *length= share->table_name_length;
5844
  return (uchar *) share->table_name;
5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883
}

/*
  Example of simple lock controls. The "share" it creates is structure we
  will pass to each partition handler. Do you have to have one of these?
  Well, you have pieces that are used for locking, and they are needed to
  function.
*/

static PARTITION_SHARE *get_share(const char *table_name, TABLE *table)
{
  PARTITION_SHARE *share;
  uint length;
  char *tmp_name;

  /*
    So why does this exist? There is no way currently to init a storage
    engine.
    Innodb and BDB both have modifications to the server to allow them to
    do this. Since you will not want to do this, this is probably the next
    best method.
  */
  if (!partition_init)
  {
    /* Hijack a mutex for init'ing the storage engine */
    pthread_mutex_lock(&LOCK_mysql_create_db);
    if (!partition_init)
    {
      partition_init++;
      VOID(pthread_mutex_init(&partition_mutex, MY_MUTEX_INIT_FAST));
      (void) hash_init(&partition_open_tables, system_charset_info, 32, 0, 0,
		       (hash_get_key) partition_get_key, 0, 0);
    }
    pthread_mutex_unlock(&LOCK_mysql_create_db);
  }
  pthread_mutex_lock(&partition_mutex);
  length= (uint) strlen(table_name);

  if (!(share= (PARTITION_SHARE *) hash_search(&partition_open_tables,
5884
					       (uchar *) table_name, length)))
5885 5886 5887
  {
    if (!(share= (PARTITION_SHARE *)
	  my_multi_malloc(MYF(MY_WME | MY_ZEROFILL),
5888 5889
			  &share, (uint) sizeof(*share),
			  &tmp_name, (uint) length + 1, NullS)))
5890 5891 5892 5893 5894 5895 5896 5897 5898
    {
      pthread_mutex_unlock(&partition_mutex);
      return NULL;
    }

    share->use_count= 0;
    share->table_name_length= length;
    share->table_name= tmp_name;
    strmov(share->table_name, table_name);
5899
    if (my_hash_insert(&partition_open_tables, (uchar *) share))
5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910
      goto error;
    thr_lock_init(&share->lock);
    pthread_mutex_init(&share->mutex, MY_MUTEX_INIT_FAST);
  }
  share->use_count++;
  pthread_mutex_unlock(&partition_mutex);

  return share;

error:
  pthread_mutex_unlock(&partition_mutex);
5911
  my_free((uchar*) share, MYF(0));
5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927

  return NULL;
}


/*
  Free lock controls. We call this whenever we close a table. If the table
  had the last reference to the share then we free memory associated with
  it.
*/

static int free_share(PARTITION_SHARE *share)
{
  pthread_mutex_lock(&partition_mutex);
  if (!--share->use_count)
  {
5928
    hash_delete(&partition_open_tables, (uchar *) share);
5929 5930
    thr_lock_delete(&share->lock);
    pthread_mutex_destroy(&share->mutex);
5931
    my_free((uchar*) share, MYF(0));
5932 5933 5934 5935 5936 5937
  }
  pthread_mutex_unlock(&partition_mutex);

  return 0;
}
#endif /* NOT_USED */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5938

5939
struct st_mysql_storage_engine partition_storage_engine=
5940
{ MYSQL_HANDLERTON_INTERFACE_VERSION };
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5941 5942 5943 5944

mysql_declare_plugin(partition)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
5945 5946
  &partition_storage_engine,
  "partition",
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5947
  "Mikael Ronstrom, MySQL AB",
5948
  "Partition Storage Engine Helper",
5949
  PLUGIN_LICENSE_GPL,
5950
  partition_initialize, /* Plugin Init */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5951
  NULL, /* Plugin Deinit */
monty@mysql.com's avatar
monty@mysql.com committed
5952
  0x0100, /* 1.0 */
5953 5954 5955
  NULL,                       /* status variables                */
  NULL,                       /* system variables                */
  NULL                        /* config options                  */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5956 5957 5958 5959
}
mysql_declare_plugin_end;

#endif