| 1 | /* Copyright (c) 2005, 2017, Oracle and/or its affiliates. |
| 2 | Copyright (c) 2009, 2018, MariaDB |
| 3 | |
| 4 | This program is free software; you can redistribute it and/or modify |
| 5 | it under the terms of the GNU General Public License as published by |
| 6 | the Free Software Foundation; version 2 of the License. |
| 7 | |
| 8 | This program is distributed in the hope that it will be useful, |
| 9 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 11 | GNU General Public License for more details. |
| 12 | |
| 13 | You should have received a copy of the GNU General Public License |
| 14 | along with this program; if not, write to the Free Software |
| 15 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA */ |
| 16 | |
| 17 | /* |
| 18 | This file is a container for general functionality related |
| 19 | to partitioning introduced in MySQL version 5.1. It contains functionality |
| 20 | used by all handlers that support partitioning, such as |
| 21 | the partitioning handler itself and the NDB handler. |
| 22 | (Much of the code in this file has been split into partition_info.cc and |
| 23 | the header files partition_info.h + partition_element.h + sql_partition.h) |
| 24 | |
| 25 | The first version was written by Mikael Ronstrom 2004-2006. |
| 26 | Various parts of the optimizer code was written by Sergey Petrunia. |
| 27 | Code have been maintained by Mattias Jonsson. |
| 28 | The second version was written by Mikael Ronstrom 2006-2007 with some |
| 29 | final fixes for partition pruning in 2008-2009 with assistance from Sergey |
| 30 | Petrunia and Mattias Jonsson. |
| 31 | |
| 32 | The first version supports RANGE partitioning, LIST partitioning, HASH |
| 33 | partitioning and composite partitioning (hereafter called subpartitioning) |
| 34 | where each RANGE/LIST partitioning is HASH partitioned. The hash function |
| 35 | can either be supplied by the user or by only a list of fields (also |
| 36 | called KEY partitioning), where the MySQL server will use an internal |
| 37 | hash function. |
| 38 | There are quite a few defaults that can be used as well. |
| 39 | |
| 40 | The second version introduces a new variant of RANGE and LIST partitioning |
| 41 | which is often referred to as column lists in the code variables. This |
| 42 | enables a user to specify a set of columns and their concatenated value |
| 43 | as the partition value. By comparing the concatenation of these values |
| 44 | the proper partition can be choosen. |
| 45 | */ |
| 46 | |
| 47 | /* Some general useful functions */ |
| 48 | |
| 49 | #define MYSQL_LEX 1 |
| 50 | #include "mariadb.h" |
| 51 | #include "sql_priv.h" |
| 52 | #include "sql_partition.h" |
| 53 | #include "key.h" // key_restore |
| 54 | #include "sql_parse.h" // parse_sql |
| 55 | #include "sql_cache.h" // query_cache_invalidate3 |
| 56 | #include "lock.h" // mysql_lock_remove |
| 57 | #include "sql_show.h" // append_identifier |
| 58 | #include <m_ctype.h> |
| 59 | #include "transaction.h" |
| 60 | #include "debug_sync.h" |
| 61 | |
| 62 | #include "sql_base.h" // close_all_tables_for_name |
| 63 | #include "sql_table.h" // build_table_filename, |
| 64 | // build_table_shadow_filename, |
| 65 | // table_to_filename |
| 66 | // mysql_*_alter_copy_data |
| 67 | #include "opt_range.h" // store_key_image_to_rec |
| 68 | #include "sql_alter.h" // Alter_table_ctx |
| 69 | #include "sql_select.h" |
| 70 | #include "sql_tablespace.h" // check_tablespace_name |
| 71 | #include "tztime.h" // my_tz_OFFSET0 |
| 72 | |
| 73 | #include <algorithm> |
| 74 | using std::max; |
| 75 | using std::min; |
| 76 | |
| 77 | #ifdef WITH_PARTITION_STORAGE_ENGINE |
| 78 | #include "ha_partition.h" |
| 79 | |
| 80 | #define ERROR_INJECT_CRASH(code) \ |
| 81 | DBUG_EVALUATE_IF(code, (DBUG_SUICIDE(), 0), 0) |
| 82 | #define ERROR_INJECT_ERROR(code) \ |
| 83 | DBUG_EVALUATE_IF(code, (my_error(ER_UNKNOWN_ERROR, MYF(0)), TRUE), 0) |
| 84 | |
| 85 | /* |
| 86 | Partition related functions declarations and some static constants; |
| 87 | */ |
| 88 | static int get_partition_id_list_col(partition_info *, uint32 *, longlong *); |
| 89 | static int get_partition_id_list(partition_info *, uint32 *, longlong *); |
| 90 | static int get_partition_id_range_col(partition_info *, uint32 *, longlong *); |
| 91 | static int get_partition_id_range(partition_info *, uint32 *, longlong *); |
| 92 | static int vers_get_partition_id(partition_info *, uint32 *, longlong *); |
| 93 | static int get_part_id_charset_func_part(partition_info *, uint32 *, longlong *); |
| 94 | static int get_part_id_charset_func_subpart(partition_info *, uint32 *); |
| 95 | static int get_partition_id_hash_nosub(partition_info *, uint32 *, longlong *); |
| 96 | static int get_partition_id_key_nosub(partition_info *, uint32 *, longlong *); |
| 97 | static int get_partition_id_linear_hash_nosub(partition_info *, uint32 *, longlong *); |
| 98 | static int get_partition_id_linear_key_nosub(partition_info *, uint32 *, longlong *); |
| 99 | static int get_partition_id_with_sub(partition_info *, uint32 *, longlong *); |
| 100 | static int get_partition_id_hash_sub(partition_info *part_info, uint32 *part_id); |
| 101 | static int get_partition_id_key_sub(partition_info *part_info, uint32 *part_id); |
| 102 | static int get_partition_id_linear_hash_sub(partition_info *part_info, uint32 *part_id); |
| 103 | static int get_partition_id_linear_key_sub(partition_info *part_info, uint32 *part_id); |
| 104 | static uint32 get_next_partition_via_walking(PARTITION_ITERATOR*); |
| 105 | static void set_up_range_analysis_info(partition_info *part_info); |
| 106 | static uint32 get_next_subpartition_via_walking(PARTITION_ITERATOR*); |
| 107 | #endif |
| 108 | |
| 109 | uint32 get_next_partition_id_range(PARTITION_ITERATOR* part_iter); |
| 110 | uint32 get_next_partition_id_list(PARTITION_ITERATOR* part_iter); |
| 111 | |
| 112 | #ifdef WITH_PARTITION_STORAGE_ENGINE |
| 113 | static int get_part_iter_for_interval_via_mapping(partition_info *, bool, |
| 114 | uint32 *, uchar *, uchar *, uint, uint, uint, PARTITION_ITERATOR *); |
| 115 | static int get_part_iter_for_interval_cols_via_map(partition_info *, bool, |
| 116 | uint32 *, uchar *, uchar *, uint, uint, uint, PARTITION_ITERATOR *); |
| 117 | static int get_part_iter_for_interval_via_walking(partition_info *, bool, |
| 118 | uint32 *, uchar *, uchar *, uint, uint, uint, PARTITION_ITERATOR *); |
| 119 | static int cmp_rec_and_tuple(part_column_list_val *val, uint32 nvals_in_rec); |
| 120 | static int cmp_rec_and_tuple_prune(part_column_list_val *val, |
| 121 | uint32 n_vals_in_rec, |
| 122 | bool is_left_endpoint, |
| 123 | bool include_endpoint); |
| 124 | |
| 125 | /* |
| 126 | Convert constants in VALUES definition to the character set the |
| 127 | corresponding field uses. |
| 128 | |
| 129 | SYNOPSIS |
| 130 | convert_charset_partition_constant() |
| 131 | item Item to convert |
| 132 | cs Character set to convert to |
| 133 | |
| 134 | RETURN VALUE |
| 135 | NULL Error |
| 136 | item New converted item |
| 137 | */ |
| 138 | |
| 139 | Item* convert_charset_partition_constant(Item *item, CHARSET_INFO *cs) |
| 140 | { |
| 141 | THD *thd= current_thd; |
| 142 | Name_resolution_context *context= &thd->lex->current_select->context; |
| 143 | TABLE_LIST *save_list= context->table_list; |
| 144 | const char *save_where= thd->where; |
| 145 | |
| 146 | item= item->safe_charset_converter(thd, cs); |
| 147 | context->table_list= NULL; |
| 148 | thd->where= "convert character set partition constant" ; |
| 149 | if (!item || item->fix_fields(thd, (Item**)NULL)) |
| 150 | item= NULL; |
| 151 | thd->where= save_where; |
| 152 | context->table_list= save_list; |
| 153 | return item; |
| 154 | } |
| 155 | |
| 156 | |
| 157 | /** |
| 158 | A support function to check if a name is in a list of strings. |
| 159 | |
| 160 | @param name String searched for |
| 161 | @param list_names A list of names searched in |
| 162 | |
| 163 | @return True if if the name is in the list. |
| 164 | @retval true String found |
| 165 | @retval false String not found |
| 166 | */ |
| 167 | |
| 168 | static bool is_name_in_list(const char *name, List<const char> list_names) |
| 169 | { |
| 170 | List_iterator<const char> names_it(list_names); |
| 171 | uint num_names= list_names.elements; |
| 172 | uint i= 0; |
| 173 | |
| 174 | do |
| 175 | { |
| 176 | const char *list_name= names_it++; |
| 177 | if (!(my_strcasecmp(system_charset_info, name, list_name))) |
| 178 | return TRUE; |
| 179 | } while (++i < num_names); |
| 180 | return FALSE; |
| 181 | } |
| 182 | |
| 183 | |
| 184 | |
| 185 | /* |
| 186 | Set-up defaults for partitions. |
| 187 | |
| 188 | SYNOPSIS |
| 189 | partition_default_handling() |
| 190 | table Table object |
| 191 | part_info Partition info to set up |
| 192 | is_create_table_ind Is this part of a table creation |
| 193 | normalized_path Normalized path name of table and database |
| 194 | |
| 195 | RETURN VALUES |
| 196 | TRUE Error |
| 197 | FALSE Success |
| 198 | */ |
| 199 | |
| 200 | bool partition_default_handling(THD *thd, TABLE *table, partition_info *part_info, |
| 201 | bool is_create_table_ind, |
| 202 | const char *normalized_path) |
| 203 | { |
| 204 | DBUG_ENTER("partition_default_handling" ); |
| 205 | |
| 206 | if (!is_create_table_ind) |
| 207 | { |
| 208 | if (part_info->use_default_num_partitions) |
| 209 | { |
| 210 | if (table->file->get_no_parts(normalized_path, &part_info->num_parts)) |
| 211 | { |
| 212 | DBUG_RETURN(TRUE); |
| 213 | } |
| 214 | } |
| 215 | else if (part_info->is_sub_partitioned() && |
| 216 | part_info->use_default_num_subpartitions) |
| 217 | { |
| 218 | uint num_parts; |
| 219 | if (table->file->get_no_parts(normalized_path, &num_parts)) |
| 220 | { |
| 221 | DBUG_RETURN(TRUE); |
| 222 | } |
| 223 | DBUG_ASSERT(part_info->num_parts > 0); |
| 224 | DBUG_ASSERT((num_parts % part_info->num_parts) == 0); |
| 225 | part_info->num_subparts= num_parts / part_info->num_parts; |
| 226 | } |
| 227 | } |
| 228 | part_info->set_up_defaults_for_partitioning(thd, table->file, |
| 229 | NULL, 0U); |
| 230 | DBUG_RETURN(FALSE); |
| 231 | } |
| 232 | |
| 233 | |
| 234 | /* |
| 235 | A useful routine used by update/delete_row for partition handlers to |
| 236 | calculate the partition id. |
| 237 | |
| 238 | SYNOPSIS |
| 239 | get_part_for_buf() |
| 240 | buf Buffer of old record |
| 241 | rec0 Reference to table->record[0] |
| 242 | part_info Reference to partition information |
| 243 | out:part_id The returned partition id to delete from |
| 244 | |
| 245 | RETURN VALUE |
| 246 | 0 Success |
| 247 | > 0 Error code |
| 248 | |
| 249 | DESCRIPTION |
| 250 | Dependent on whether buf is not record[0] we need to prepare the |
| 251 | fields. Then we call the function pointer get_partition_id to |
| 252 | calculate the partition id. |
| 253 | */ |
| 254 | |
| 255 | int get_part_for_buf(const uchar *buf, const uchar *rec0, |
| 256 | partition_info *part_info, uint32 *part_id) |
| 257 | { |
| 258 | int error; |
| 259 | longlong func_value; |
| 260 | DBUG_ENTER("get_part_for_buf" ); |
| 261 | |
| 262 | if (buf == rec0) |
| 263 | { |
| 264 | error= part_info->get_partition_id(part_info, part_id, &func_value); |
| 265 | if (unlikely((error))) |
| 266 | goto err; |
| 267 | DBUG_PRINT("info" , ("Partition %d" , *part_id)); |
| 268 | } |
| 269 | else |
| 270 | { |
| 271 | Field **part_field_array= part_info->full_part_field_array; |
| 272 | part_info->table->move_fields(part_field_array, buf, rec0); |
| 273 | error= part_info->get_partition_id(part_info, part_id, &func_value); |
| 274 | part_info->table->move_fields(part_field_array, rec0, buf); |
| 275 | if (unlikely(error)) |
| 276 | goto err; |
| 277 | DBUG_PRINT("info" , ("Partition %d (path2)" , *part_id)); |
| 278 | } |
| 279 | DBUG_RETURN(0); |
| 280 | err: |
| 281 | part_info->err_value= func_value; |
| 282 | DBUG_RETURN(error); |
| 283 | } |
| 284 | |
| 285 | |
| 286 | /* |
| 287 | This method is used to set-up both partition and subpartitioning |
| 288 | field array and used for all types of partitioning. |
| 289 | It is part of the logic around fix_partition_func. |
| 290 | |
| 291 | SYNOPSIS |
| 292 | set_up_field_array() |
| 293 | table TABLE object for which partition fields are set-up |
| 294 | sub_part Is the table subpartitioned as well |
| 295 | |
| 296 | RETURN VALUE |
| 297 | TRUE Error, some field didn't meet requirements |
| 298 | FALSE Ok, partition field array set-up |
| 299 | |
| 300 | DESCRIPTION |
| 301 | |
| 302 | A great number of functions below here is part of the fix_partition_func |
| 303 | method. It is used to set up the partition structures for execution from |
| 304 | openfrm. It is called at the end of the openfrm when the table struct has |
| 305 | been set-up apart from the partition information. |
| 306 | It involves: |
| 307 | 1) Setting arrays of fields for the partition functions. |
| 308 | 2) Setting up binary search array for LIST partitioning |
| 309 | 3) Setting up array for binary search for RANGE partitioning |
| 310 | 4) Setting up key_map's to assist in quick evaluation whether one |
| 311 | can deduce anything from a given index of what partition to use |
| 312 | 5) Checking whether a set of partitions can be derived from a range on |
| 313 | a field in the partition function. |
| 314 | As part of doing this there is also a great number of error controls. |
| 315 | This is actually the place where most of the things are checked for |
| 316 | partition information when creating a table. |
| 317 | Things that are checked includes |
| 318 | 1) All fields of partition function in Primary keys and unique indexes |
| 319 | (if not supported) |
| 320 | |
| 321 | |
| 322 | Create an array of partition fields (NULL terminated). Before this method |
| 323 | is called fix_fields or find_table_in_sef has been called to set |
| 324 | GET_FIXED_FIELDS_FLAG on all fields that are part of the partition |
| 325 | function. |
| 326 | */ |
| 327 | |
| 328 | static bool set_up_field_array(THD *thd, TABLE *table, |
| 329 | bool is_sub_part) |
| 330 | { |
| 331 | Field **ptr, *field, **field_array; |
| 332 | uint num_fields= 0; |
| 333 | uint size_field_array; |
| 334 | uint i= 0; |
| 335 | uint inx; |
| 336 | partition_info *part_info= table->part_info; |
| 337 | int result= FALSE; |
| 338 | DBUG_ENTER("set_up_field_array" ); |
| 339 | |
| 340 | ptr= table->field; |
| 341 | while ((field= *(ptr++))) |
| 342 | { |
| 343 | if (field->flags & GET_FIXED_FIELDS_FLAG) |
| 344 | num_fields++; |
| 345 | } |
| 346 | if (unlikely(num_fields > MAX_REF_PARTS)) |
| 347 | { |
| 348 | char *err_str; |
| 349 | if (is_sub_part) |
| 350 | err_str= (char*)"subpartition function" ; |
| 351 | else |
| 352 | err_str= (char*)"partition function" ; |
| 353 | my_error(ER_TOO_MANY_PARTITION_FUNC_FIELDS_ERROR, MYF(0), err_str); |
| 354 | DBUG_RETURN(TRUE); |
| 355 | } |
| 356 | if (num_fields == 0) |
| 357 | { |
| 358 | /* |
| 359 | We are using hidden key as partitioning field |
| 360 | */ |
| 361 | DBUG_ASSERT(!is_sub_part); |
| 362 | DBUG_RETURN(FALSE); |
| 363 | } |
| 364 | size_field_array= (num_fields+1)*sizeof(Field*); |
| 365 | field_array= (Field**) thd->calloc(size_field_array); |
| 366 | if (unlikely(!field_array)) |
| 367 | DBUG_RETURN(TRUE); |
| 368 | |
| 369 | ptr= table->field; |
| 370 | while ((field= *(ptr++))) |
| 371 | { |
| 372 | if (field->flags & GET_FIXED_FIELDS_FLAG) |
| 373 | { |
| 374 | field->flags&= ~GET_FIXED_FIELDS_FLAG; |
| 375 | field->flags|= FIELD_IN_PART_FUNC_FLAG; |
| 376 | if (likely(!result)) |
| 377 | { |
| 378 | if (!is_sub_part && part_info->column_list) |
| 379 | { |
| 380 | List_iterator<const char> it(part_info->part_field_list); |
| 381 | const char *field_name; |
| 382 | |
| 383 | DBUG_ASSERT(num_fields == part_info->part_field_list.elements); |
| 384 | inx= 0; |
| 385 | do |
| 386 | { |
| 387 | field_name= it++; |
| 388 | if (!my_strcasecmp(system_charset_info, |
| 389 | field_name, |
| 390 | field->field_name.str)) |
| 391 | break; |
| 392 | } while (++inx < num_fields); |
| 393 | if (inx == num_fields) |
| 394 | { |
| 395 | /* |
| 396 | Should not occur since it should already been checked in either |
| 397 | add_column_list_values, handle_list_of_fields, |
| 398 | check_partition_info etc. |
| 399 | */ |
| 400 | DBUG_ASSERT(0); |
| 401 | my_error(ER_FIELD_NOT_FOUND_PART_ERROR, MYF(0)); |
| 402 | result= TRUE; |
| 403 | continue; |
| 404 | } |
| 405 | } |
| 406 | else |
| 407 | inx= i; |
| 408 | field_array[inx]= field; |
| 409 | i++; |
| 410 | |
| 411 | /* |
| 412 | We check that the fields are proper. It is required for each |
| 413 | field in a partition function to: |
| 414 | 1) Not be a BLOB of any type |
| 415 | A BLOB takes too long time to evaluate so we don't want it for |
| 416 | performance reasons. |
| 417 | */ |
| 418 | |
| 419 | if (unlikely(field->flags & BLOB_FLAG)) |
| 420 | { |
| 421 | my_error(ER_BLOB_FIELD_IN_PART_FUNC_ERROR, MYF(0)); |
| 422 | result= TRUE; |
| 423 | } |
| 424 | } |
| 425 | } |
| 426 | } |
| 427 | field_array[num_fields]= 0; |
| 428 | if (!is_sub_part) |
| 429 | { |
| 430 | part_info->part_field_array= field_array; |
| 431 | part_info->num_part_fields= num_fields; |
| 432 | } |
| 433 | else |
| 434 | { |
| 435 | part_info->subpart_field_array= field_array; |
| 436 | part_info->num_subpart_fields= num_fields; |
| 437 | } |
| 438 | DBUG_RETURN(result); |
| 439 | } |
| 440 | |
| 441 | |
| 442 | |
| 443 | /* |
| 444 | Create a field array including all fields of both the partitioning and the |
| 445 | subpartitioning functions. |
| 446 | |
| 447 | SYNOPSIS |
| 448 | create_full_part_field_array() |
| 449 | thd Thread handle |
| 450 | table TABLE object for which partition fields are set-up |
| 451 | part_info Reference to partitioning data structure |
| 452 | |
| 453 | RETURN VALUE |
| 454 | TRUE Memory allocation of field array failed |
| 455 | FALSE Ok |
| 456 | |
| 457 | DESCRIPTION |
| 458 | If there is no subpartitioning then the same array is used as for the |
| 459 | partitioning. Otherwise a new array is built up using the flag |
| 460 | FIELD_IN_PART_FUNC in the field object. |
| 461 | This function is called from fix_partition_func |
| 462 | */ |
| 463 | |
| 464 | static bool create_full_part_field_array(THD *thd, TABLE *table, |
| 465 | partition_info *part_info) |
| 466 | { |
| 467 | bool result= FALSE; |
| 468 | Field **ptr; |
| 469 | my_bitmap_map *bitmap_buf; |
| 470 | DBUG_ENTER("create_full_part_field_array" ); |
| 471 | |
| 472 | if (!part_info->is_sub_partitioned()) |
| 473 | { |
| 474 | part_info->full_part_field_array= part_info->part_field_array; |
| 475 | part_info->num_full_part_fields= part_info->num_part_fields; |
| 476 | } |
| 477 | else |
| 478 | { |
| 479 | Field *field, **field_array; |
| 480 | uint num_part_fields=0, size_field_array; |
| 481 | ptr= table->field; |
| 482 | while ((field= *(ptr++))) |
| 483 | { |
| 484 | if (field->flags & FIELD_IN_PART_FUNC_FLAG) |
| 485 | num_part_fields++; |
| 486 | } |
| 487 | size_field_array= (num_part_fields+1)*sizeof(Field*); |
| 488 | field_array= (Field**) thd->calloc(size_field_array); |
| 489 | if (unlikely(!field_array)) |
| 490 | { |
| 491 | result= TRUE; |
| 492 | goto end; |
| 493 | } |
| 494 | num_part_fields= 0; |
| 495 | ptr= table->field; |
| 496 | while ((field= *(ptr++))) |
| 497 | { |
| 498 | if (field->flags & FIELD_IN_PART_FUNC_FLAG) |
| 499 | field_array[num_part_fields++]= field; |
| 500 | } |
| 501 | field_array[num_part_fields]=0; |
| 502 | part_info->full_part_field_array= field_array; |
| 503 | part_info->num_full_part_fields= num_part_fields; |
| 504 | } |
| 505 | |
| 506 | /* |
| 507 | Initialize the set of all fields used in partition and subpartition |
| 508 | expression. Required for testing of partition fields in write_set |
| 509 | when updating. We need to set all bits in read_set because the row |
| 510 | may need to be inserted in a different [sub]partition. |
| 511 | */ |
| 512 | if (!(bitmap_buf= (my_bitmap_map*) |
| 513 | thd->alloc(bitmap_buffer_size(table->s->fields)))) |
| 514 | { |
| 515 | result= TRUE; |
| 516 | goto end; |
| 517 | } |
| 518 | if (unlikely(my_bitmap_init(&part_info->full_part_field_set, bitmap_buf, |
| 519 | table->s->fields, FALSE))) |
| 520 | { |
| 521 | result= TRUE; |
| 522 | goto end; |
| 523 | } |
| 524 | /* |
| 525 | full_part_field_array may be NULL if storage engine supports native |
| 526 | partitioning. |
| 527 | */ |
| 528 | table->vcol_set= table->read_set= &part_info->full_part_field_set; |
| 529 | if ((ptr= part_info->full_part_field_array)) |
| 530 | for (; *ptr; ptr++) |
| 531 | { |
| 532 | if ((*ptr)->vcol_info) |
| 533 | table->mark_virtual_col(*ptr); |
| 534 | else |
| 535 | bitmap_fast_test_and_set(table->read_set, (*ptr)->field_index); |
| 536 | } |
| 537 | table->default_column_bitmaps(); |
| 538 | |
| 539 | end: |
| 540 | DBUG_RETURN(result); |
| 541 | } |
| 542 | |
| 543 | |
| 544 | /* |
| 545 | |
| 546 | Clear flag GET_FIXED_FIELDS_FLAG in all fields of a key previously set by |
| 547 | set_indicator_in_key_fields (always used in pairs). |
| 548 | |
| 549 | SYNOPSIS |
| 550 | clear_indicator_in_key_fields() |
| 551 | key_info Reference to find the key fields |
| 552 | |
| 553 | RETURN VALUE |
| 554 | NONE |
| 555 | |
| 556 | DESCRIPTION |
| 557 | These support routines is used to set/reset an indicator of all fields |
| 558 | in a certain key. It is used in conjunction with another support routine |
| 559 | that traverse all fields in the PF to find if all or some fields in the |
| 560 | PF is part of the key. This is used to check primary keys and unique |
| 561 | keys involve all fields in PF (unless supported) and to derive the |
| 562 | key_map's used to quickly decide whether the index can be used to |
| 563 | derive which partitions are needed to scan. |
| 564 | */ |
| 565 | |
| 566 | static void clear_indicator_in_key_fields(KEY *key_info) |
| 567 | { |
| 568 | KEY_PART_INFO *key_part; |
| 569 | uint key_parts= key_info->user_defined_key_parts, i; |
| 570 | for (i= 0, key_part=key_info->key_part; i < key_parts; i++, key_part++) |
| 571 | key_part->field->flags&= (~GET_FIXED_FIELDS_FLAG); |
| 572 | } |
| 573 | |
| 574 | |
| 575 | /* |
| 576 | Set flag GET_FIXED_FIELDS_FLAG in all fields of a key. |
| 577 | |
| 578 | SYNOPSIS |
| 579 | set_indicator_in_key_fields |
| 580 | key_info Reference to find the key fields |
| 581 | |
| 582 | RETURN VALUE |
| 583 | NONE |
| 584 | */ |
| 585 | |
| 586 | static void set_indicator_in_key_fields(KEY *key_info) |
| 587 | { |
| 588 | KEY_PART_INFO *key_part; |
| 589 | uint key_parts= key_info->user_defined_key_parts, i; |
| 590 | for (i= 0, key_part=key_info->key_part; i < key_parts; i++, key_part++) |
| 591 | key_part->field->flags|= GET_FIXED_FIELDS_FLAG; |
| 592 | } |
| 593 | |
| 594 | |
| 595 | /* |
| 596 | Check if all or some fields in partition field array is part of a key |
| 597 | previously used to tag key fields. |
| 598 | |
| 599 | SYNOPSIS |
| 600 | check_fields_in_PF() |
| 601 | ptr Partition field array |
| 602 | out:all_fields Is all fields of partition field array used in key |
| 603 | out:some_fields Is some fields of partition field array used in key |
| 604 | |
| 605 | RETURN VALUE |
| 606 | all_fields, some_fields |
| 607 | */ |
| 608 | |
| 609 | static void check_fields_in_PF(Field **ptr, bool *all_fields, |
| 610 | bool *some_fields) |
| 611 | { |
| 612 | DBUG_ENTER("check_fields_in_PF" ); |
| 613 | |
| 614 | *all_fields= TRUE; |
| 615 | *some_fields= FALSE; |
| 616 | if ((!ptr) || !(*ptr)) |
| 617 | { |
| 618 | *all_fields= FALSE; |
| 619 | DBUG_VOID_RETURN; |
| 620 | } |
| 621 | do |
| 622 | { |
| 623 | /* Check if the field of the PF is part of the current key investigated */ |
| 624 | if ((*ptr)->flags & GET_FIXED_FIELDS_FLAG) |
| 625 | *some_fields= TRUE; |
| 626 | else |
| 627 | *all_fields= FALSE; |
| 628 | } while (*(++ptr)); |
| 629 | DBUG_VOID_RETURN; |
| 630 | } |
| 631 | |
| 632 | |
| 633 | /* |
| 634 | Clear flag GET_FIXED_FIELDS_FLAG in all fields of the table. |
| 635 | This routine is used for error handling purposes. |
| 636 | |
| 637 | SYNOPSIS |
| 638 | clear_field_flag() |
| 639 | table TABLE object for which partition fields are set-up |
| 640 | |
| 641 | RETURN VALUE |
| 642 | NONE |
| 643 | */ |
| 644 | |
| 645 | static void clear_field_flag(TABLE *table) |
| 646 | { |
| 647 | Field **ptr; |
| 648 | DBUG_ENTER("clear_field_flag" ); |
| 649 | |
| 650 | for (ptr= table->field; *ptr; ptr++) |
| 651 | (*ptr)->flags&= (~GET_FIXED_FIELDS_FLAG); |
| 652 | DBUG_VOID_RETURN; |
| 653 | } |
| 654 | |
| 655 | |
| 656 | /* |
| 657 | find_field_in_table_sef finds the field given its name. All fields get |
| 658 | GET_FIXED_FIELDS_FLAG set. |
| 659 | |
| 660 | SYNOPSIS |
| 661 | handle_list_of_fields() |
| 662 | it A list of field names for the partition function |
| 663 | table TABLE object for which partition fields are set-up |
| 664 | part_info Reference to partitioning data structure |
| 665 | sub_part Is the table subpartitioned as well |
| 666 | |
| 667 | RETURN VALUE |
| 668 | TRUE Fields in list of fields not part of table |
| 669 | FALSE All fields ok and array created |
| 670 | |
| 671 | DESCRIPTION |
| 672 | This routine sets-up the partition field array for KEY partitioning, it |
| 673 | also verifies that all fields in the list of fields is actually a part of |
| 674 | the table. |
| 675 | |
| 676 | */ |
| 677 | |
| 678 | |
| 679 | static bool handle_list_of_fields(THD *thd, List_iterator<const char> it, |
| 680 | TABLE *table, |
| 681 | partition_info *part_info, |
| 682 | bool is_sub_part) |
| 683 | { |
| 684 | Field *field; |
| 685 | bool result; |
| 686 | const char *field_name; |
| 687 | bool is_list_empty= TRUE; |
| 688 | DBUG_ENTER("handle_list_of_fields" ); |
| 689 | |
| 690 | while ((field_name= it++)) |
| 691 | { |
| 692 | is_list_empty= FALSE; |
| 693 | field= find_field_in_table_sef(table, field_name); |
| 694 | if (likely(field != 0)) |
| 695 | field->flags|= GET_FIXED_FIELDS_FLAG; |
| 696 | else |
| 697 | { |
| 698 | my_error(ER_FIELD_NOT_FOUND_PART_ERROR, MYF(0)); |
| 699 | clear_field_flag(table); |
| 700 | result= TRUE; |
| 701 | goto end; |
| 702 | } |
| 703 | } |
| 704 | if (is_list_empty && part_info->part_type == HASH_PARTITION) |
| 705 | { |
| 706 | uint primary_key= table->s->primary_key; |
| 707 | if (primary_key != MAX_KEY) |
| 708 | { |
| 709 | uint num_key_parts= table->key_info[primary_key].user_defined_key_parts, i; |
| 710 | /* |
| 711 | In the case of an empty list we use primary key as partition key. |
| 712 | */ |
| 713 | for (i= 0; i < num_key_parts; i++) |
| 714 | { |
| 715 | Field *field= table->key_info[primary_key].key_part[i].field; |
| 716 | field->flags|= GET_FIXED_FIELDS_FLAG; |
| 717 | } |
| 718 | } |
| 719 | else |
| 720 | { |
| 721 | if (table->s->db_type()->partition_flags && |
| 722 | (table->s->db_type()->partition_flags() & HA_USE_AUTO_PARTITION) && |
| 723 | (table->s->db_type()->partition_flags() & HA_CAN_PARTITION)) |
| 724 | { |
| 725 | /* |
| 726 | This engine can handle automatic partitioning and there is no |
| 727 | primary key. In this case we rely on that the engine handles |
| 728 | partitioning based on a hidden key. Thus we allocate no |
| 729 | array for partitioning fields. |
| 730 | */ |
| 731 | DBUG_RETURN(FALSE); |
| 732 | } |
| 733 | else |
| 734 | { |
| 735 | my_error(ER_FIELD_NOT_FOUND_PART_ERROR, MYF(0)); |
| 736 | DBUG_RETURN(TRUE); |
| 737 | } |
| 738 | } |
| 739 | } |
| 740 | result= set_up_field_array(thd, table, is_sub_part); |
| 741 | end: |
| 742 | DBUG_RETURN(result); |
| 743 | } |
| 744 | |
| 745 | |
| 746 | /* |
| 747 | Support function to check if all VALUES * (expression) is of the |
| 748 | right sign (no signed constants when unsigned partition function) |
| 749 | |
| 750 | SYNOPSIS |
| 751 | check_signed_flag() |
| 752 | part_info Partition info object |
| 753 | |
| 754 | RETURN VALUES |
| 755 | 0 No errors due to sign errors |
| 756 | >0 Sign error |
| 757 | */ |
| 758 | |
| 759 | int check_signed_flag(partition_info *part_info) |
| 760 | { |
| 761 | int error= 0; |
| 762 | uint i= 0; |
| 763 | if (part_info->part_type != HASH_PARTITION && |
| 764 | part_info->part_expr->unsigned_flag) |
| 765 | { |
| 766 | List_iterator<partition_element> part_it(part_info->partitions); |
| 767 | do |
| 768 | { |
| 769 | partition_element *part_elem= part_it++; |
| 770 | |
| 771 | if (part_elem->signed_flag) |
| 772 | { |
| 773 | my_error(ER_PARTITION_CONST_DOMAIN_ERROR, MYF(0)); |
| 774 | error= ER_PARTITION_CONST_DOMAIN_ERROR; |
| 775 | break; |
| 776 | } |
| 777 | } while (++i < part_info->num_parts); |
| 778 | } |
| 779 | return error; |
| 780 | } |
| 781 | |
| 782 | /* |
| 783 | init_lex_with_single_table and end_lex_with_single_table |
| 784 | are now in sql_lex.cc |
| 785 | */ |
| 786 | |
| 787 | /* |
| 788 | The function uses a new feature in fix_fields where the flag |
| 789 | GET_FIXED_FIELDS_FLAG is set for all fields in the item tree. |
| 790 | This field must always be reset before returning from the function |
| 791 | since it is used for other purposes as well. |
| 792 | |
| 793 | SYNOPSIS |
| 794 | fix_fields_part_func() |
| 795 | thd The thread object |
| 796 | func_expr The item tree reference of the partition function |
| 797 | table The table object |
| 798 | part_info Reference to partitioning data structure |
| 799 | is_sub_part Is the table subpartitioned as well |
| 800 | is_create_table_ind Indicator of whether openfrm was called as part of |
| 801 | CREATE or ALTER TABLE |
| 802 | |
| 803 | RETURN VALUE |
| 804 | TRUE An error occurred, something was wrong with the |
| 805 | partition function. |
| 806 | FALSE Ok, a partition field array was created |
| 807 | |
| 808 | DESCRIPTION |
| 809 | This function is used to build an array of partition fields for the |
| 810 | partitioning function and subpartitioning function. The partitioning |
| 811 | function is an item tree that must reference at least one field in the |
| 812 | table. This is checked first in the parser that the function doesn't |
| 813 | contain non-cacheable parts (like a random function) and by checking |
| 814 | here that the function isn't a constant function. |
| 815 | |
| 816 | Calculate the number of fields in the partition function. |
| 817 | Use it allocate memory for array of Field pointers. |
| 818 | Initialise array of field pointers. Use information set when |
| 819 | calling fix_fields and reset it immediately after. |
| 820 | The get_fields_in_item_tree activates setting of bit in flags |
| 821 | on the field object. |
| 822 | */ |
| 823 | |
| 824 | static bool fix_fields_part_func(THD *thd, Item* func_expr, TABLE *table, |
| 825 | bool is_sub_part, bool is_create_table_ind) |
| 826 | { |
| 827 | partition_info *part_info= table->part_info; |
| 828 | bool result= TRUE; |
| 829 | int error; |
| 830 | LEX *old_lex= thd->lex; |
| 831 | LEX lex; |
| 832 | DBUG_ENTER("fix_fields_part_func" ); |
| 833 | |
| 834 | if (init_lex_with_single_table(thd, table, &lex)) |
| 835 | goto end; |
| 836 | table->get_fields_in_item_tree= true; |
| 837 | |
| 838 | func_expr->walk(&Item::change_context_processor, 0, &lex.select_lex.context); |
| 839 | thd->where= "partition function" ; |
| 840 | /* |
| 841 | In execution we must avoid the use of thd->change_item_tree since |
| 842 | we might release memory before statement is completed. We do this |
| 843 | by temporarily setting the stmt_arena->mem_root to be the mem_root |
| 844 | of the table object, this also ensures that any memory allocated |
| 845 | during fix_fields will not be released at end of execution of this |
| 846 | statement. Thus the item tree will remain valid also in subsequent |
| 847 | executions of this table object. We do however not at the moment |
| 848 | support allocations during execution of val_int so any item class |
| 849 | that does this during val_int must be disallowed as partition |
| 850 | function. |
| 851 | SEE Bug #21658 |
| 852 | |
| 853 | This is a tricky call to prepare for since it can have a large number |
| 854 | of interesting side effects, both desirable and undesirable. |
| 855 | */ |
| 856 | { |
| 857 | const bool save_agg_field= thd->lex->current_select->non_agg_field_used(); |
| 858 | const bool save_agg_func= thd->lex->current_select->agg_func_used(); |
| 859 | const nesting_map saved_allow_sum_func= thd->lex->allow_sum_func; |
| 860 | thd->lex->allow_sum_func= 0; |
| 861 | |
| 862 | if (likely(!(error= func_expr->fix_fields(thd, (Item**)&func_expr)))) |
| 863 | func_expr->walk(&Item::post_fix_fields_part_expr_processor, 0, NULL); |
| 864 | |
| 865 | /* |
| 866 | Restore agg_field/agg_func and allow_sum_func, |
| 867 | fix_fields should not affect mysql_select later, see Bug#46923. |
| 868 | */ |
| 869 | thd->lex->current_select->set_non_agg_field_used(save_agg_field); |
| 870 | thd->lex->current_select->set_agg_func_used(save_agg_func); |
| 871 | thd->lex->allow_sum_func= saved_allow_sum_func; |
| 872 | } |
| 873 | if (unlikely(error)) |
| 874 | { |
| 875 | DBUG_PRINT("info" , ("Field in partition function not part of table" )); |
| 876 | clear_field_flag(table); |
| 877 | goto end; |
| 878 | } |
| 879 | if (unlikely(func_expr->const_item())) |
| 880 | { |
| 881 | my_error(ER_WRONG_EXPR_IN_PARTITION_FUNC_ERROR, MYF(0)); |
| 882 | clear_field_flag(table); |
| 883 | goto end; |
| 884 | } |
| 885 | |
| 886 | /* |
| 887 | We don't allow creating partitions with expressions with non matching |
| 888 | arguments as a (sub)partitioning function, |
| 889 | but we want to allow such expressions when opening existing tables for |
| 890 | easier maintenance. This exception should be deprecated at some point |
| 891 | in future so that we always throw an error. |
| 892 | */ |
| 893 | if (func_expr->walk(&Item::check_valid_arguments_processor, 0, NULL)) |
| 894 | { |
| 895 | if (is_create_table_ind) |
| 896 | { |
| 897 | my_error(ER_WRONG_EXPR_IN_PARTITION_FUNC_ERROR, MYF(0)); |
| 898 | goto end; |
| 899 | } |
| 900 | else |
| 901 | push_warning(thd, Sql_condition::WARN_LEVEL_WARN, |
| 902 | ER_WRONG_EXPR_IN_PARTITION_FUNC_ERROR, |
| 903 | ER_THD(thd, ER_WRONG_EXPR_IN_PARTITION_FUNC_ERROR)); |
| 904 | } |
| 905 | |
| 906 | if (unlikely((!is_sub_part) && (error= check_signed_flag(part_info)))) |
| 907 | goto end; |
| 908 | result= set_up_field_array(thd, table, is_sub_part); |
| 909 | end: |
| 910 | end_lex_with_single_table(thd, table, old_lex); |
| 911 | func_expr->walk(&Item::change_context_processor, 0, 0); |
| 912 | DBUG_RETURN(result); |
| 913 | } |
| 914 | |
| 915 | |
| 916 | /* |
| 917 | Check that the primary key contains all partition fields if defined |
| 918 | |
| 919 | SYNOPSIS |
| 920 | check_primary_key() |
| 921 | table TABLE object for which partition fields are set-up |
| 922 | |
| 923 | RETURN VALUES |
| 924 | TRUE Not all fields in partitioning function was part |
| 925 | of primary key |
| 926 | FALSE Ok, all fields of partitioning function were part |
| 927 | of primary key |
| 928 | |
| 929 | DESCRIPTION |
| 930 | This function verifies that if there is a primary key that it contains |
| 931 | all the fields of the partition function. |
| 932 | This is a temporary limitation that will hopefully be removed after a |
| 933 | while. |
| 934 | */ |
| 935 | |
| 936 | static bool check_primary_key(TABLE *table) |
| 937 | { |
| 938 | uint primary_key= table->s->primary_key; |
| 939 | bool all_fields, some_fields; |
| 940 | bool result= FALSE; |
| 941 | DBUG_ENTER("check_primary_key" ); |
| 942 | |
| 943 | if (primary_key < MAX_KEY) |
| 944 | { |
| 945 | set_indicator_in_key_fields(table->key_info+primary_key); |
| 946 | check_fields_in_PF(table->part_info->full_part_field_array, |
| 947 | &all_fields, &some_fields); |
| 948 | clear_indicator_in_key_fields(table->key_info+primary_key); |
| 949 | if (unlikely(!all_fields)) |
| 950 | { |
| 951 | my_error(ER_UNIQUE_KEY_NEED_ALL_FIELDS_IN_PF,MYF(0),"PRIMARY KEY" ); |
| 952 | result= TRUE; |
| 953 | } |
| 954 | } |
| 955 | DBUG_RETURN(result); |
| 956 | } |
| 957 | |
| 958 | |
| 959 | /* |
| 960 | Check that unique keys contains all partition fields |
| 961 | |
| 962 | SYNOPSIS |
| 963 | check_unique_keys() |
| 964 | table TABLE object for which partition fields are set-up |
| 965 | |
| 966 | RETURN VALUES |
| 967 | TRUE Not all fields in partitioning function was part |
| 968 | of all unique keys |
| 969 | FALSE Ok, all fields of partitioning function were part |
| 970 | of unique keys |
| 971 | |
| 972 | DESCRIPTION |
| 973 | This function verifies that if there is a unique index that it contains |
| 974 | all the fields of the partition function. |
| 975 | This is a temporary limitation that will hopefully be removed after a |
| 976 | while. |
| 977 | */ |
| 978 | |
| 979 | static bool check_unique_keys(TABLE *table) |
| 980 | { |
| 981 | bool all_fields, some_fields; |
| 982 | bool result= FALSE; |
| 983 | uint keys= table->s->keys; |
| 984 | uint i; |
| 985 | DBUG_ENTER("check_unique_keys" ); |
| 986 | |
| 987 | for (i= 0; i < keys; i++) |
| 988 | { |
| 989 | if (table->key_info[i].flags & HA_NOSAME) //Unique index |
| 990 | { |
| 991 | set_indicator_in_key_fields(table->key_info+i); |
| 992 | check_fields_in_PF(table->part_info->full_part_field_array, |
| 993 | &all_fields, &some_fields); |
| 994 | clear_indicator_in_key_fields(table->key_info+i); |
| 995 | if (unlikely(!all_fields)) |
| 996 | { |
| 997 | my_error(ER_UNIQUE_KEY_NEED_ALL_FIELDS_IN_PF,MYF(0),"UNIQUE INDEX" ); |
| 998 | result= TRUE; |
| 999 | break; |
| 1000 | } |
| 1001 | } |
| 1002 | } |
| 1003 | DBUG_RETURN(result); |
| 1004 | } |
| 1005 | |
| 1006 | |
| 1007 | /* |
| 1008 | An important optimisation is whether a range on a field can select a subset |
| 1009 | of the partitions. |
| 1010 | A prerequisite for this to happen is that the PF is a growing function OR |
| 1011 | a shrinking function. |
| 1012 | This can never happen for a multi-dimensional PF. Thus this can only happen |
| 1013 | with PF with at most one field involved in the PF. |
| 1014 | The idea is that if the function is a growing function and you know that |
| 1015 | the field of the PF is 4 <= A <= 6 then we can convert this to a range |
| 1016 | in the PF instead by setting the range to PF(4) <= PF(A) <= PF(6). In the |
| 1017 | case of RANGE PARTITIONING and LIST PARTITIONING this can be used to |
| 1018 | calculate a set of partitions rather than scanning all of them. |
| 1019 | Thus the following prerequisites are there to check if sets of partitions |
| 1020 | can be found. |
| 1021 | 1) Only possible for RANGE and LIST partitioning (not for subpartitioning) |
| 1022 | 2) Only possible if PF only contains 1 field |
| 1023 | 3) Possible if PF is a growing function of the field |
| 1024 | 4) Possible if PF is a shrinking function of the field |
| 1025 | OBSERVATION: |
| 1026 | 1) IF f1(A) is a growing function AND f2(A) is a growing function THEN |
| 1027 | f1(A) + f2(A) is a growing function |
| 1028 | f1(A) * f2(A) is a growing function if f1(A) >= 0 and f2(A) >= 0 |
| 1029 | 2) IF f1(A) is a growing function and f2(A) is a shrinking function THEN |
| 1030 | f1(A) / f2(A) is a growing function if f1(A) >= 0 and f2(A) > 0 |
| 1031 | 3) IF A is a growing function then a function f(A) that removes the |
| 1032 | least significant portion of A is a growing function |
| 1033 | E.g. DATE(datetime) is a growing function |
| 1034 | MONTH(datetime) is not a growing/shrinking function |
| 1035 | 4) IF f1(A) is a growing function and f2(A) is a growing function THEN |
| 1036 | f1(f2(A)) and f2(f1(A)) are also growing functions |
| 1037 | 5) IF f1(A) is a shrinking function and f2(A) is a growing function THEN |
| 1038 | f1(f2(A)) is a shrinking function and f2(f1(A)) is a shrinking function |
| 1039 | 6) f1(A) = A is a growing function |
| 1040 | 7) f1(A) = A*a + b (where a and b are constants) is a growing function |
| 1041 | |
| 1042 | By analysing the item tree of the PF we can use these deducements and |
| 1043 | derive whether the PF is a growing function or a shrinking function or |
| 1044 | neither of it. |
| 1045 | |
| 1046 | If the PF is range capable then a flag is set on the table object |
| 1047 | indicating this to notify that we can use also ranges on the field |
| 1048 | of the PF to deduce a set of partitions if the fields of the PF were |
| 1049 | not all fully bound. |
| 1050 | |
| 1051 | SYNOPSIS |
| 1052 | check_range_capable_PF() |
| 1053 | table TABLE object for which partition fields are set-up |
| 1054 | |
| 1055 | DESCRIPTION |
| 1056 | Support for this is not implemented yet. |
| 1057 | */ |
| 1058 | |
| 1059 | void check_range_capable_PF(TABLE *table) |
| 1060 | { |
| 1061 | DBUG_ENTER("check_range_capable_PF" ); |
| 1062 | |
| 1063 | DBUG_VOID_RETURN; |
| 1064 | } |
| 1065 | |
| 1066 | |
| 1067 | /** |
| 1068 | Set up partition bitmaps |
| 1069 | |
| 1070 | @param thd Thread object |
| 1071 | @param part_info Reference to partitioning data structure |
| 1072 | |
| 1073 | @return Operation status |
| 1074 | @retval TRUE Memory allocation failure |
| 1075 | @retval FALSE Success |
| 1076 | |
| 1077 | Allocate memory for bitmaps of the partitioned table |
| 1078 | and initialise it. |
| 1079 | */ |
| 1080 | |
| 1081 | static bool set_up_partition_bitmaps(THD *thd, partition_info *part_info) |
| 1082 | { |
| 1083 | uint32 *bitmap_buf; |
| 1084 | uint bitmap_bits= part_info->num_subparts? |
| 1085 | (part_info->num_subparts* part_info->num_parts): |
| 1086 | part_info->num_parts; |
| 1087 | uint bitmap_bytes= bitmap_buffer_size(bitmap_bits); |
| 1088 | DBUG_ENTER("set_up_partition_bitmaps" ); |
| 1089 | |
| 1090 | DBUG_ASSERT(!part_info->bitmaps_are_initialized); |
| 1091 | |
| 1092 | /* Allocate for both read and lock_partitions */ |
| 1093 | if (unlikely(!(bitmap_buf= |
| 1094 | (uint32*) alloc_root(&part_info->table->mem_root, |
| 1095 | bitmap_bytes * 2)))) |
| 1096 | DBUG_RETURN(TRUE); |
| 1097 | |
| 1098 | my_bitmap_init(&part_info->read_partitions, bitmap_buf, bitmap_bits, FALSE); |
| 1099 | /* Use the second half of the allocated buffer for lock_partitions */ |
| 1100 | my_bitmap_init(&part_info->lock_partitions, bitmap_buf + (bitmap_bytes / 4), |
| 1101 | bitmap_bits, FALSE); |
| 1102 | part_info->bitmaps_are_initialized= TRUE; |
| 1103 | part_info->set_partition_bitmaps(NULL); |
| 1104 | DBUG_RETURN(FALSE); |
| 1105 | } |
| 1106 | |
| 1107 | |
| 1108 | /* |
| 1109 | Set up partition key maps |
| 1110 | |
| 1111 | SYNOPSIS |
| 1112 | set_up_partition_key_maps() |
| 1113 | table TABLE object for which partition fields are set-up |
| 1114 | part_info Reference to partitioning data structure |
| 1115 | |
| 1116 | RETURN VALUES |
| 1117 | None |
| 1118 | |
| 1119 | DESCRIPTION |
| 1120 | This function sets up a couple of key maps to be able to quickly check |
| 1121 | if an index ever can be used to deduce the partition fields or even |
| 1122 | a part of the fields of the partition function. |
| 1123 | We set up the following key_map's. |
| 1124 | PF = Partition Function |
| 1125 | 1) All fields of the PF is set even by equal on the first fields in the |
| 1126 | key |
| 1127 | 2) All fields of the PF is set if all fields of the key is set |
| 1128 | 3) At least one field in the PF is set if all fields is set |
| 1129 | 4) At least one field in the PF is part of the key |
| 1130 | */ |
| 1131 | |
| 1132 | static void set_up_partition_key_maps(TABLE *table, |
| 1133 | partition_info *part_info) |
| 1134 | { |
| 1135 | uint keys= table->s->keys; |
| 1136 | uint i; |
| 1137 | bool all_fields, some_fields; |
| 1138 | DBUG_ENTER("set_up_partition_key_maps" ); |
| 1139 | |
| 1140 | part_info->all_fields_in_PF.clear_all(); |
| 1141 | part_info->all_fields_in_PPF.clear_all(); |
| 1142 | part_info->all_fields_in_SPF.clear_all(); |
| 1143 | part_info->some_fields_in_PF.clear_all(); |
| 1144 | for (i= 0; i < keys; i++) |
| 1145 | { |
| 1146 | set_indicator_in_key_fields(table->key_info+i); |
| 1147 | check_fields_in_PF(part_info->full_part_field_array, |
| 1148 | &all_fields, &some_fields); |
| 1149 | if (all_fields) |
| 1150 | part_info->all_fields_in_PF.set_bit(i); |
| 1151 | if (some_fields) |
| 1152 | part_info->some_fields_in_PF.set_bit(i); |
| 1153 | if (part_info->is_sub_partitioned()) |
| 1154 | { |
| 1155 | check_fields_in_PF(part_info->part_field_array, |
| 1156 | &all_fields, &some_fields); |
| 1157 | if (all_fields) |
| 1158 | part_info->all_fields_in_PPF.set_bit(i); |
| 1159 | check_fields_in_PF(part_info->subpart_field_array, |
| 1160 | &all_fields, &some_fields); |
| 1161 | if (all_fields) |
| 1162 | part_info->all_fields_in_SPF.set_bit(i); |
| 1163 | } |
| 1164 | clear_indicator_in_key_fields(table->key_info+i); |
| 1165 | } |
| 1166 | DBUG_VOID_RETURN; |
| 1167 | } |
| 1168 | |
| 1169 | static bool check_no_constants(THD *, partition_info*) |
| 1170 | { |
| 1171 | return FALSE; |
| 1172 | } |
| 1173 | |
| 1174 | /* |
| 1175 | Support routines for check_list_constants used by qsort to sort the |
| 1176 | constant list expressions. One routine for integers and one for |
| 1177 | column lists. |
| 1178 | |
| 1179 | SYNOPSIS |
| 1180 | list_part_cmp() |
| 1181 | a First list constant to compare with |
| 1182 | b Second list constant to compare with |
| 1183 | |
| 1184 | RETURN VALUE |
| 1185 | +1 a > b |
| 1186 | 0 a == b |
| 1187 | -1 a < b |
| 1188 | */ |
| 1189 | |
| 1190 | extern "C" |
| 1191 | int partition_info_list_part_cmp(const void* a, const void* b) |
| 1192 | { |
| 1193 | longlong a1= ((LIST_PART_ENTRY*)a)->list_value; |
| 1194 | longlong b1= ((LIST_PART_ENTRY*)b)->list_value; |
| 1195 | if (a1 < b1) |
| 1196 | return -1; |
| 1197 | else if (a1 > b1) |
| 1198 | return +1; |
| 1199 | else |
| 1200 | return 0; |
| 1201 | } |
| 1202 | |
| 1203 | |
| 1204 | /* |
| 1205 | Compare two lists of column values in RANGE/LIST partitioning |
| 1206 | SYNOPSIS |
| 1207 | partition_info_compare_column_values() |
| 1208 | first First column list argument |
| 1209 | second Second column list argument |
| 1210 | RETURN VALUES |
| 1211 | 0 Equal |
| 1212 | -1 First argument is smaller |
| 1213 | +1 First argument is larger |
| 1214 | */ |
| 1215 | |
| 1216 | extern "C" |
| 1217 | int partition_info_compare_column_values(const void *first_arg, |
| 1218 | const void *second_arg) |
| 1219 | { |
| 1220 | const part_column_list_val *first= (part_column_list_val*)first_arg; |
| 1221 | const part_column_list_val *second= (part_column_list_val*)second_arg; |
| 1222 | partition_info *part_info= first->part_info; |
| 1223 | Field **field; |
| 1224 | |
| 1225 | for (field= part_info->part_field_array; *field; |
| 1226 | field++, first++, second++) |
| 1227 | { |
| 1228 | if (first->max_value || second->max_value) |
| 1229 | { |
| 1230 | if (first->max_value && second->max_value) |
| 1231 | return 0; |
| 1232 | if (second->max_value) |
| 1233 | return -1; |
| 1234 | else |
| 1235 | return +1; |
| 1236 | } |
| 1237 | if (first->null_value || second->null_value) |
| 1238 | { |
| 1239 | if (first->null_value && second->null_value) |
| 1240 | continue; |
| 1241 | if (second->null_value) |
| 1242 | return +1; |
| 1243 | else |
| 1244 | return -1; |
| 1245 | } |
| 1246 | int res= (*field)->cmp((const uchar*)first->column_value, |
| 1247 | (const uchar*)second->column_value); |
| 1248 | if (res) |
| 1249 | return res; |
| 1250 | } |
| 1251 | return 0; |
| 1252 | } |
| 1253 | |
| 1254 | |
| 1255 | /* |
| 1256 | This routine allocates an array for all range constants to achieve a fast |
| 1257 | check what partition a certain value belongs to. At the same time it does |
| 1258 | also check that the range constants are defined in increasing order and |
| 1259 | that the expressions are constant integer expressions. |
| 1260 | |
| 1261 | SYNOPSIS |
| 1262 | check_range_constants() |
| 1263 | thd Thread object |
| 1264 | |
| 1265 | RETURN VALUE |
| 1266 | TRUE An error occurred during creation of range constants |
| 1267 | FALSE Successful creation of range constant mapping |
| 1268 | |
| 1269 | DESCRIPTION |
| 1270 | This routine is called from check_partition_info to get a quick error |
| 1271 | before we came too far into the CREATE TABLE process. It is also called |
| 1272 | from fix_partition_func every time we open the .frm file. It is only |
| 1273 | called for RANGE PARTITIONed tables. |
| 1274 | */ |
| 1275 | |
| 1276 | static bool check_range_constants(THD *thd, partition_info *part_info) |
| 1277 | { |
| 1278 | partition_element* part_def; |
| 1279 | bool first= TRUE; |
| 1280 | uint i; |
| 1281 | List_iterator<partition_element> it(part_info->partitions); |
| 1282 | bool result= TRUE; |
| 1283 | DBUG_ENTER("check_range_constants" ); |
| 1284 | DBUG_PRINT("enter" , ("RANGE with %d parts, column_list = %u" , |
| 1285 | part_info->num_parts, part_info->column_list)); |
| 1286 | |
| 1287 | if (part_info->column_list) |
| 1288 | { |
| 1289 | part_column_list_val *loc_range_col_array; |
| 1290 | part_column_list_val *UNINIT_VAR(current_largest_col_val); |
| 1291 | uint num_column_values= part_info->part_field_list.elements; |
| 1292 | uint size_entries= sizeof(part_column_list_val) * num_column_values; |
| 1293 | part_info->range_col_array= (part_column_list_val*) |
| 1294 | thd->calloc(part_info->num_parts * size_entries); |
| 1295 | if (unlikely(part_info->range_col_array == NULL)) |
| 1296 | goto end; |
| 1297 | |
| 1298 | loc_range_col_array= part_info->range_col_array; |
| 1299 | i= 0; |
| 1300 | do |
| 1301 | { |
| 1302 | part_def= it++; |
| 1303 | { |
| 1304 | List_iterator<part_elem_value> list_val_it(part_def->list_val_list); |
| 1305 | part_elem_value *range_val= list_val_it++; |
| 1306 | part_column_list_val *col_val= range_val->col_val_array; |
| 1307 | |
| 1308 | if (part_info->fix_column_value_functions(thd, range_val, i)) |
| 1309 | goto end; |
| 1310 | memcpy(loc_range_col_array, (const void*)col_val, size_entries); |
| 1311 | loc_range_col_array+= num_column_values; |
| 1312 | if (!first) |
| 1313 | { |
| 1314 | if (partition_info_compare_column_values(current_largest_col_val, |
| 1315 | col_val) >= 0) |
| 1316 | goto range_not_increasing_error; |
| 1317 | } |
| 1318 | current_largest_col_val= col_val; |
| 1319 | } |
| 1320 | first= FALSE; |
| 1321 | } while (++i < part_info->num_parts); |
| 1322 | } |
| 1323 | else |
| 1324 | { |
| 1325 | longlong UNINIT_VAR(current_largest); |
| 1326 | longlong part_range_value; |
| 1327 | bool signed_flag= !part_info->part_expr->unsigned_flag; |
| 1328 | |
| 1329 | part_info->range_int_array= (longlong*) |
| 1330 | thd->alloc(part_info->num_parts * sizeof(longlong)); |
| 1331 | if (unlikely(part_info->range_int_array == NULL)) |
| 1332 | goto end; |
| 1333 | |
| 1334 | i= 0; |
| 1335 | do |
| 1336 | { |
| 1337 | part_def= it++; |
| 1338 | if ((i != part_info->num_parts - 1) || !part_info->defined_max_value) |
| 1339 | { |
| 1340 | part_range_value= part_def->range_value; |
| 1341 | if (!signed_flag) |
| 1342 | part_range_value-= 0x8000000000000000ULL; |
| 1343 | } |
| 1344 | else |
| 1345 | part_range_value= LONGLONG_MAX; |
| 1346 | |
| 1347 | if (!first) |
| 1348 | { |
| 1349 | if (current_largest > part_range_value || |
| 1350 | (current_largest == part_range_value && |
| 1351 | (part_range_value < LONGLONG_MAX || |
| 1352 | i != part_info->num_parts - 1 || |
| 1353 | !part_info->defined_max_value))) |
| 1354 | goto range_not_increasing_error; |
| 1355 | } |
| 1356 | part_info->range_int_array[i]= part_range_value; |
| 1357 | current_largest= part_range_value; |
| 1358 | first= FALSE; |
| 1359 | } while (++i < part_info->num_parts); |
| 1360 | } |
| 1361 | result= FALSE; |
| 1362 | end: |
| 1363 | DBUG_RETURN(result); |
| 1364 | |
| 1365 | range_not_increasing_error: |
| 1366 | my_error(ER_RANGE_NOT_INCREASING_ERROR, MYF(0)); |
| 1367 | goto end; |
| 1368 | } |
| 1369 | |
| 1370 | |
| 1371 | /* |
| 1372 | This routine allocates an array for all list constants to achieve a fast |
| 1373 | check what partition a certain value belongs to. At the same time it does |
| 1374 | also check that there are no duplicates among the list constants and that |
| 1375 | that the list expressions are constant integer expressions. |
| 1376 | |
| 1377 | SYNOPSIS |
| 1378 | check_list_constants() |
| 1379 | thd Thread object |
| 1380 | |
| 1381 | RETURN VALUE |
| 1382 | TRUE An error occurred during creation of list constants |
| 1383 | FALSE Successful creation of list constant mapping |
| 1384 | |
| 1385 | DESCRIPTION |
| 1386 | This routine is called from check_partition_info to get a quick error |
| 1387 | before we came too far into the CREATE TABLE process. It is also called |
| 1388 | from fix_partition_func every time we open the .frm file. It is only |
| 1389 | called for LIST PARTITIONed tables. |
| 1390 | */ |
| 1391 | |
| 1392 | static bool check_list_constants(THD *thd, partition_info *part_info) |
| 1393 | { |
| 1394 | uint i, size_entries, num_column_values; |
| 1395 | uint list_index= 0; |
| 1396 | part_elem_value *list_value; |
| 1397 | bool result= TRUE; |
| 1398 | longlong type_add, calc_value; |
| 1399 | void *curr_value; |
| 1400 | void *UNINIT_VAR(prev_value); |
| 1401 | partition_element* part_def; |
| 1402 | bool found_null= FALSE; |
| 1403 | qsort_cmp compare_func; |
| 1404 | void *ptr; |
| 1405 | List_iterator<partition_element> list_func_it(part_info->partitions); |
| 1406 | DBUG_ENTER("check_list_constants" ); |
| 1407 | |
| 1408 | DBUG_ASSERT(part_info->part_type == LIST_PARTITION); |
| 1409 | |
| 1410 | part_info->num_list_values= 0; |
| 1411 | /* |
| 1412 | We begin by calculating the number of list values that have been |
| 1413 | defined in the first step. |
| 1414 | |
| 1415 | We use this number to allocate a properly sized array of structs |
| 1416 | to keep the partition id and the value to use in that partition. |
| 1417 | In the second traversal we assign them values in the struct array. |
| 1418 | |
| 1419 | Finally we sort the array of structs in order of values to enable |
| 1420 | a quick binary search for the proper value to discover the |
| 1421 | partition id. |
| 1422 | After sorting the array we check that there are no duplicates in the |
| 1423 | list. |
| 1424 | */ |
| 1425 | |
| 1426 | i= 0; |
| 1427 | do |
| 1428 | { |
| 1429 | part_def= list_func_it++; |
| 1430 | if (part_def->has_null_value) |
| 1431 | { |
| 1432 | if (found_null) |
| 1433 | { |
| 1434 | my_error(ER_MULTIPLE_DEF_CONST_IN_LIST_PART_ERROR, MYF(0)); |
| 1435 | goto end; |
| 1436 | } |
| 1437 | part_info->has_null_value= TRUE; |
| 1438 | part_info->has_null_part_id= i; |
| 1439 | found_null= TRUE; |
| 1440 | } |
| 1441 | part_info->num_list_values+= part_def->list_val_list.elements; |
| 1442 | } while (++i < part_info->num_parts); |
| 1443 | list_func_it.rewind(); |
| 1444 | num_column_values= part_info->part_field_list.elements; |
| 1445 | size_entries= part_info->column_list ? |
| 1446 | (num_column_values * sizeof(part_column_list_val)) : |
| 1447 | sizeof(LIST_PART_ENTRY); |
| 1448 | if (!(ptr= thd->calloc((part_info->num_list_values+1) * size_entries))) |
| 1449 | goto end; |
| 1450 | if (part_info->column_list) |
| 1451 | { |
| 1452 | part_column_list_val *loc_list_col_array; |
| 1453 | loc_list_col_array= (part_column_list_val*)ptr; |
| 1454 | part_info->list_col_array= (part_column_list_val*)ptr; |
| 1455 | compare_func= partition_info_compare_column_values; |
| 1456 | i= 0; |
| 1457 | do |
| 1458 | { |
| 1459 | part_def= list_func_it++; |
| 1460 | if (part_def->max_value) |
| 1461 | { |
| 1462 | // DEFAULT is not a real value so let's exclude it from sorting. |
| 1463 | DBUG_ASSERT(part_info->num_list_values); |
| 1464 | part_info->num_list_values--; |
| 1465 | continue; |
| 1466 | } |
| 1467 | List_iterator<part_elem_value> list_val_it2(part_def->list_val_list); |
| 1468 | while ((list_value= list_val_it2++)) |
| 1469 | { |
| 1470 | part_column_list_val *col_val= list_value->col_val_array; |
| 1471 | if (part_info->fix_column_value_functions(thd, list_value, i)) |
| 1472 | DBUG_RETURN(result); |
| 1473 | memcpy(loc_list_col_array, (const void*)col_val, size_entries); |
| 1474 | loc_list_col_array+= num_column_values; |
| 1475 | } |
| 1476 | } while (++i < part_info->num_parts); |
| 1477 | } |
| 1478 | else |
| 1479 | { |
| 1480 | compare_func= partition_info_list_part_cmp; |
| 1481 | part_info->list_array= (LIST_PART_ENTRY*)ptr; |
| 1482 | i= 0; |
| 1483 | /* |
| 1484 | Fix to be able to reuse signed sort functions also for unsigned |
| 1485 | partition functions. |
| 1486 | */ |
| 1487 | type_add= (longlong)(part_info->part_expr->unsigned_flag ? |
| 1488 | 0x8000000000000000ULL : |
| 1489 | 0ULL); |
| 1490 | |
| 1491 | do |
| 1492 | { |
| 1493 | part_def= list_func_it++; |
| 1494 | if (part_def->max_value) |
| 1495 | { |
| 1496 | // DEFAULT is not a real value so let's exclude it from sorting. |
| 1497 | DBUG_ASSERT(part_info->num_list_values); |
| 1498 | part_info->num_list_values--; |
| 1499 | continue; |
| 1500 | } |
| 1501 | List_iterator<part_elem_value> list_val_it2(part_def->list_val_list); |
| 1502 | while ((list_value= list_val_it2++)) |
| 1503 | { |
| 1504 | calc_value= list_value->value - type_add; |
| 1505 | part_info->list_array[list_index].list_value= calc_value; |
| 1506 | part_info->list_array[list_index++].partition_id= i; |
| 1507 | } |
| 1508 | } while (++i < part_info->num_parts); |
| 1509 | } |
| 1510 | DBUG_ASSERT(part_info->fixed); |
| 1511 | if (part_info->num_list_values) |
| 1512 | { |
| 1513 | bool first= TRUE; |
| 1514 | /* |
| 1515 | list_array and list_col_array are unions, so this works for both |
| 1516 | variants of LIST partitioning. |
| 1517 | */ |
| 1518 | my_qsort(part_info->list_array, part_info->num_list_values, size_entries, |
| 1519 | compare_func); |
| 1520 | |
| 1521 | i= 0; |
| 1522 | do |
| 1523 | { |
| 1524 | DBUG_ASSERT(i < part_info->num_list_values); |
| 1525 | curr_value= part_info->column_list |
| 1526 | ? (void*)&part_info->list_col_array[num_column_values * i] |
| 1527 | : (void*)&part_info->list_array[i]; |
| 1528 | if (likely(first || compare_func(curr_value, prev_value))) |
| 1529 | { |
| 1530 | prev_value= curr_value; |
| 1531 | first= FALSE; |
| 1532 | } |
| 1533 | else |
| 1534 | { |
| 1535 | my_error(ER_MULTIPLE_DEF_CONST_IN_LIST_PART_ERROR, MYF(0)); |
| 1536 | goto end; |
| 1537 | } |
| 1538 | } while (++i < part_info->num_list_values); |
| 1539 | } |
| 1540 | result= FALSE; |
| 1541 | end: |
| 1542 | DBUG_RETURN(result); |
| 1543 | } |
| 1544 | |
| 1545 | |
| 1546 | /* Set partition boundaries when rotating by INTERVAL */ |
| 1547 | static bool check_vers_constants(THD *thd, partition_info *part_info) |
| 1548 | { |
| 1549 | uint hist_parts= part_info->num_parts - 1; |
| 1550 | Vers_part_info *vers_info= part_info->vers_info; |
| 1551 | vers_info->hist_part= part_info->partitions.head(); |
| 1552 | vers_info->now_part= part_info->partitions.elem(hist_parts); |
| 1553 | |
| 1554 | if (!vers_info->interval.is_set()) |
| 1555 | return 0; |
| 1556 | |
| 1557 | part_info->range_int_array= |
| 1558 | (longlong*) thd->alloc(hist_parts * sizeof(longlong)); |
| 1559 | |
| 1560 | MYSQL_TIME ltime; |
| 1561 | List_iterator<partition_element> it(part_info->partitions); |
| 1562 | partition_element *el; |
| 1563 | my_tz_OFFSET0->gmt_sec_to_TIME(<ime, vers_info->interval.start); |
| 1564 | while ((el= it++)->id < hist_parts) |
| 1565 | { |
| 1566 | if (date_add_interval(<ime, vers_info->interval.type, |
| 1567 | vers_info->interval.step)) |
| 1568 | goto err; |
| 1569 | uint error= 0; |
| 1570 | part_info->range_int_array[el->id]= el->range_value= |
| 1571 | my_tz_OFFSET0->TIME_to_gmt_sec(<ime, &error); |
| 1572 | if (error) |
| 1573 | goto err; |
| 1574 | if (vers_info->hist_part->range_value <= thd->query_start()) |
| 1575 | vers_info->hist_part= el; |
| 1576 | } |
| 1577 | return 0; |
| 1578 | err: |
| 1579 | my_error(ER_DATA_OUT_OF_RANGE, MYF(0), "TIMESTAMP" , "INTERVAL" ); |
| 1580 | return 1; |
| 1581 | } |
| 1582 | |
| 1583 | |
| 1584 | /* |
| 1585 | Set up function pointers for partition function |
| 1586 | |
| 1587 | SYNOPSIS |
| 1588 | set_up_partition_func_pointers() |
| 1589 | part_info Reference to partitioning data structure |
| 1590 | |
| 1591 | RETURN VALUE |
| 1592 | NONE |
| 1593 | |
| 1594 | DESCRIPTION |
| 1595 | Set-up all function pointers for calculation of partition id, |
| 1596 | subpartition id and the upper part in subpartitioning. This is to speed up |
| 1597 | execution of get_partition_id which is executed once every record to be |
| 1598 | written and deleted and twice for updates. |
| 1599 | */ |
| 1600 | |
| 1601 | static void set_up_partition_func_pointers(partition_info *part_info) |
| 1602 | { |
| 1603 | DBUG_ENTER("set_up_partition_func_pointers" ); |
| 1604 | |
| 1605 | if (part_info->is_sub_partitioned()) |
| 1606 | { |
| 1607 | part_info->get_partition_id= get_partition_id_with_sub; |
| 1608 | if (part_info->part_type == RANGE_PARTITION) |
| 1609 | { |
| 1610 | if (part_info->column_list) |
| 1611 | part_info->get_part_partition_id= get_partition_id_range_col; |
| 1612 | else |
| 1613 | part_info->get_part_partition_id= get_partition_id_range; |
| 1614 | if (part_info->list_of_subpart_fields) |
| 1615 | { |
| 1616 | if (part_info->linear_hash_ind) |
| 1617 | part_info->get_subpartition_id= get_partition_id_linear_key_sub; |
| 1618 | else |
| 1619 | part_info->get_subpartition_id= get_partition_id_key_sub; |
| 1620 | } |
| 1621 | else |
| 1622 | { |
| 1623 | if (part_info->linear_hash_ind) |
| 1624 | part_info->get_subpartition_id= get_partition_id_linear_hash_sub; |
| 1625 | else |
| 1626 | part_info->get_subpartition_id= get_partition_id_hash_sub; |
| 1627 | } |
| 1628 | } |
| 1629 | else if (part_info->part_type == VERSIONING_PARTITION) |
| 1630 | { |
| 1631 | part_info->get_part_partition_id= vers_get_partition_id; |
| 1632 | if (part_info->list_of_subpart_fields) |
| 1633 | { |
| 1634 | if (part_info->linear_hash_ind) |
| 1635 | part_info->get_subpartition_id= get_partition_id_linear_key_sub; |
| 1636 | else |
| 1637 | part_info->get_subpartition_id= get_partition_id_key_sub; |
| 1638 | } |
| 1639 | else |
| 1640 | { |
| 1641 | if (part_info->linear_hash_ind) |
| 1642 | part_info->get_subpartition_id= get_partition_id_linear_hash_sub; |
| 1643 | else |
| 1644 | part_info->get_subpartition_id= get_partition_id_hash_sub; |
| 1645 | } |
| 1646 | } |
| 1647 | else /* LIST Partitioning */ |
| 1648 | { |
| 1649 | if (part_info->column_list) |
| 1650 | part_info->get_part_partition_id= get_partition_id_list_col; |
| 1651 | else |
| 1652 | part_info->get_part_partition_id= get_partition_id_list; |
| 1653 | if (part_info->list_of_subpart_fields) |
| 1654 | { |
| 1655 | if (part_info->linear_hash_ind) |
| 1656 | part_info->get_subpartition_id= get_partition_id_linear_key_sub; |
| 1657 | else |
| 1658 | part_info->get_subpartition_id= get_partition_id_key_sub; |
| 1659 | } |
| 1660 | else |
| 1661 | { |
| 1662 | if (part_info->linear_hash_ind) |
| 1663 | part_info->get_subpartition_id= get_partition_id_linear_hash_sub; |
| 1664 | else |
| 1665 | part_info->get_subpartition_id= get_partition_id_hash_sub; |
| 1666 | } |
| 1667 | } |
| 1668 | } |
| 1669 | else /* No subpartitioning */ |
| 1670 | { |
| 1671 | part_info->get_part_partition_id= NULL; |
| 1672 | part_info->get_subpartition_id= NULL; |
| 1673 | if (part_info->part_type == RANGE_PARTITION) |
| 1674 | { |
| 1675 | if (part_info->column_list) |
| 1676 | part_info->get_partition_id= get_partition_id_range_col; |
| 1677 | else |
| 1678 | part_info->get_partition_id= get_partition_id_range; |
| 1679 | } |
| 1680 | else if (part_info->part_type == LIST_PARTITION) |
| 1681 | { |
| 1682 | if (part_info->column_list) |
| 1683 | part_info->get_partition_id= get_partition_id_list_col; |
| 1684 | else |
| 1685 | part_info->get_partition_id= get_partition_id_list; |
| 1686 | } |
| 1687 | else if (part_info->part_type == VERSIONING_PARTITION) |
| 1688 | { |
| 1689 | part_info->get_partition_id= vers_get_partition_id; |
| 1690 | } |
| 1691 | else /* HASH partitioning */ |
| 1692 | { |
| 1693 | if (part_info->list_of_part_fields) |
| 1694 | { |
| 1695 | if (part_info->linear_hash_ind) |
| 1696 | part_info->get_partition_id= get_partition_id_linear_key_nosub; |
| 1697 | else |
| 1698 | part_info->get_partition_id= get_partition_id_key_nosub; |
| 1699 | } |
| 1700 | else |
| 1701 | { |
| 1702 | if (part_info->linear_hash_ind) |
| 1703 | part_info->get_partition_id= get_partition_id_linear_hash_nosub; |
| 1704 | else |
| 1705 | part_info->get_partition_id= get_partition_id_hash_nosub; |
| 1706 | } |
| 1707 | } |
| 1708 | } |
| 1709 | /* |
| 1710 | We need special functions to handle character sets since they require copy |
| 1711 | of field pointers and restore afterwards. For subpartitioned tables we do |
| 1712 | the copy and restore individually on the part and subpart parts. For non- |
| 1713 | subpartitioned tables we use the same functions as used for the parts part |
| 1714 | of subpartioning. |
| 1715 | Thus for subpartitioned tables the get_partition_id is always |
| 1716 | get_partition_id_with_sub, even when character sets exists. |
| 1717 | */ |
| 1718 | if (part_info->part_charset_field_array) |
| 1719 | { |
| 1720 | if (part_info->is_sub_partitioned()) |
| 1721 | { |
| 1722 | DBUG_ASSERT(part_info->get_part_partition_id); |
| 1723 | if (!part_info->column_list) |
| 1724 | { |
| 1725 | part_info->get_part_partition_id_charset= |
| 1726 | part_info->get_part_partition_id; |
| 1727 | part_info->get_part_partition_id= get_part_id_charset_func_part; |
| 1728 | } |
| 1729 | } |
| 1730 | else |
| 1731 | { |
| 1732 | DBUG_ASSERT(part_info->get_partition_id); |
| 1733 | if (!part_info->column_list) |
| 1734 | { |
| 1735 | part_info->get_part_partition_id_charset= part_info->get_partition_id; |
| 1736 | part_info->get_part_partition_id= get_part_id_charset_func_part; |
| 1737 | } |
| 1738 | } |
| 1739 | } |
| 1740 | if (part_info->subpart_charset_field_array) |
| 1741 | { |
| 1742 | DBUG_ASSERT(part_info->get_subpartition_id); |
| 1743 | part_info->get_subpartition_id_charset= |
| 1744 | part_info->get_subpartition_id; |
| 1745 | part_info->get_subpartition_id= get_part_id_charset_func_subpart; |
| 1746 | } |
| 1747 | if (part_info->part_type == RANGE_PARTITION) |
| 1748 | part_info->check_constants= check_range_constants; |
| 1749 | else if (part_info->part_type == LIST_PARTITION) |
| 1750 | part_info->check_constants= check_list_constants; |
| 1751 | else if (part_info->part_type == VERSIONING_PARTITION) |
| 1752 | part_info->check_constants= check_vers_constants; |
| 1753 | else |
| 1754 | part_info->check_constants= check_no_constants; |
| 1755 | DBUG_VOID_RETURN; |
| 1756 | } |
| 1757 | |
| 1758 | |
| 1759 | /* |
| 1760 | For linear hashing we need a mask which is on the form 2**n - 1 where |
| 1761 | 2**n >= num_parts. Thus if num_parts is 6 then mask is 2**3 - 1 = 8 - 1 = 7. |
| 1762 | |
| 1763 | SYNOPSIS |
| 1764 | set_linear_hash_mask() |
| 1765 | part_info Reference to partitioning data structure |
| 1766 | num_parts Number of parts in linear hash partitioning |
| 1767 | |
| 1768 | RETURN VALUE |
| 1769 | NONE |
| 1770 | */ |
| 1771 | |
| 1772 | void set_linear_hash_mask(partition_info *part_info, uint num_parts) |
| 1773 | { |
| 1774 | uint mask; |
| 1775 | |
| 1776 | for (mask= 1; mask < num_parts; mask<<=1) |
| 1777 | ; |
| 1778 | part_info->linear_hash_mask= mask - 1; |
| 1779 | } |
| 1780 | |
| 1781 | |
| 1782 | /* |
| 1783 | This function calculates the partition id provided the result of the hash |
| 1784 | function using linear hashing parameters, mask and number of partitions. |
| 1785 | |
| 1786 | SYNOPSIS |
| 1787 | get_part_id_from_linear_hash() |
| 1788 | hash_value Hash value calculated by HASH function or KEY function |
| 1789 | mask Mask calculated previously by set_linear_hash_mask |
| 1790 | num_parts Number of partitions in HASH partitioned part |
| 1791 | |
| 1792 | RETURN VALUE |
| 1793 | part_id The calculated partition identity (starting at 0) |
| 1794 | |
| 1795 | DESCRIPTION |
| 1796 | The partition is calculated according to the theory of linear hashing. |
| 1797 | See e.g. Linear hashing: a new tool for file and table addressing, |
| 1798 | Reprinted from VLDB-80 in Readings Database Systems, 2nd ed, M. Stonebraker |
| 1799 | (ed.), Morgan Kaufmann 1994. |
| 1800 | */ |
| 1801 | |
| 1802 | static uint32 get_part_id_from_linear_hash(longlong hash_value, uint mask, |
| 1803 | uint num_parts) |
| 1804 | { |
| 1805 | uint32 part_id= (uint32)(hash_value & mask); |
| 1806 | |
| 1807 | if (part_id >= num_parts) |
| 1808 | { |
| 1809 | uint new_mask= ((mask + 1) >> 1) - 1; |
| 1810 | part_id= (uint32)(hash_value & new_mask); |
| 1811 | } |
| 1812 | return part_id; |
| 1813 | } |
| 1814 | |
| 1815 | |
| 1816 | /* |
| 1817 | Check if a particular field is in need of character set |
| 1818 | handling for partition functions. |
| 1819 | |
| 1820 | SYNOPSIS |
| 1821 | field_is_partition_charset() |
| 1822 | field The field to check |
| 1823 | |
| 1824 | RETURN VALUES |
| 1825 | FALSE Not in need of character set handling |
| 1826 | TRUE In need of character set handling |
| 1827 | */ |
| 1828 | |
| 1829 | bool field_is_partition_charset(Field *field) |
| 1830 | { |
| 1831 | if (!(field->type() == MYSQL_TYPE_STRING) && |
| 1832 | !(field->type() == MYSQL_TYPE_VARCHAR)) |
| 1833 | return FALSE; |
| 1834 | { |
| 1835 | CHARSET_INFO *cs= field->charset(); |
| 1836 | if (!(field->type() == MYSQL_TYPE_STRING) || |
| 1837 | !(cs->state & MY_CS_BINSORT)) |
| 1838 | return TRUE; |
| 1839 | return FALSE; |
| 1840 | } |
| 1841 | } |
| 1842 | |
| 1843 | |
| 1844 | /* |
| 1845 | Check that partition function doesn't contain any forbidden |
| 1846 | character sets and collations. |
| 1847 | |
| 1848 | SYNOPSIS |
| 1849 | check_part_func_fields() |
| 1850 | ptr Array of Field pointers |
| 1851 | ok_with_charsets Will we report allowed charset |
| 1852 | fields as ok |
| 1853 | RETURN VALUES |
| 1854 | FALSE Success |
| 1855 | TRUE Error |
| 1856 | |
| 1857 | DESCRIPTION |
| 1858 | We will check in this routine that the fields of the partition functions |
| 1859 | do not contain unallowed parts. It can also be used to check if there |
| 1860 | are fields that require special care by calling my_strnxfrm before |
| 1861 | calling the functions to calculate partition id. |
| 1862 | */ |
| 1863 | |
| 1864 | bool check_part_func_fields(Field **ptr, bool ok_with_charsets) |
| 1865 | { |
| 1866 | Field *field; |
| 1867 | DBUG_ENTER("check_part_func_fields" ); |
| 1868 | |
| 1869 | while ((field= *(ptr++))) |
| 1870 | { |
| 1871 | /* |
| 1872 | For CHAR/VARCHAR fields we need to take special precautions. |
| 1873 | Binary collation with CHAR is automatically supported. Other |
| 1874 | types need some kind of standardisation function handling |
| 1875 | */ |
| 1876 | if (field_is_partition_charset(field)) |
| 1877 | { |
| 1878 | CHARSET_INFO *cs= field->charset(); |
| 1879 | if (!ok_with_charsets || |
| 1880 | cs->mbmaxlen > 1 || |
| 1881 | cs->strxfrm_multiply > 1) |
| 1882 | { |
| 1883 | DBUG_RETURN(TRUE); |
| 1884 | } |
| 1885 | } |
| 1886 | } |
| 1887 | DBUG_RETURN(FALSE); |
| 1888 | } |
| 1889 | |
| 1890 | |
| 1891 | /* |
| 1892 | fix partition functions |
| 1893 | |
| 1894 | SYNOPSIS |
| 1895 | fix_partition_func() |
| 1896 | thd The thread object |
| 1897 | table TABLE object for which partition fields are set-up |
| 1898 | is_create_table_ind Indicator of whether openfrm was called as part of |
| 1899 | CREATE or ALTER TABLE |
| 1900 | |
| 1901 | RETURN VALUE |
| 1902 | TRUE Error |
| 1903 | FALSE Success |
| 1904 | |
| 1905 | DESCRIPTION |
| 1906 | The name parameter contains the full table name and is used to get the |
| 1907 | database name of the table which is used to set-up a correct |
| 1908 | TABLE_LIST object for use in fix_fields. |
| 1909 | |
| 1910 | NOTES |
| 1911 | This function is called as part of opening the table by opening the .frm |
| 1912 | file. It is a part of CREATE TABLE to do this so it is quite permissible |
| 1913 | that errors due to erroneus syntax isn't found until we come here. |
| 1914 | If the user has used a non-existing field in the table is one such example |
| 1915 | of an error that is not discovered until here. |
| 1916 | */ |
| 1917 | |
| 1918 | bool fix_partition_func(THD *thd, TABLE *table, bool is_create_table_ind) |
| 1919 | { |
| 1920 | bool result= TRUE; |
| 1921 | partition_info *part_info= table->part_info; |
| 1922 | enum_column_usage saved_column_usage= thd->column_usage; |
| 1923 | DBUG_ENTER("fix_partition_func" ); |
| 1924 | |
| 1925 | if (part_info->fixed) |
| 1926 | { |
| 1927 | DBUG_RETURN(FALSE); |
| 1928 | } |
| 1929 | thd->column_usage= COLUMNS_WRITE; |
| 1930 | DBUG_PRINT("info" , ("thd->column_usage: %d" , thd->column_usage)); |
| 1931 | |
| 1932 | if (!is_create_table_ind || |
| 1933 | thd->lex->sql_command != SQLCOM_CREATE_TABLE) |
| 1934 | { |
| 1935 | if (partition_default_handling(thd, table, part_info, |
| 1936 | is_create_table_ind, |
| 1937 | table->s->normalized_path.str)) |
| 1938 | { |
| 1939 | DBUG_RETURN(TRUE); |
| 1940 | } |
| 1941 | } |
| 1942 | if (part_info->is_sub_partitioned()) |
| 1943 | { |
| 1944 | DBUG_ASSERT(part_info->subpart_type == HASH_PARTITION); |
| 1945 | /* |
| 1946 | Subpartition is defined. We need to verify that subpartitioning |
| 1947 | function is correct. |
| 1948 | */ |
| 1949 | if (part_info->linear_hash_ind) |
| 1950 | set_linear_hash_mask(part_info, part_info->num_subparts); |
| 1951 | if (part_info->list_of_subpart_fields) |
| 1952 | { |
| 1953 | List_iterator<const char> it(part_info->subpart_field_list); |
| 1954 | if (unlikely(handle_list_of_fields(thd, it, table, part_info, TRUE))) |
| 1955 | goto end; |
| 1956 | } |
| 1957 | else |
| 1958 | { |
| 1959 | if (unlikely(fix_fields_part_func(thd, part_info->subpart_expr, |
| 1960 | table, TRUE, is_create_table_ind))) |
| 1961 | goto end; |
| 1962 | if (unlikely(part_info->subpart_expr->result_type() != INT_RESULT)) |
| 1963 | { |
| 1964 | part_info->report_part_expr_error(TRUE); |
| 1965 | goto end; |
| 1966 | } |
| 1967 | } |
| 1968 | } |
| 1969 | DBUG_ASSERT(part_info->part_type != NOT_A_PARTITION); |
| 1970 | DBUG_ASSERT(part_info->part_type != VERSIONING_PARTITION || part_info->column_list); |
| 1971 | /* |
| 1972 | Partition is defined. We need to verify that partitioning |
| 1973 | function is correct. |
| 1974 | */ |
| 1975 | set_up_partition_func_pointers(part_info); |
| 1976 | if (part_info->part_type == HASH_PARTITION) |
| 1977 | { |
| 1978 | if (part_info->linear_hash_ind) |
| 1979 | set_linear_hash_mask(part_info, part_info->num_parts); |
| 1980 | if (part_info->list_of_part_fields) |
| 1981 | { |
| 1982 | List_iterator<const char> it(part_info->part_field_list); |
| 1983 | if (unlikely(handle_list_of_fields(thd, it, table, part_info, FALSE))) |
| 1984 | goto end; |
| 1985 | } |
| 1986 | else |
| 1987 | { |
| 1988 | if (unlikely(fix_fields_part_func(thd, part_info->part_expr, |
| 1989 | table, FALSE, is_create_table_ind))) |
| 1990 | goto end; |
| 1991 | if (unlikely(part_info->part_expr->result_type() != INT_RESULT)) |
| 1992 | { |
| 1993 | part_info->report_part_expr_error(FALSE); |
| 1994 | goto end; |
| 1995 | } |
| 1996 | } |
| 1997 | part_info->fixed= TRUE; |
| 1998 | } |
| 1999 | else |
| 2000 | { |
| 2001 | if (part_info->column_list) |
| 2002 | { |
| 2003 | if (part_info->part_type == VERSIONING_PARTITION && |
| 2004 | part_info->vers_setup_expression(thd)) |
| 2005 | goto end; |
| 2006 | List_iterator<const char> it(part_info->part_field_list); |
| 2007 | if (unlikely(handle_list_of_fields(thd, it, table, part_info, FALSE))) |
| 2008 | goto end; |
| 2009 | } |
| 2010 | else |
| 2011 | { |
| 2012 | if (unlikely(fix_fields_part_func(thd, part_info->part_expr, |
| 2013 | table, FALSE, is_create_table_ind))) |
| 2014 | goto end; |
| 2015 | } |
| 2016 | part_info->fixed= TRUE; |
| 2017 | if (part_info->check_constants(thd, part_info)) |
| 2018 | goto end; |
| 2019 | if (unlikely(part_info->num_parts < 1)) |
| 2020 | { |
| 2021 | const char *error_str= part_info->part_type == LIST_PARTITION |
| 2022 | ? "LIST" : "RANGE" ; |
| 2023 | my_error(ER_PARTITIONS_MUST_BE_DEFINED_ERROR, MYF(0), error_str); |
| 2024 | goto end; |
| 2025 | } |
| 2026 | if (unlikely(!part_info->column_list && |
| 2027 | part_info->part_expr->result_type() != INT_RESULT)) |
| 2028 | { |
| 2029 | part_info->report_part_expr_error(FALSE); |
| 2030 | goto end; |
| 2031 | } |
| 2032 | } |
| 2033 | if (((part_info->part_type != HASH_PARTITION || |
| 2034 | part_info->list_of_part_fields == FALSE) && |
| 2035 | !part_info->column_list && |
| 2036 | check_part_func_fields(part_info->part_field_array, TRUE)) || |
| 2037 | (part_info->list_of_subpart_fields == FALSE && |
| 2038 | part_info->is_sub_partitioned() && |
| 2039 | check_part_func_fields(part_info->subpart_field_array, TRUE))) |
| 2040 | { |
| 2041 | /* |
| 2042 | Range/List/HASH (but not KEY) and not COLUMNS or HASH subpartitioning |
| 2043 | with columns in the partitioning expression using unallowed charset. |
| 2044 | */ |
| 2045 | my_error(ER_PARTITION_FUNCTION_IS_NOT_ALLOWED, MYF(0)); |
| 2046 | goto end; |
| 2047 | } |
| 2048 | if (unlikely(create_full_part_field_array(thd, table, part_info))) |
| 2049 | goto end; |
| 2050 | if (unlikely(check_primary_key(table))) |
| 2051 | goto end; |
| 2052 | if (unlikely((!(table->s->db_type()->partition_flags && |
| 2053 | (table->s->db_type()->partition_flags() & HA_CAN_PARTITION_UNIQUE))) && |
| 2054 | check_unique_keys(table))) |
| 2055 | goto end; |
| 2056 | if (unlikely(set_up_partition_bitmaps(thd, part_info))) |
| 2057 | goto end; |
| 2058 | if (unlikely(part_info->set_up_charset_field_preps(thd))) |
| 2059 | { |
| 2060 | my_error(ER_PARTITION_FUNCTION_IS_NOT_ALLOWED, MYF(0)); |
| 2061 | goto end; |
| 2062 | } |
| 2063 | if (unlikely(part_info->check_partition_field_length())) |
| 2064 | { |
| 2065 | my_error(ER_PARTITION_FIELDS_TOO_LONG, MYF(0)); |
| 2066 | goto end; |
| 2067 | } |
| 2068 | check_range_capable_PF(table); |
| 2069 | set_up_partition_key_maps(table, part_info); |
| 2070 | set_up_range_analysis_info(part_info); |
| 2071 | table->file->set_part_info(part_info); |
| 2072 | result= FALSE; |
| 2073 | end: |
| 2074 | thd->column_usage= saved_column_usage; |
| 2075 | DBUG_PRINT("info" , ("thd->column_usage: %d" , thd->column_usage)); |
| 2076 | DBUG_RETURN(result); |
| 2077 | } |
| 2078 | |
| 2079 | |
| 2080 | /* |
| 2081 | The code below is support routines for the reverse parsing of the |
| 2082 | partitioning syntax. This feature is very useful to generate syntax for |
| 2083 | all default values to avoid all default checking when opening the frm |
| 2084 | file. It is also used when altering the partitioning by use of various |
| 2085 | ALTER TABLE commands. Finally it is used for SHOW CREATE TABLES. |
| 2086 | */ |
| 2087 | |
| 2088 | static int add_part_field_list(THD *thd, String *str, List<const char> field_list) |
| 2089 | { |
| 2090 | int err= 0; |
| 2091 | const char *field_name; |
| 2092 | List_iterator<const char> part_it(field_list); |
| 2093 | |
| 2094 | err+= str->append('('); |
| 2095 | while ((field_name= part_it++)) |
| 2096 | { |
| 2097 | err+= append_identifier(thd, str, field_name, strlen(field_name)); |
| 2098 | err+= str->append(','); |
| 2099 | } |
| 2100 | if (field_list.elements) |
| 2101 | str->length(str->length()-1); |
| 2102 | err+= str->append(')'); |
| 2103 | return err; |
| 2104 | } |
| 2105 | |
| 2106 | /* |
| 2107 | Must escape strings in partitioned tables frm-files, |
| 2108 | parsing it later with mysql_unpack_partition will fail otherwise. |
| 2109 | */ |
| 2110 | |
| 2111 | static int add_keyword_string(String *str, const char *keyword, |
| 2112 | bool quoted, const char *keystr) |
| 2113 | { |
| 2114 | int err= str->append(' '); |
| 2115 | err+= str->append(keyword); |
| 2116 | |
| 2117 | str->append(STRING_WITH_LEN(" = " )); |
| 2118 | if (quoted) |
| 2119 | { |
| 2120 | err+= str->append('\''); |
| 2121 | err+= str->append_for_single_quote(keystr); |
| 2122 | err+= str->append('\''); |
| 2123 | } |
| 2124 | else |
| 2125 | err+= str->append(keystr); |
| 2126 | return err; |
| 2127 | } |
| 2128 | |
| 2129 | |
| 2130 | /** |
| 2131 | @brief Truncate the partition file name from a path it it exists. |
| 2132 | |
| 2133 | @note A partition file name will contian one or more '#' characters. |
| 2134 | One of the occurances of '#' will be either "#P#" or "#p#" depending |
| 2135 | on whether the storage engine has converted the filename to lower case. |
| 2136 | */ |
| 2137 | void truncate_partition_filename(char *path) |
| 2138 | { |
| 2139 | if (path) |
| 2140 | { |
| 2141 | char* last_slash= strrchr(path, FN_LIBCHAR); |
| 2142 | |
| 2143 | if (!last_slash) |
| 2144 | last_slash= strrchr(path, FN_LIBCHAR2); |
| 2145 | |
| 2146 | if (last_slash) |
| 2147 | { |
| 2148 | /* Look for a partition-type filename */ |
| 2149 | for (char* pound= strchr(last_slash, '#'); |
| 2150 | pound; pound = strchr(pound + 1, '#')) |
| 2151 | { |
| 2152 | if ((pound[1] == 'P' || pound[1] == 'p') && pound[2] == '#') |
| 2153 | { |
| 2154 | last_slash[0] = '\0'; /* truncate the file name */ |
| 2155 | break; |
| 2156 | } |
| 2157 | } |
| 2158 | } |
| 2159 | } |
| 2160 | } |
| 2161 | |
| 2162 | /** |
| 2163 | @brief Output a filepath. Similar to add_keyword_string except it |
| 2164 | also converts \ to / on Windows and skips the partition file name at |
| 2165 | the end if found. |
| 2166 | |
| 2167 | @note When Mysql sends a DATA DIRECTORY from SQL for partitions it does |
| 2168 | not use a file name, but it does for DATA DIRECTORY on a non-partitioned |
| 2169 | table. So when the storage engine is asked for the DATA DIRECTORY string |
| 2170 | after a restart through Handler::update_create_options(), the storage |
| 2171 | engine may include the filename. |
| 2172 | */ |
| 2173 | static int add_keyword_path(String *str, const char *keyword, |
| 2174 | const char *path) |
| 2175 | { |
| 2176 | char temp_path[FN_REFLEN]; |
| 2177 | strcpy(temp_path, path); |
| 2178 | #ifdef __WIN__ |
| 2179 | /* Convert \ to / to be able to create table on unix */ |
| 2180 | char *pos, *end; |
| 2181 | size_t length= strlen(temp_path); |
| 2182 | for (pos= temp_path, end= pos+length ; pos < end ; pos++) |
| 2183 | { |
| 2184 | if (*pos == '\\') |
| 2185 | *pos = '/'; |
| 2186 | } |
| 2187 | #endif |
| 2188 | |
| 2189 | /* |
| 2190 | If the partition file name with its "#P#" identifier |
| 2191 | is found after the last slash, truncate that filename. |
| 2192 | */ |
| 2193 | truncate_partition_filename(temp_path); |
| 2194 | |
| 2195 | return add_keyword_string(str, keyword, true, temp_path); |
| 2196 | } |
| 2197 | |
| 2198 | static int add_keyword_int(String *str, const char *keyword, longlong num) |
| 2199 | { |
| 2200 | int err= str->append(' '); |
| 2201 | err+= str->append(keyword); |
| 2202 | str->append(STRING_WITH_LEN(" = " )); |
| 2203 | return err + str->append_longlong(num); |
| 2204 | } |
| 2205 | |
| 2206 | static int add_partition_options(String *str, partition_element *p_elem) |
| 2207 | { |
| 2208 | int err= 0; |
| 2209 | |
| 2210 | if (p_elem->tablespace_name) |
| 2211 | err+= add_keyword_string(str,"TABLESPACE" , false, p_elem->tablespace_name); |
| 2212 | if (p_elem->nodegroup_id != UNDEF_NODEGROUP) |
| 2213 | err+= add_keyword_int(str,"NODEGROUP" ,(longlong)p_elem->nodegroup_id); |
| 2214 | if (p_elem->part_max_rows) |
| 2215 | err+= add_keyword_int(str,"MAX_ROWS" ,(longlong)p_elem->part_max_rows); |
| 2216 | if (p_elem->part_min_rows) |
| 2217 | err+= add_keyword_int(str,"MIN_ROWS" ,(longlong)p_elem->part_min_rows); |
| 2218 | if (!(current_thd->variables.sql_mode & MODE_NO_DIR_IN_CREATE)) |
| 2219 | { |
| 2220 | if (p_elem->data_file_name) |
| 2221 | err+= add_keyword_path(str, "DATA DIRECTORY" , p_elem->data_file_name); |
| 2222 | if (p_elem->index_file_name) |
| 2223 | err+= add_keyword_path(str, "INDEX DIRECTORY" , p_elem->index_file_name); |
| 2224 | } |
| 2225 | if (p_elem->part_comment) |
| 2226 | err+= add_keyword_string(str, "COMMENT" , true, p_elem->part_comment); |
| 2227 | if (p_elem->connect_string.length) |
| 2228 | err+= add_keyword_string(str, "CONNECTION" , true, |
| 2229 | p_elem->connect_string.str); |
| 2230 | err += add_keyword_string(str, "ENGINE" , false, |
| 2231 | ha_resolve_storage_engine_name(p_elem->engine_type)); |
| 2232 | return err; |
| 2233 | } |
| 2234 | |
| 2235 | |
| 2236 | /* |
| 2237 | Check partition fields for result type and if they need |
| 2238 | to check the character set. |
| 2239 | |
| 2240 | SYNOPSIS |
| 2241 | check_part_field() |
| 2242 | sql_type Type provided by user |
| 2243 | field_name Name of field, used for error handling |
| 2244 | result_type Out value: Result type of field |
| 2245 | need_cs_check Out value: Do we need character set check |
| 2246 | |
| 2247 | RETURN VALUES |
| 2248 | TRUE Error |
| 2249 | FALSE Ok |
| 2250 | */ |
| 2251 | |
| 2252 | static int check_part_field(enum_field_types sql_type, |
| 2253 | const char *field_name, |
| 2254 | Item_result *result_type, |
| 2255 | bool *need_cs_check) |
| 2256 | { |
| 2257 | if (sql_type >= MYSQL_TYPE_TINY_BLOB && |
| 2258 | sql_type <= MYSQL_TYPE_BLOB) |
| 2259 | { |
| 2260 | my_error(ER_BLOB_FIELD_IN_PART_FUNC_ERROR, MYF(0)); |
| 2261 | return TRUE; |
| 2262 | } |
| 2263 | switch (sql_type) |
| 2264 | { |
| 2265 | case MYSQL_TYPE_TINY: |
| 2266 | case MYSQL_TYPE_SHORT: |
| 2267 | case MYSQL_TYPE_LONG: |
| 2268 | case MYSQL_TYPE_LONGLONG: |
| 2269 | case MYSQL_TYPE_INT24: |
| 2270 | *result_type= INT_RESULT; |
| 2271 | *need_cs_check= FALSE; |
| 2272 | return FALSE; |
| 2273 | case MYSQL_TYPE_NEWDATE: |
| 2274 | case MYSQL_TYPE_DATE: |
| 2275 | case MYSQL_TYPE_TIME: |
| 2276 | case MYSQL_TYPE_DATETIME: |
| 2277 | case MYSQL_TYPE_TIME2: |
| 2278 | case MYSQL_TYPE_DATETIME2: |
| 2279 | *result_type= STRING_RESULT; |
| 2280 | *need_cs_check= TRUE; |
| 2281 | return FALSE; |
| 2282 | case MYSQL_TYPE_VARCHAR: |
| 2283 | case MYSQL_TYPE_STRING: |
| 2284 | case MYSQL_TYPE_VAR_STRING: |
| 2285 | *result_type= STRING_RESULT; |
| 2286 | *need_cs_check= TRUE; |
| 2287 | return FALSE; |
| 2288 | case MYSQL_TYPE_NEWDECIMAL: |
| 2289 | case MYSQL_TYPE_DECIMAL: |
| 2290 | case MYSQL_TYPE_TIMESTAMP: |
| 2291 | case MYSQL_TYPE_TIMESTAMP2: |
| 2292 | case MYSQL_TYPE_NULL: |
| 2293 | case MYSQL_TYPE_FLOAT: |
| 2294 | case MYSQL_TYPE_DOUBLE: |
| 2295 | case MYSQL_TYPE_BIT: |
| 2296 | case MYSQL_TYPE_ENUM: |
| 2297 | case MYSQL_TYPE_SET: |
| 2298 | case MYSQL_TYPE_GEOMETRY: |
| 2299 | goto error; |
| 2300 | default: |
| 2301 | goto error; |
| 2302 | } |
| 2303 | error: |
| 2304 | my_error(ER_FIELD_TYPE_NOT_ALLOWED_AS_PARTITION_FIELD, MYF(0), |
| 2305 | field_name); |
| 2306 | return TRUE; |
| 2307 | } |
| 2308 | |
| 2309 | |
| 2310 | /* |
| 2311 | Find the given field's Create_field object using name of field |
| 2312 | |
| 2313 | SYNOPSIS |
| 2314 | get_sql_field() |
| 2315 | field_name Field name |
| 2316 | alter_info Info from ALTER TABLE/CREATE TABLE |
| 2317 | |
| 2318 | RETURN VALUE |
| 2319 | sql_field Object filled in by parser about field |
| 2320 | NULL No field found |
| 2321 | */ |
| 2322 | |
| 2323 | static Create_field* get_sql_field(const char *field_name, |
| 2324 | Alter_info *alter_info) |
| 2325 | { |
| 2326 | List_iterator<Create_field> it(alter_info->create_list); |
| 2327 | Create_field *sql_field; |
| 2328 | DBUG_ENTER("get_sql_field" ); |
| 2329 | |
| 2330 | while ((sql_field= it++)) |
| 2331 | { |
| 2332 | if (!(my_strcasecmp(system_charset_info, |
| 2333 | sql_field->field_name.str, |
| 2334 | field_name))) |
| 2335 | { |
| 2336 | DBUG_RETURN(sql_field); |
| 2337 | } |
| 2338 | } |
| 2339 | DBUG_RETURN(NULL); |
| 2340 | } |
| 2341 | |
| 2342 | |
| 2343 | static int add_column_list_values(String *str, partition_info *part_info, |
| 2344 | part_elem_value *list_value, |
| 2345 | HA_CREATE_INFO *create_info, |
| 2346 | Alter_info *alter_info) |
| 2347 | { |
| 2348 | int err= 0; |
| 2349 | uint i; |
| 2350 | List_iterator<const char> it(part_info->part_field_list); |
| 2351 | uint num_elements= part_info->part_field_list.elements; |
| 2352 | bool use_parenthesis= (part_info->part_type == LIST_PARTITION && |
| 2353 | part_info->num_columns > 1U); |
| 2354 | |
| 2355 | if (use_parenthesis) |
| 2356 | err+= str->append('('); |
| 2357 | for (i= 0; i < num_elements; i++) |
| 2358 | { |
| 2359 | part_column_list_val *col_val= &list_value->col_val_array[i]; |
| 2360 | const char *field_name= it++; |
| 2361 | if (col_val->max_value) |
| 2362 | err+= str->append(STRING_WITH_LEN("MAXVALUE" )); |
| 2363 | else if (col_val->null_value) |
| 2364 | err+= str->append(STRING_WITH_LEN("NULL" )); |
| 2365 | else |
| 2366 | { |
| 2367 | Item *item_expr= col_val->item_expression; |
| 2368 | if (item_expr->null_value) |
| 2369 | err+= str->append(STRING_WITH_LEN("NULL" )); |
| 2370 | else |
| 2371 | { |
| 2372 | CHARSET_INFO *field_cs; |
| 2373 | bool need_cs_check= FALSE; |
| 2374 | Item_result result_type= STRING_RESULT; |
| 2375 | |
| 2376 | /* |
| 2377 | This function is called at a very early stage, even before |
| 2378 | we have prepared the sql_field objects. Thus we have to |
| 2379 | find the proper sql_field object and get the character set |
| 2380 | from that object. |
| 2381 | */ |
| 2382 | if (create_info) |
| 2383 | { |
| 2384 | Create_field *sql_field; |
| 2385 | |
| 2386 | if (!(sql_field= get_sql_field(field_name, |
| 2387 | alter_info))) |
| 2388 | { |
| 2389 | my_error(ER_FIELD_NOT_FOUND_PART_ERROR, MYF(0)); |
| 2390 | return 1; |
| 2391 | } |
| 2392 | if (check_part_field(sql_field->real_field_type(), |
| 2393 | sql_field->field_name.str, |
| 2394 | &result_type, |
| 2395 | &need_cs_check)) |
| 2396 | return 1; |
| 2397 | if (need_cs_check) |
| 2398 | field_cs= get_sql_field_charset(sql_field, create_info); |
| 2399 | else |
| 2400 | field_cs= NULL; |
| 2401 | } |
| 2402 | else |
| 2403 | { |
| 2404 | Field *field= part_info->part_field_array[i]; |
| 2405 | result_type= field->result_type(); |
| 2406 | if (check_part_field(field->real_type(), |
| 2407 | field->field_name.str, |
| 2408 | &result_type, |
| 2409 | &need_cs_check)) |
| 2410 | return 1; |
| 2411 | DBUG_ASSERT(result_type == field->result_type()); |
| 2412 | if (need_cs_check) |
| 2413 | field_cs= field->charset(); |
| 2414 | else |
| 2415 | field_cs= NULL; |
| 2416 | } |
| 2417 | if (result_type != item_expr->result_type()) |
| 2418 | { |
| 2419 | my_error(ER_WRONG_TYPE_COLUMN_VALUE_ERROR, MYF(0)); |
| 2420 | return 1; |
| 2421 | } |
| 2422 | if (field_cs && field_cs != item_expr->collation.collation) |
| 2423 | { |
| 2424 | if (!(item_expr= convert_charset_partition_constant(item_expr, |
| 2425 | field_cs))) |
| 2426 | { |
| 2427 | my_error(ER_PARTITION_FUNCTION_IS_NOT_ALLOWED, MYF(0)); |
| 2428 | return 1; |
| 2429 | } |
| 2430 | } |
| 2431 | { |
| 2432 | StringBuffer<MAX_KEY_LENGTH> buf; |
| 2433 | String val_conv, *res; |
| 2434 | val_conv.set_charset(system_charset_info); |
| 2435 | res= item_expr->val_str(&buf); |
| 2436 | if (get_cs_converted_part_value_from_string(current_thd, |
| 2437 | item_expr, res, |
| 2438 | &val_conv, field_cs, |
| 2439 | (bool)(alter_info != NULL))) |
| 2440 | return 1; |
| 2441 | err+= str->append(val_conv); |
| 2442 | } |
| 2443 | } |
| 2444 | } |
| 2445 | if (i != (num_elements - 1)) |
| 2446 | err+= str->append(','); |
| 2447 | } |
| 2448 | if (use_parenthesis) |
| 2449 | err+= str->append(')'); |
| 2450 | return err; |
| 2451 | } |
| 2452 | |
| 2453 | static int add_partition_values(String *str, partition_info *part_info, |
| 2454 | partition_element *p_elem, |
| 2455 | HA_CREATE_INFO *create_info, |
| 2456 | Alter_info *alter_info) |
| 2457 | { |
| 2458 | int err= 0; |
| 2459 | |
| 2460 | if (part_info->part_type == RANGE_PARTITION) |
| 2461 | { |
| 2462 | err+= str->append(STRING_WITH_LEN(" VALUES LESS THAN " )); |
| 2463 | if (part_info->column_list) |
| 2464 | { |
| 2465 | List_iterator<part_elem_value> list_val_it(p_elem->list_val_list); |
| 2466 | part_elem_value *list_value= list_val_it++; |
| 2467 | err+= str->append('('); |
| 2468 | err+= add_column_list_values(str, part_info, list_value, |
| 2469 | create_info, alter_info); |
| 2470 | err+= str->append(')'); |
| 2471 | } |
| 2472 | else |
| 2473 | { |
| 2474 | if (!p_elem->max_value) |
| 2475 | { |
| 2476 | err+= str->append('('); |
| 2477 | if (p_elem->signed_flag) |
| 2478 | err+= str->append_longlong(p_elem->range_value); |
| 2479 | else |
| 2480 | err+= str->append_ulonglong(p_elem->range_value); |
| 2481 | err+= str->append(')'); |
| 2482 | } |
| 2483 | else |
| 2484 | err+= str->append(STRING_WITH_LEN("MAXVALUE" )); |
| 2485 | } |
| 2486 | } |
| 2487 | else if (part_info->part_type == LIST_PARTITION) |
| 2488 | { |
| 2489 | uint i; |
| 2490 | List_iterator<part_elem_value> list_val_it(p_elem->list_val_list); |
| 2491 | |
| 2492 | if (p_elem->max_value) |
| 2493 | { |
| 2494 | DBUG_ASSERT(part_info->defined_max_value || |
| 2495 | current_thd->lex->sql_command == SQLCOM_ALTER_TABLE); |
| 2496 | err+= str->append(STRING_WITH_LEN(" DEFAULT" )); |
| 2497 | return err; |
| 2498 | } |
| 2499 | |
| 2500 | err+= str->append(STRING_WITH_LEN(" VALUES IN " )); |
| 2501 | uint num_items= p_elem->list_val_list.elements; |
| 2502 | |
| 2503 | err+= str->append('('); |
| 2504 | if (p_elem->has_null_value) |
| 2505 | { |
| 2506 | err+= str->append(STRING_WITH_LEN("NULL" )); |
| 2507 | if (num_items == 0) |
| 2508 | { |
| 2509 | err+= str->append(')'); |
| 2510 | goto end; |
| 2511 | } |
| 2512 | err+= str->append(','); |
| 2513 | } |
| 2514 | i= 0; |
| 2515 | do |
| 2516 | { |
| 2517 | part_elem_value *list_value= list_val_it++; |
| 2518 | |
| 2519 | if (part_info->column_list) |
| 2520 | err+= add_column_list_values(str, part_info, list_value, |
| 2521 | create_info, alter_info); |
| 2522 | else |
| 2523 | { |
| 2524 | if (!list_value->unsigned_flag) |
| 2525 | err+= str->append_longlong(list_value->value); |
| 2526 | else |
| 2527 | err+= str->append_ulonglong(list_value->value); |
| 2528 | } |
| 2529 | if (i != (num_items-1)) |
| 2530 | err+= str->append(','); |
| 2531 | } while (++i < num_items); |
| 2532 | err+= str->append(')'); |
| 2533 | } |
| 2534 | else if (part_info->part_type == VERSIONING_PARTITION) |
| 2535 | { |
| 2536 | switch (p_elem->type()) |
| 2537 | { |
| 2538 | case partition_element::CURRENT: |
| 2539 | err+= str->append(STRING_WITH_LEN(" CURRENT" )); |
| 2540 | break; |
| 2541 | case partition_element::HISTORY: |
| 2542 | err+= str->append(STRING_WITH_LEN(" HISTORY" )); |
| 2543 | break; |
| 2544 | default: |
| 2545 | DBUG_ASSERT(0 && "wrong p_elem->type" ); |
| 2546 | } |
| 2547 | } |
| 2548 | end: |
| 2549 | return err; |
| 2550 | } |
| 2551 | |
| 2552 | |
| 2553 | /** |
| 2554 | Add 'KEY' word, with optional 'ALGORTIHM = N'. |
| 2555 | |
| 2556 | @param str String to write to. |
| 2557 | @param part_info partition_info holding the used key_algorithm |
| 2558 | |
| 2559 | @return Operation status. |
| 2560 | @retval 0 Success |
| 2561 | @retval != 0 Failure |
| 2562 | */ |
| 2563 | |
| 2564 | static int add_key_with_algorithm(String *str, partition_info *part_info) |
| 2565 | { |
| 2566 | int err= 0; |
| 2567 | err+= str->append(STRING_WITH_LEN("KEY " )); |
| 2568 | |
| 2569 | if (part_info->key_algorithm == partition_info::KEY_ALGORITHM_51) |
| 2570 | { |
| 2571 | err+= str->append(STRING_WITH_LEN("ALGORITHM = " )); |
| 2572 | err+= str->append_longlong(part_info->key_algorithm); |
| 2573 | err+= str->append(' '); |
| 2574 | } |
| 2575 | return err; |
| 2576 | } |
| 2577 | |
| 2578 | |
| 2579 | /* |
| 2580 | Generate the partition syntax from the partition data structure. |
| 2581 | Useful for support of generating defaults, SHOW CREATE TABLES |
| 2582 | and easy partition management. |
| 2583 | |
| 2584 | SYNOPSIS |
| 2585 | generate_partition_syntax() |
| 2586 | part_info The partitioning data structure |
| 2587 | buf_length A pointer to the returned buffer length |
| 2588 | show_partition_options Should we display partition options |
| 2589 | create_info Info generated by parser |
| 2590 | alter_info Info generated by parser |
| 2591 | |
| 2592 | RETURN VALUES |
| 2593 | NULL error |
| 2594 | buf, buf_length Buffer and its length |
| 2595 | |
| 2596 | DESCRIPTION |
| 2597 | Here we will generate the full syntax for the given command where all |
| 2598 | defaults have been expanded. By so doing the it is also possible to |
| 2599 | make lots of checks of correctness while at it. |
| 2600 | This could will also be reused for SHOW CREATE TABLES and also for all |
| 2601 | type ALTER TABLE commands focusing on changing the PARTITION structure |
| 2602 | in any fashion. |
| 2603 | |
| 2604 | The code is optimised for minimal code size since it is not used in any |
| 2605 | common queries. |
| 2606 | */ |
| 2607 | |
| 2608 | char *generate_partition_syntax(THD *thd, partition_info *part_info, |
| 2609 | uint *buf_length, |
| 2610 | bool show_partition_options, |
| 2611 | HA_CREATE_INFO *create_info, |
| 2612 | Alter_info *alter_info) |
| 2613 | { |
| 2614 | uint i,j, tot_num_parts, num_subparts; |
| 2615 | partition_element *part_elem; |
| 2616 | int err= 0; |
| 2617 | List_iterator<partition_element> part_it(part_info->partitions); |
| 2618 | StringBuffer<1024> str; |
| 2619 | DBUG_ENTER("generate_partition_syntax" ); |
| 2620 | |
| 2621 | err+= str.append(STRING_WITH_LEN(" PARTITION BY " )); |
| 2622 | switch (part_info->part_type) |
| 2623 | { |
| 2624 | case RANGE_PARTITION: |
| 2625 | err+= str.append(STRING_WITH_LEN("RANGE " )); |
| 2626 | break; |
| 2627 | case LIST_PARTITION: |
| 2628 | err+= str.append(STRING_WITH_LEN("LIST " )); |
| 2629 | break; |
| 2630 | case HASH_PARTITION: |
| 2631 | if (part_info->linear_hash_ind) |
| 2632 | err+= str.append(STRING_WITH_LEN("LINEAR " )); |
| 2633 | if (part_info->list_of_part_fields) |
| 2634 | { |
| 2635 | err+= add_key_with_algorithm(&str, part_info); |
| 2636 | err+= add_part_field_list(thd, &str, part_info->part_field_list); |
| 2637 | } |
| 2638 | else |
| 2639 | err+= str.append(STRING_WITH_LEN("HASH " )); |
| 2640 | break; |
| 2641 | case VERSIONING_PARTITION: |
| 2642 | err+= str.append(STRING_WITH_LEN("SYSTEM_TIME " )); |
| 2643 | break; |
| 2644 | default: |
| 2645 | DBUG_ASSERT(0); |
| 2646 | /* We really shouldn't get here, no use in continuing from here */ |
| 2647 | my_error(ER_OUT_OF_RESOURCES, MYF(ME_FATALERROR)); |
| 2648 | DBUG_RETURN(NULL); |
| 2649 | } |
| 2650 | if (part_info->part_type == VERSIONING_PARTITION) |
| 2651 | { |
| 2652 | Vers_part_info *vers_info= part_info->vers_info; |
| 2653 | DBUG_ASSERT(vers_info); |
| 2654 | if (vers_info->interval.is_set()) |
| 2655 | { |
| 2656 | err+= str.append(STRING_WITH_LEN("INTERVAL " )); |
| 2657 | err+= append_interval(&str, vers_info->interval.type, |
| 2658 | vers_info->interval.step); |
| 2659 | if (create_info) // not SHOW CREATE |
| 2660 | { |
| 2661 | err+= str.append(STRING_WITH_LEN(" STARTS " )); |
| 2662 | err+= str.append_ulonglong(vers_info->interval.start); |
| 2663 | } |
| 2664 | } |
| 2665 | if (vers_info->limit) |
| 2666 | { |
| 2667 | err+= str.append(STRING_WITH_LEN("LIMIT " )); |
| 2668 | err+= str.append_ulonglong(vers_info->limit); |
| 2669 | } |
| 2670 | } |
| 2671 | else if (part_info->part_expr) |
| 2672 | { |
| 2673 | err+= str.append('('); |
| 2674 | part_info->part_expr->print_for_table_def(&str); |
| 2675 | err+= str.append(')'); |
| 2676 | } |
| 2677 | else if (part_info->column_list) |
| 2678 | { |
| 2679 | err+= str.append(STRING_WITH_LEN(" COLUMNS" )); |
| 2680 | err+= add_part_field_list(thd, &str, part_info->part_field_list); |
| 2681 | } |
| 2682 | if ((!part_info->use_default_num_partitions) && |
| 2683 | part_info->use_default_partitions) |
| 2684 | { |
| 2685 | err+= str.append(STRING_WITH_LEN("\nPARTITIONS " )); |
| 2686 | err+= str.append_ulonglong(part_info->num_parts); |
| 2687 | } |
| 2688 | if (part_info->is_sub_partitioned()) |
| 2689 | { |
| 2690 | err+= str.append(STRING_WITH_LEN("\nSUBPARTITION BY " )); |
| 2691 | /* Must be hash partitioning for subpartitioning */ |
| 2692 | if (part_info->linear_hash_ind) |
| 2693 | err+= str.append(STRING_WITH_LEN("LINEAR " )); |
| 2694 | if (part_info->list_of_subpart_fields) |
| 2695 | { |
| 2696 | err+= add_key_with_algorithm(&str, part_info); |
| 2697 | err+= add_part_field_list(thd, &str, part_info->subpart_field_list); |
| 2698 | } |
| 2699 | else |
| 2700 | err+= str.append(STRING_WITH_LEN("HASH " )); |
| 2701 | if (part_info->subpart_expr) |
| 2702 | { |
| 2703 | err+= str.append('('); |
| 2704 | part_info->subpart_expr->print_for_table_def(&str); |
| 2705 | err+= str.append(')'); |
| 2706 | } |
| 2707 | if ((!part_info->use_default_num_subpartitions) && |
| 2708 | part_info->use_default_subpartitions) |
| 2709 | { |
| 2710 | err+= str.append(STRING_WITH_LEN("\nSUBPARTITIONS " )); |
| 2711 | err+= str.append_ulonglong(part_info->num_subparts); |
| 2712 | } |
| 2713 | } |
| 2714 | tot_num_parts= part_info->partitions.elements; |
| 2715 | num_subparts= part_info->num_subparts; |
| 2716 | |
| 2717 | if (!part_info->use_default_partitions) |
| 2718 | { |
| 2719 | bool first= TRUE; |
| 2720 | err+= str.append(STRING_WITH_LEN("\n(" )); |
| 2721 | i= 0; |
| 2722 | do |
| 2723 | { |
| 2724 | part_elem= part_it++; |
| 2725 | if (part_elem->part_state != PART_TO_BE_DROPPED && |
| 2726 | part_elem->part_state != PART_REORGED_DROPPED) |
| 2727 | { |
| 2728 | if (!first) |
| 2729 | err+= str.append(STRING_WITH_LEN(",\n " )); |
| 2730 | first= FALSE; |
| 2731 | err+= str.append(STRING_WITH_LEN("PARTITION " )); |
| 2732 | err+= append_identifier(thd, &str, part_elem->partition_name, |
| 2733 | strlen(part_elem->partition_name)); |
| 2734 | err+= add_partition_values(&str, part_info, part_elem, |
| 2735 | create_info, alter_info); |
| 2736 | if (!part_info->is_sub_partitioned() || |
| 2737 | part_info->use_default_subpartitions) |
| 2738 | { |
| 2739 | if (show_partition_options) |
| 2740 | err+= add_partition_options(&str, part_elem); |
| 2741 | } |
| 2742 | else |
| 2743 | { |
| 2744 | err+= str.append(STRING_WITH_LEN("\n (" )); |
| 2745 | List_iterator<partition_element> sub_it(part_elem->subpartitions); |
| 2746 | j= 0; |
| 2747 | do |
| 2748 | { |
| 2749 | part_elem= sub_it++; |
| 2750 | err+= str.append(STRING_WITH_LEN("SUBPARTITION " )); |
| 2751 | err+= append_identifier(thd, &str, part_elem->partition_name, |
| 2752 | strlen(part_elem->partition_name)); |
| 2753 | if (show_partition_options) |
| 2754 | err+= add_partition_options(&str, part_elem); |
| 2755 | if (j != (num_subparts-1)) |
| 2756 | err+= str.append(STRING_WITH_LEN(",\n " )); |
| 2757 | else |
| 2758 | err+= str.append(')'); |
| 2759 | } while (++j < num_subparts); |
| 2760 | } |
| 2761 | } |
| 2762 | if (i == (tot_num_parts-1)) |
| 2763 | err+= str.append(')'); |
| 2764 | } while (++i < tot_num_parts); |
| 2765 | } |
| 2766 | if (err) |
| 2767 | DBUG_RETURN(NULL); |
| 2768 | *buf_length= str.length(); |
| 2769 | DBUG_RETURN(thd->strmake(str.ptr(), str.length())); |
| 2770 | } |
| 2771 | |
| 2772 | |
| 2773 | /* |
| 2774 | Check if partition key fields are modified and if it can be handled by the |
| 2775 | underlying storage engine. |
| 2776 | |
| 2777 | SYNOPSIS |
| 2778 | partition_key_modified |
| 2779 | table TABLE object for which partition fields are set-up |
| 2780 | fields Bitmap representing fields to be modified |
| 2781 | |
| 2782 | RETURN VALUES |
| 2783 | TRUE Need special handling of UPDATE |
| 2784 | FALSE Normal UPDATE handling is ok |
| 2785 | */ |
| 2786 | |
| 2787 | bool partition_key_modified(TABLE *table, const MY_BITMAP *fields) |
| 2788 | { |
| 2789 | Field **fld; |
| 2790 | partition_info *part_info= table->part_info; |
| 2791 | DBUG_ENTER("partition_key_modified" ); |
| 2792 | |
| 2793 | if (!part_info) |
| 2794 | DBUG_RETURN(FALSE); |
| 2795 | if (table->s->db_type()->partition_flags && |
| 2796 | (table->s->db_type()->partition_flags() & HA_CAN_UPDATE_PARTITION_KEY)) |
| 2797 | DBUG_RETURN(FALSE); |
| 2798 | for (fld= part_info->full_part_field_array; *fld; fld++) |
| 2799 | if (bitmap_is_set(fields, (*fld)->field_index)) |
| 2800 | DBUG_RETURN(TRUE); |
| 2801 | DBUG_RETURN(FALSE); |
| 2802 | } |
| 2803 | |
| 2804 | |
| 2805 | /* |
| 2806 | A function to handle correct handling of NULL values in partition |
| 2807 | functions. |
| 2808 | SYNOPSIS |
| 2809 | part_val_int() |
| 2810 | item_expr The item expression to evaluate |
| 2811 | out:result The value of the partition function, |
| 2812 | LONGLONG_MIN if any null value in function |
| 2813 | RETURN VALUES |
| 2814 | TRUE Error in val_int() |
| 2815 | FALSE ok |
| 2816 | */ |
| 2817 | |
| 2818 | static inline int part_val_int(Item *item_expr, longlong *result) |
| 2819 | { |
| 2820 | *result= item_expr->val_int(); |
| 2821 | if (item_expr->null_value) |
| 2822 | { |
| 2823 | if (unlikely(current_thd->is_error())) |
| 2824 | return TRUE; |
| 2825 | *result= LONGLONG_MIN; |
| 2826 | } |
| 2827 | return FALSE; |
| 2828 | } |
| 2829 | |
| 2830 | |
| 2831 | /* |
| 2832 | The next set of functions are used to calculate the partition identity. |
| 2833 | A handler sets up a variable that corresponds to one of these functions |
| 2834 | to be able to quickly call it whenever the partition id needs to calculated |
| 2835 | based on the record in table->record[0] (or set up to fake that). |
| 2836 | There are 4 functions for hash partitioning and 2 for RANGE/LIST partitions. |
| 2837 | In addition there are 4 variants for RANGE subpartitioning and 4 variants |
| 2838 | for LIST subpartitioning thus in total there are 14 variants of this |
| 2839 | function. |
| 2840 | |
| 2841 | We have a set of support functions for these 14 variants. There are 4 |
| 2842 | variants of hash functions and there is a function for each. The KEY |
| 2843 | partitioning uses the function calculate_key_hash_value to calculate the hash |
| 2844 | value based on an array of fields. The linear hash variants uses the |
| 2845 | method get_part_id_from_linear_hash to get the partition id using the |
| 2846 | hash value and some parameters calculated from the number of partitions. |
| 2847 | */ |
| 2848 | |
| 2849 | /* |
| 2850 | A simple support function to calculate part_id given local part and |
| 2851 | sub part. |
| 2852 | |
| 2853 | SYNOPSIS |
| 2854 | get_part_id_for_sub() |
| 2855 | loc_part_id Local partition id |
| 2856 | sub_part_id Subpartition id |
| 2857 | num_subparts Number of subparts |
| 2858 | */ |
| 2859 | |
| 2860 | inline |
| 2861 | static uint32 get_part_id_for_sub(uint32 loc_part_id, uint32 sub_part_id, |
| 2862 | uint num_subparts) |
| 2863 | { |
| 2864 | return (uint32)((loc_part_id * num_subparts) + sub_part_id); |
| 2865 | } |
| 2866 | |
| 2867 | |
| 2868 | /* |
| 2869 | Calculate part_id for (SUB)PARTITION BY HASH |
| 2870 | |
| 2871 | SYNOPSIS |
| 2872 | get_part_id_hash() |
| 2873 | num_parts Number of hash partitions |
| 2874 | part_expr Item tree of hash function |
| 2875 | out:part_id The returned partition id |
| 2876 | out:func_value Value of hash function |
| 2877 | |
| 2878 | RETURN VALUE |
| 2879 | != 0 Error code |
| 2880 | FALSE Success |
| 2881 | */ |
| 2882 | |
| 2883 | static int get_part_id_hash(uint num_parts, |
| 2884 | Item *part_expr, |
| 2885 | uint32 *part_id, |
| 2886 | longlong *func_value) |
| 2887 | { |
| 2888 | longlong int_hash_id; |
| 2889 | DBUG_ENTER("get_part_id_hash" ); |
| 2890 | |
| 2891 | if (part_val_int(part_expr, func_value)) |
| 2892 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 2893 | |
| 2894 | int_hash_id= *func_value % num_parts; |
| 2895 | |
| 2896 | *part_id= int_hash_id < 0 ? (uint32) -int_hash_id : (uint32) int_hash_id; |
| 2897 | DBUG_RETURN(FALSE); |
| 2898 | } |
| 2899 | |
| 2900 | |
| 2901 | /* |
| 2902 | Calculate part_id for (SUB)PARTITION BY LINEAR HASH |
| 2903 | |
| 2904 | SYNOPSIS |
| 2905 | get_part_id_linear_hash() |
| 2906 | part_info A reference to the partition_info struct where all the |
| 2907 | desired information is given |
| 2908 | num_parts Number of hash partitions |
| 2909 | part_expr Item tree of hash function |
| 2910 | out:part_id The returned partition id |
| 2911 | out:func_value Value of hash function |
| 2912 | |
| 2913 | RETURN VALUE |
| 2914 | != 0 Error code |
| 2915 | 0 OK |
| 2916 | */ |
| 2917 | |
| 2918 | static int get_part_id_linear_hash(partition_info *part_info, |
| 2919 | uint num_parts, |
| 2920 | Item *part_expr, |
| 2921 | uint32 *part_id, |
| 2922 | longlong *func_value) |
| 2923 | { |
| 2924 | DBUG_ENTER("get_part_id_linear_hash" ); |
| 2925 | |
| 2926 | if (part_val_int(part_expr, func_value)) |
| 2927 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 2928 | |
| 2929 | *part_id= get_part_id_from_linear_hash(*func_value, |
| 2930 | part_info->linear_hash_mask, |
| 2931 | num_parts); |
| 2932 | DBUG_RETURN(FALSE); |
| 2933 | } |
| 2934 | |
| 2935 | |
| 2936 | /** |
| 2937 | Calculate part_id for (SUB)PARTITION BY KEY |
| 2938 | |
| 2939 | @param file Handler to storage engine |
| 2940 | @param field_array Array of fields for PARTTION KEY |
| 2941 | @param num_parts Number of KEY partitions |
| 2942 | @param func_value[out] Returns calculated hash value |
| 2943 | |
| 2944 | @return Calculated partition id |
| 2945 | */ |
| 2946 | |
| 2947 | inline |
| 2948 | static uint32 get_part_id_key(handler *file, |
| 2949 | Field **field_array, |
| 2950 | uint num_parts, |
| 2951 | longlong *func_value) |
| 2952 | { |
| 2953 | DBUG_ENTER("get_part_id_key" ); |
| 2954 | *func_value= ha_partition::calculate_key_hash_value(field_array); |
| 2955 | DBUG_RETURN((uint32) (*func_value % num_parts)); |
| 2956 | } |
| 2957 | |
| 2958 | |
| 2959 | /* |
| 2960 | Calculate part_id for (SUB)PARTITION BY LINEAR KEY |
| 2961 | |
| 2962 | SYNOPSIS |
| 2963 | get_part_id_linear_key() |
| 2964 | part_info A reference to the partition_info struct where all the |
| 2965 | desired information is given |
| 2966 | field_array Array of fields for PARTTION KEY |
| 2967 | num_parts Number of KEY partitions |
| 2968 | |
| 2969 | RETURN VALUE |
| 2970 | Calculated partition id |
| 2971 | */ |
| 2972 | |
| 2973 | inline |
| 2974 | static uint32 get_part_id_linear_key(partition_info *part_info, |
| 2975 | Field **field_array, |
| 2976 | uint num_parts, |
| 2977 | longlong *func_value) |
| 2978 | { |
| 2979 | DBUG_ENTER("get_part_id_linear_key" ); |
| 2980 | |
| 2981 | *func_value= ha_partition::calculate_key_hash_value(field_array); |
| 2982 | DBUG_RETURN(get_part_id_from_linear_hash(*func_value, |
| 2983 | part_info->linear_hash_mask, |
| 2984 | num_parts)); |
| 2985 | } |
| 2986 | |
| 2987 | /* |
| 2988 | Copy to field buffers and set up field pointers |
| 2989 | |
| 2990 | SYNOPSIS |
| 2991 | copy_to_part_field_buffers() |
| 2992 | ptr Array of fields to copy |
| 2993 | field_bufs Array of field buffers to copy to |
| 2994 | restore_ptr Array of pointers to restore to |
| 2995 | |
| 2996 | RETURN VALUES |
| 2997 | NONE |
| 2998 | DESCRIPTION |
| 2999 | This routine is used to take the data from field pointer, convert |
| 3000 | it to a standard format and store this format in a field buffer |
| 3001 | allocated for this purpose. Next the field pointers are moved to |
| 3002 | point to the field buffers. There is a separate to restore the |
| 3003 | field pointers after this call. |
| 3004 | */ |
| 3005 | |
| 3006 | static void copy_to_part_field_buffers(Field **ptr, |
| 3007 | uchar **field_bufs, |
| 3008 | uchar **restore_ptr) |
| 3009 | { |
| 3010 | Field *field; |
| 3011 | while ((field= *(ptr++))) |
| 3012 | { |
| 3013 | *restore_ptr= field->ptr; |
| 3014 | restore_ptr++; |
| 3015 | if (!field->maybe_null() || !field->is_null()) |
| 3016 | { |
| 3017 | CHARSET_INFO *cs= field->charset(); |
| 3018 | uint max_len= field->pack_length(); |
| 3019 | uint data_len= field->data_length(); |
| 3020 | uchar *field_buf= *field_bufs; |
| 3021 | /* |
| 3022 | We only use the field buffer for VARCHAR and CHAR strings |
| 3023 | which isn't of a binary collation. We also only use the |
| 3024 | field buffer for fields which are not currently NULL. |
| 3025 | The field buffer will store a normalised string. We use |
| 3026 | the strnxfrm method to normalise the string. |
| 3027 | */ |
| 3028 | if (field->type() == MYSQL_TYPE_VARCHAR) |
| 3029 | { |
| 3030 | uint len_bytes= ((Field_varstring*)field)->length_bytes; |
| 3031 | my_strnxfrm(cs, field_buf + len_bytes, max_len, |
| 3032 | field->ptr + len_bytes, data_len); |
| 3033 | if (len_bytes == 1) |
| 3034 | *field_buf= (uchar) data_len; |
| 3035 | else |
| 3036 | int2store(field_buf, data_len); |
| 3037 | } |
| 3038 | else |
| 3039 | { |
| 3040 | my_strnxfrm(cs, field_buf, max_len, |
| 3041 | field->ptr, max_len); |
| 3042 | } |
| 3043 | field->ptr= field_buf; |
| 3044 | } |
| 3045 | field_bufs++; |
| 3046 | } |
| 3047 | return; |
| 3048 | } |
| 3049 | |
| 3050 | /* |
| 3051 | Restore field pointers |
| 3052 | SYNOPSIS |
| 3053 | restore_part_field_pointers() |
| 3054 | ptr Array of fields to restore |
| 3055 | restore_ptr Array of field pointers to restore to |
| 3056 | |
| 3057 | RETURN VALUES |
| 3058 | */ |
| 3059 | |
| 3060 | static void restore_part_field_pointers(Field **ptr, uchar **restore_ptr) |
| 3061 | { |
| 3062 | Field *field; |
| 3063 | while ((field= *(ptr++))) |
| 3064 | { |
| 3065 | field->ptr= *restore_ptr; |
| 3066 | restore_ptr++; |
| 3067 | } |
| 3068 | return; |
| 3069 | } |
| 3070 | |
| 3071 | /* |
| 3072 | This function is used to calculate the partition id where all partition |
| 3073 | fields have been prepared to point to a record where the partition field |
| 3074 | values are bound. |
| 3075 | |
| 3076 | SYNOPSIS |
| 3077 | get_partition_id() |
| 3078 | part_info A reference to the partition_info struct where all the |
| 3079 | desired information is given |
| 3080 | out:part_id The partition id is returned through this pointer |
| 3081 | out:func_value Value of partition function (longlong) |
| 3082 | |
| 3083 | RETURN VALUE |
| 3084 | part_id Partition id of partition that would contain |
| 3085 | row with given values of PF-fields |
| 3086 | HA_ERR_NO_PARTITION_FOUND The fields of the partition function didn't |
| 3087 | fit into any partition and thus the values of |
| 3088 | the PF-fields are not allowed. |
| 3089 | |
| 3090 | DESCRIPTION |
| 3091 | A routine used from write_row, update_row and delete_row from any |
| 3092 | handler supporting partitioning. It is also a support routine for |
| 3093 | get_partition_set used to find the set of partitions needed to scan |
| 3094 | for a certain index scan or full table scan. |
| 3095 | |
| 3096 | It is actually 9 different variants of this function which are called |
| 3097 | through a function pointer. |
| 3098 | |
| 3099 | get_partition_id_list |
| 3100 | get_partition_id_list_col |
| 3101 | get_partition_id_range |
| 3102 | get_partition_id_range_col |
| 3103 | get_partition_id_hash_nosub |
| 3104 | get_partition_id_key_nosub |
| 3105 | get_partition_id_linear_hash_nosub |
| 3106 | get_partition_id_linear_key_nosub |
| 3107 | get_partition_id_with_sub |
| 3108 | */ |
| 3109 | |
| 3110 | /* |
| 3111 | This function is used to calculate the main partition to use in the case of |
| 3112 | subpartitioning and we don't know enough to get the partition identity in |
| 3113 | total. |
| 3114 | |
| 3115 | SYNOPSIS |
| 3116 | get_part_partition_id() |
| 3117 | part_info A reference to the partition_info struct where all the |
| 3118 | desired information is given |
| 3119 | out:part_id The partition id is returned through this pointer |
| 3120 | out:func_value The value calculated by partition function |
| 3121 | |
| 3122 | RETURN VALUE |
| 3123 | HA_ERR_NO_PARTITION_FOUND The fields of the partition function didn't |
| 3124 | fit into any partition and thus the values of |
| 3125 | the PF-fields are not allowed. |
| 3126 | 0 OK |
| 3127 | |
| 3128 | DESCRIPTION |
| 3129 | |
| 3130 | It is actually 8 different variants of this function which are called |
| 3131 | through a function pointer. |
| 3132 | |
| 3133 | get_partition_id_list |
| 3134 | get_partition_id_list_col |
| 3135 | get_partition_id_range |
| 3136 | get_partition_id_range_col |
| 3137 | get_partition_id_hash_nosub |
| 3138 | get_partition_id_key_nosub |
| 3139 | get_partition_id_linear_hash_nosub |
| 3140 | get_partition_id_linear_key_nosub |
| 3141 | */ |
| 3142 | |
| 3143 | static int get_part_id_charset_func_part(partition_info *part_info, |
| 3144 | uint32 *part_id, |
| 3145 | longlong *func_value) |
| 3146 | { |
| 3147 | int res; |
| 3148 | DBUG_ENTER("get_part_id_charset_func_part" ); |
| 3149 | |
| 3150 | copy_to_part_field_buffers(part_info->part_charset_field_array, |
| 3151 | part_info->part_field_buffers, |
| 3152 | part_info->restore_part_field_ptrs); |
| 3153 | res= part_info->get_part_partition_id_charset(part_info, |
| 3154 | part_id, func_value); |
| 3155 | restore_part_field_pointers(part_info->part_charset_field_array, |
| 3156 | part_info->restore_part_field_ptrs); |
| 3157 | DBUG_RETURN(res); |
| 3158 | } |
| 3159 | |
| 3160 | |
| 3161 | static int get_part_id_charset_func_subpart(partition_info *part_info, |
| 3162 | uint32 *part_id) |
| 3163 | { |
| 3164 | int res; |
| 3165 | DBUG_ENTER("get_part_id_charset_func_subpart" ); |
| 3166 | |
| 3167 | copy_to_part_field_buffers(part_info->subpart_charset_field_array, |
| 3168 | part_info->subpart_field_buffers, |
| 3169 | part_info->restore_subpart_field_ptrs); |
| 3170 | res= part_info->get_subpartition_id_charset(part_info, part_id); |
| 3171 | restore_part_field_pointers(part_info->subpart_charset_field_array, |
| 3172 | part_info->restore_subpart_field_ptrs); |
| 3173 | DBUG_RETURN(res); |
| 3174 | } |
| 3175 | |
| 3176 | int get_partition_id_list_col(partition_info *part_info, |
| 3177 | uint32 *part_id, |
| 3178 | longlong *func_value) |
| 3179 | { |
| 3180 | part_column_list_val *list_col_array= part_info->list_col_array; |
| 3181 | uint num_columns= part_info->part_field_list.elements; |
| 3182 | int list_index, cmp; |
| 3183 | int min_list_index= 0; |
| 3184 | int max_list_index= part_info->num_list_values - 1; |
| 3185 | DBUG_ENTER("get_partition_id_list_col" ); |
| 3186 | |
| 3187 | while (max_list_index >= min_list_index) |
| 3188 | { |
| 3189 | list_index= (max_list_index + min_list_index) >> 1; |
| 3190 | cmp= cmp_rec_and_tuple(list_col_array + list_index*num_columns, |
| 3191 | num_columns); |
| 3192 | if (cmp > 0) |
| 3193 | min_list_index= list_index + 1; |
| 3194 | else if (cmp < 0) |
| 3195 | { |
| 3196 | if (!list_index) |
| 3197 | goto notfound; |
| 3198 | max_list_index= list_index - 1; |
| 3199 | } |
| 3200 | else |
| 3201 | { |
| 3202 | *part_id= (uint32)list_col_array[list_index*num_columns].partition_id; |
| 3203 | DBUG_RETURN(0); |
| 3204 | } |
| 3205 | } |
| 3206 | notfound: |
| 3207 | if (part_info->defined_max_value) |
| 3208 | { |
| 3209 | *part_id= part_info->default_partition_id; |
| 3210 | DBUG_RETURN(0); |
| 3211 | } |
| 3212 | *part_id= 0; |
| 3213 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 3214 | } |
| 3215 | |
| 3216 | |
| 3217 | int get_partition_id_list(partition_info *part_info, |
| 3218 | uint32 *part_id, |
| 3219 | longlong *func_value) |
| 3220 | { |
| 3221 | LIST_PART_ENTRY *list_array= part_info->list_array; |
| 3222 | int list_index; |
| 3223 | int min_list_index= 0; |
| 3224 | int max_list_index= part_info->num_list_values - 1; |
| 3225 | longlong part_func_value; |
| 3226 | int error= part_val_int(part_info->part_expr, &part_func_value); |
| 3227 | longlong list_value; |
| 3228 | bool unsigned_flag= part_info->part_expr->unsigned_flag; |
| 3229 | DBUG_ENTER("get_partition_id_list" ); |
| 3230 | |
| 3231 | if (error) |
| 3232 | goto notfound; |
| 3233 | |
| 3234 | if (part_info->part_expr->null_value) |
| 3235 | { |
| 3236 | if (part_info->has_null_value) |
| 3237 | { |
| 3238 | *part_id= part_info->has_null_part_id; |
| 3239 | DBUG_RETURN(0); |
| 3240 | } |
| 3241 | goto notfound; |
| 3242 | } |
| 3243 | *func_value= part_func_value; |
| 3244 | if (unsigned_flag) |
| 3245 | part_func_value-= 0x8000000000000000ULL; |
| 3246 | while (max_list_index >= min_list_index) |
| 3247 | { |
| 3248 | list_index= (max_list_index + min_list_index) >> 1; |
| 3249 | list_value= list_array[list_index].list_value; |
| 3250 | if (list_value < part_func_value) |
| 3251 | min_list_index= list_index + 1; |
| 3252 | else if (list_value > part_func_value) |
| 3253 | { |
| 3254 | if (!list_index) |
| 3255 | goto notfound; |
| 3256 | max_list_index= list_index - 1; |
| 3257 | } |
| 3258 | else |
| 3259 | { |
| 3260 | *part_id= (uint32)list_array[list_index].partition_id; |
| 3261 | DBUG_RETURN(0); |
| 3262 | } |
| 3263 | } |
| 3264 | notfound: |
| 3265 | if (part_info->defined_max_value) |
| 3266 | { |
| 3267 | *part_id= part_info->default_partition_id; |
| 3268 | DBUG_RETURN(0); |
| 3269 | } |
| 3270 | *part_id= 0; |
| 3271 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 3272 | } |
| 3273 | |
| 3274 | |
| 3275 | uint32 get_partition_id_cols_list_for_endpoint(partition_info *part_info, |
| 3276 | bool left_endpoint, |
| 3277 | bool include_endpoint, |
| 3278 | uint32 nparts) |
| 3279 | { |
| 3280 | part_column_list_val *list_col_array= part_info->list_col_array; |
| 3281 | uint num_columns= part_info->part_field_list.elements; |
| 3282 | uint list_index; |
| 3283 | uint min_list_index= 0; |
| 3284 | int cmp; |
| 3285 | /* Notice that max_list_index = last_index + 1 here! */ |
| 3286 | uint max_list_index= part_info->num_list_values; |
| 3287 | DBUG_ENTER("get_partition_id_cols_list_for_endpoint" ); |
| 3288 | |
| 3289 | /* Find the matching partition (including taking endpoint into account). */ |
| 3290 | do |
| 3291 | { |
| 3292 | /* Midpoint, adjusted down, so it can never be >= max_list_index. */ |
| 3293 | list_index= (max_list_index + min_list_index) >> 1; |
| 3294 | cmp= cmp_rec_and_tuple_prune(list_col_array + list_index*num_columns, |
| 3295 | nparts, left_endpoint, include_endpoint); |
| 3296 | if (cmp > 0) |
| 3297 | { |
| 3298 | min_list_index= list_index + 1; |
| 3299 | } |
| 3300 | else |
| 3301 | { |
| 3302 | max_list_index= list_index; |
| 3303 | if (cmp == 0) |
| 3304 | break; |
| 3305 | } |
| 3306 | } while (max_list_index > min_list_index); |
| 3307 | list_index= max_list_index; |
| 3308 | |
| 3309 | /* Given value must be LESS THAN or EQUAL to the found partition. */ |
| 3310 | DBUG_ASSERT(list_index == part_info->num_list_values || |
| 3311 | (0 >= cmp_rec_and_tuple_prune(list_col_array + |
| 3312 | list_index*num_columns, |
| 3313 | nparts, left_endpoint, |
| 3314 | include_endpoint))); |
| 3315 | /* Given value must be GREATER THAN the previous partition. */ |
| 3316 | DBUG_ASSERT(list_index == 0 || |
| 3317 | (0 < cmp_rec_and_tuple_prune(list_col_array + |
| 3318 | (list_index - 1)*num_columns, |
| 3319 | nparts, left_endpoint, |
| 3320 | include_endpoint))); |
| 3321 | |
| 3322 | /* Include the right endpoint if not already passed end of array. */ |
| 3323 | if (!left_endpoint && include_endpoint && cmp == 0 && |
| 3324 | list_index < part_info->num_list_values) |
| 3325 | list_index++; |
| 3326 | |
| 3327 | DBUG_RETURN(list_index); |
| 3328 | } |
| 3329 | |
| 3330 | |
| 3331 | /** |
| 3332 | Find the sub-array part_info->list_array that corresponds to given interval. |
| 3333 | |
| 3334 | @param part_info Partitioning info (partitioning type must be LIST) |
| 3335 | @param left_endpoint TRUE - the interval is [a; +inf) or (a; +inf) |
| 3336 | FALSE - the interval is (-inf; a] or (-inf; a) |
| 3337 | @param include_endpoint TRUE iff the interval includes the endpoint |
| 3338 | |
| 3339 | This function finds the sub-array of part_info->list_array where values of |
| 3340 | list_array[idx].list_value are contained within the specifed interval. |
| 3341 | list_array is ordered by list_value, so |
| 3342 | 1. For [a; +inf) or (a; +inf)-type intervals (left_endpoint==TRUE), the |
| 3343 | sought sub-array starts at some index idx and continues till array end. |
| 3344 | The function returns first number idx, such that |
| 3345 | list_array[idx].list_value is contained within the passed interval. |
| 3346 | |
| 3347 | 2. For (-inf; a] or (-inf; a)-type intervals (left_endpoint==FALSE), the |
| 3348 | sought sub-array starts at array start and continues till some last |
| 3349 | index idx. |
| 3350 | The function returns first number idx, such that |
| 3351 | list_array[idx].list_value is NOT contained within the passed interval. |
| 3352 | If all array elements are contained, part_info->num_list_values is |
| 3353 | returned. |
| 3354 | |
| 3355 | @note The caller will call this function and then will run along the |
| 3356 | sub-array of list_array to collect partition ids. If the number of list |
| 3357 | values is significantly higher then number of partitions, this could be slow |
| 3358 | and we could invent some other approach. The "run over list array" part is |
| 3359 | already wrapped in a get_next()-like function. |
| 3360 | |
| 3361 | @return The index of corresponding sub-array of part_info->list_array. |
| 3362 | */ |
| 3363 | |
| 3364 | uint32 get_list_array_idx_for_endpoint_charset(partition_info *part_info, |
| 3365 | bool left_endpoint, |
| 3366 | bool include_endpoint) |
| 3367 | { |
| 3368 | uint32 res; |
| 3369 | copy_to_part_field_buffers(part_info->part_field_array, |
| 3370 | part_info->part_field_buffers, |
| 3371 | part_info->restore_part_field_ptrs); |
| 3372 | res= get_list_array_idx_for_endpoint(part_info, left_endpoint, |
| 3373 | include_endpoint); |
| 3374 | restore_part_field_pointers(part_info->part_field_array, |
| 3375 | part_info->restore_part_field_ptrs); |
| 3376 | return res; |
| 3377 | } |
| 3378 | |
| 3379 | uint32 get_list_array_idx_for_endpoint(partition_info *part_info, |
| 3380 | bool left_endpoint, |
| 3381 | bool include_endpoint) |
| 3382 | { |
| 3383 | LIST_PART_ENTRY *list_array= part_info->list_array; |
| 3384 | uint list_index; |
| 3385 | uint min_list_index= 0, max_list_index= part_info->num_list_values - 1; |
| 3386 | longlong list_value; |
| 3387 | /* Get the partitioning function value for the endpoint */ |
| 3388 | longlong part_func_value= |
| 3389 | part_info->part_expr->val_int_endpoint(left_endpoint, &include_endpoint); |
| 3390 | bool unsigned_flag= part_info->part_expr->unsigned_flag; |
| 3391 | DBUG_ENTER("get_list_array_idx_for_endpoint" ); |
| 3392 | |
| 3393 | if (part_info->part_expr->null_value) |
| 3394 | { |
| 3395 | /* |
| 3396 | Special handling for MONOTONIC functions that can return NULL for |
| 3397 | values that are comparable. I.e. |
| 3398 | '2000-00-00' can be compared to '2000-01-01' but TO_DAYS('2000-00-00') |
| 3399 | returns NULL which cannot be compared used <, >, <=, >= etc. |
| 3400 | |
| 3401 | Otherwise, just return the the first index (lowest value). |
| 3402 | */ |
| 3403 | enum_monotonicity_info monotonic; |
| 3404 | monotonic= part_info->part_expr->get_monotonicity_info(); |
| 3405 | if (monotonic != MONOTONIC_INCREASING_NOT_NULL && |
| 3406 | monotonic != MONOTONIC_STRICT_INCREASING_NOT_NULL) |
| 3407 | { |
| 3408 | /* F(col) can not return NULL, return index with lowest value */ |
| 3409 | DBUG_RETURN(0); |
| 3410 | } |
| 3411 | } |
| 3412 | |
| 3413 | if (unsigned_flag) |
| 3414 | part_func_value-= 0x8000000000000000ULL; |
| 3415 | DBUG_ASSERT(part_info->num_list_values); |
| 3416 | do |
| 3417 | { |
| 3418 | list_index= (max_list_index + min_list_index) >> 1; |
| 3419 | list_value= list_array[list_index].list_value; |
| 3420 | if (list_value < part_func_value) |
| 3421 | min_list_index= list_index + 1; |
| 3422 | else if (list_value > part_func_value) |
| 3423 | { |
| 3424 | if (!list_index) |
| 3425 | goto notfound; |
| 3426 | max_list_index= list_index - 1; |
| 3427 | } |
| 3428 | else |
| 3429 | { |
| 3430 | DBUG_RETURN(list_index + MY_TEST(left_endpoint ^ include_endpoint)); |
| 3431 | } |
| 3432 | } while (max_list_index >= min_list_index); |
| 3433 | notfound: |
| 3434 | if (list_value < part_func_value) |
| 3435 | list_index++; |
| 3436 | DBUG_RETURN(list_index); |
| 3437 | } |
| 3438 | |
| 3439 | |
| 3440 | int get_partition_id_range_col(partition_info *part_info, |
| 3441 | uint32 *part_id, |
| 3442 | longlong *func_value) |
| 3443 | { |
| 3444 | part_column_list_val *range_col_array= part_info->range_col_array; |
| 3445 | uint num_columns= part_info->part_field_list.elements; |
| 3446 | uint max_partition= part_info->num_parts - 1; |
| 3447 | uint min_part_id= 0; |
| 3448 | uint max_part_id= max_partition; |
| 3449 | uint loc_part_id; |
| 3450 | DBUG_ENTER("get_partition_id_range_col" ); |
| 3451 | |
| 3452 | while (max_part_id > min_part_id) |
| 3453 | { |
| 3454 | loc_part_id= (max_part_id + min_part_id + 1) >> 1; |
| 3455 | if (cmp_rec_and_tuple(range_col_array + loc_part_id*num_columns, |
| 3456 | num_columns) >= 0) |
| 3457 | min_part_id= loc_part_id + 1; |
| 3458 | else |
| 3459 | max_part_id= loc_part_id - 1; |
| 3460 | } |
| 3461 | loc_part_id= max_part_id; |
| 3462 | if (loc_part_id != max_partition) |
| 3463 | if (cmp_rec_and_tuple(range_col_array + loc_part_id*num_columns, |
| 3464 | num_columns) >= 0) |
| 3465 | loc_part_id++; |
| 3466 | *part_id= (uint32)loc_part_id; |
| 3467 | if (loc_part_id == max_partition && |
| 3468 | (cmp_rec_and_tuple(range_col_array + loc_part_id*num_columns, |
| 3469 | num_columns) >= 0)) |
| 3470 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 3471 | |
| 3472 | DBUG_PRINT("exit" ,("partition: %d" , *part_id)); |
| 3473 | DBUG_RETURN(0); |
| 3474 | } |
| 3475 | |
| 3476 | |
| 3477 | int vers_get_partition_id(partition_info *part_info, uint32 *part_id, |
| 3478 | longlong *func_value) |
| 3479 | { |
| 3480 | DBUG_ENTER("vers_get_partition_id" ); |
| 3481 | Field *row_end= part_info->part_field_array[STAT_TRX_END]; |
| 3482 | Vers_part_info *vers_info= part_info->vers_info; |
| 3483 | |
| 3484 | if (row_end->is_max() || row_end->is_null()) |
| 3485 | *part_id= vers_info->now_part->id; |
| 3486 | else // row is historical |
| 3487 | { |
| 3488 | longlong *range_value= part_info->range_int_array; |
| 3489 | uint max_hist_id= part_info->num_parts - 2; |
| 3490 | uint min_hist_id= 0, loc_hist_id= vers_info->hist_part->id; |
| 3491 | ulong unused; |
| 3492 | my_time_t ts; |
| 3493 | |
| 3494 | if (!range_value) |
| 3495 | goto done; // fastpath |
| 3496 | |
| 3497 | ts= row_end->get_timestamp(&unused); |
| 3498 | if ((loc_hist_id == 0 || range_value[loc_hist_id - 1] < ts) && |
| 3499 | (loc_hist_id == max_hist_id || range_value[loc_hist_id] >= ts)) |
| 3500 | goto done; // fastpath |
| 3501 | |
| 3502 | while (max_hist_id > min_hist_id) |
| 3503 | { |
| 3504 | loc_hist_id= (max_hist_id + min_hist_id) / 2; |
| 3505 | if (range_value[loc_hist_id] <= ts) |
| 3506 | min_hist_id= loc_hist_id + 1; |
| 3507 | else |
| 3508 | max_hist_id= loc_hist_id; |
| 3509 | } |
| 3510 | loc_hist_id= max_hist_id; |
| 3511 | done: |
| 3512 | *part_id= (uint32)loc_hist_id; |
| 3513 | } |
| 3514 | DBUG_PRINT("exit" ,("partition: %d" , *part_id)); |
| 3515 | DBUG_RETURN(0); |
| 3516 | } |
| 3517 | |
| 3518 | |
| 3519 | int get_partition_id_range(partition_info *part_info, |
| 3520 | uint32 *part_id, |
| 3521 | longlong *func_value) |
| 3522 | { |
| 3523 | longlong *range_array= part_info->range_int_array; |
| 3524 | uint max_partition= part_info->num_parts - 1; |
| 3525 | uint min_part_id= 0; |
| 3526 | uint max_part_id= max_partition; |
| 3527 | uint loc_part_id; |
| 3528 | longlong part_func_value; |
| 3529 | int error= part_val_int(part_info->part_expr, &part_func_value); |
| 3530 | bool unsigned_flag= part_info->part_expr->unsigned_flag; |
| 3531 | DBUG_ENTER("get_partition_id_range" ); |
| 3532 | |
| 3533 | if (unlikely(error)) |
| 3534 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 3535 | |
| 3536 | if (part_info->part_expr->null_value) |
| 3537 | { |
| 3538 | *part_id= 0; |
| 3539 | DBUG_RETURN(0); |
| 3540 | } |
| 3541 | *func_value= part_func_value; |
| 3542 | if (unsigned_flag) |
| 3543 | part_func_value-= 0x8000000000000000ULL; |
| 3544 | /* Search for the partition containing part_func_value */ |
| 3545 | while (max_part_id > min_part_id) |
| 3546 | { |
| 3547 | loc_part_id= (max_part_id + min_part_id) / 2; |
| 3548 | if (range_array[loc_part_id] <= part_func_value) |
| 3549 | min_part_id= loc_part_id + 1; |
| 3550 | else |
| 3551 | max_part_id= loc_part_id; |
| 3552 | } |
| 3553 | loc_part_id= max_part_id; |
| 3554 | *part_id= (uint32)loc_part_id; |
| 3555 | if (loc_part_id == max_partition && |
| 3556 | part_func_value >= range_array[loc_part_id] && |
| 3557 | !part_info->defined_max_value) |
| 3558 | DBUG_RETURN(HA_ERR_NO_PARTITION_FOUND); |
| 3559 | |
| 3560 | DBUG_PRINT("exit" ,("partition: %d" , *part_id)); |
| 3561 | DBUG_RETURN(0); |
| 3562 | } |
| 3563 | |
| 3564 | |
| 3565 | /* |
| 3566 | Find the sub-array of part_info->range_int_array that covers given interval |
| 3567 | |
| 3568 | SYNOPSIS |
| 3569 | get_partition_id_range_for_endpoint() |
| 3570 | part_info Partitioning info (partitioning type must be RANGE) |
| 3571 | left_endpoint TRUE - the interval is [a; +inf) or (a; +inf) |
| 3572 | FALSE - the interval is (-inf; a] or (-inf; a). |
| 3573 | include_endpoint TRUE <=> the endpoint itself is included in the |
| 3574 | interval |
| 3575 | |
| 3576 | DESCRIPTION |
| 3577 | This function finds the sub-array of part_info->range_int_array where the |
| 3578 | elements have non-empty intersections with the given interval. |
| 3579 | |
| 3580 | A range_int_array element at index idx represents the interval |
| 3581 | |
| 3582 | [range_int_array[idx-1], range_int_array[idx]), |
| 3583 | |
| 3584 | intervals are disjoint and ordered by their right bound, so |
| 3585 | |
| 3586 | 1. For [a; +inf) or (a; +inf)-type intervals (left_endpoint==TRUE), the |
| 3587 | sought sub-array starts at some index idx and continues till array end. |
| 3588 | The function returns first number idx, such that the interval |
| 3589 | represented by range_int_array[idx] has non empty intersection with |
| 3590 | the passed interval. |
| 3591 | |
| 3592 | 2. For (-inf; a] or (-inf; a)-type intervals (left_endpoint==FALSE), the |
| 3593 | sought sub-array starts at array start and continues till some last |
| 3594 | index idx. |
| 3595 | The function returns first number idx, such that the interval |
| 3596 | represented by range_int_array[idx] has EMPTY intersection with the |
| 3597 | passed interval. |
| 3598 | If the interval represented by the last array element has non-empty |
| 3599 | intersection with the passed interval, part_info->num_parts is |
| 3600 | returned. |
| 3601 | |
| 3602 | RETURN |
| 3603 | The edge of corresponding part_info->range_int_array sub-array. |
| 3604 | */ |
| 3605 | |
| 3606 | static uint32 |
| 3607 | get_partition_id_range_for_endpoint_charset(partition_info *part_info, |
| 3608 | bool left_endpoint, |
| 3609 | bool include_endpoint) |
| 3610 | { |
| 3611 | uint32 res; |
| 3612 | copy_to_part_field_buffers(part_info->part_field_array, |
| 3613 | part_info->part_field_buffers, |
| 3614 | part_info->restore_part_field_ptrs); |
| 3615 | res= get_partition_id_range_for_endpoint(part_info, left_endpoint, |
| 3616 | include_endpoint); |
| 3617 | restore_part_field_pointers(part_info->part_field_array, |
| 3618 | part_info->restore_part_field_ptrs); |
| 3619 | return res; |
| 3620 | } |
| 3621 | |
| 3622 | uint32 get_partition_id_range_for_endpoint(partition_info *part_info, |
| 3623 | bool left_endpoint, |
| 3624 | bool include_endpoint) |
| 3625 | { |
| 3626 | longlong *range_array= part_info->range_int_array; |
| 3627 | longlong part_end_val; |
| 3628 | uint max_partition= part_info->num_parts - 1; |
| 3629 | uint min_part_id= 0, max_part_id= max_partition, loc_part_id; |
| 3630 | /* Get the partitioning function value for the endpoint */ |
| 3631 | longlong part_func_value= |
| 3632 | part_info->part_expr->val_int_endpoint(left_endpoint, &include_endpoint); |
| 3633 | |
| 3634 | bool unsigned_flag= part_info->part_expr->unsigned_flag; |
| 3635 | DBUG_ENTER("get_partition_id_range_for_endpoint" ); |
| 3636 | |
| 3637 | if (part_info->part_expr->null_value) |
| 3638 | { |
| 3639 | /* |
| 3640 | Special handling for MONOTONIC functions that can return NULL for |
| 3641 | values that are comparable. I.e. |
| 3642 | '2000-00-00' can be compared to '2000-01-01' but TO_DAYS('2000-00-00') |
| 3643 | returns NULL which cannot be compared used <, >, <=, >= etc. |
| 3644 | |
| 3645 | Otherwise, just return the first partition |
| 3646 | (may be included if not left endpoint) |
| 3647 | */ |
| 3648 | enum_monotonicity_info monotonic; |
| 3649 | monotonic= part_info->part_expr->get_monotonicity_info(); |
| 3650 | if (monotonic != MONOTONIC_INCREASING_NOT_NULL && |
| 3651 | monotonic != MONOTONIC_STRICT_INCREASING_NOT_NULL) |
| 3652 | { |
| 3653 | /* F(col) can not return NULL, return partition with lowest value */ |
| 3654 | if (!left_endpoint && include_endpoint) |
| 3655 | DBUG_RETURN(1); |
| 3656 | DBUG_RETURN(0); |
| 3657 | |
| 3658 | } |
| 3659 | } |
| 3660 | |
| 3661 | if (unsigned_flag) |
| 3662 | part_func_value-= 0x8000000000000000ULL; |
| 3663 | if (left_endpoint && !include_endpoint) |
| 3664 | part_func_value++; |
| 3665 | |
| 3666 | /* |
| 3667 | Search for the partition containing part_func_value |
| 3668 | (including the right endpoint). |
| 3669 | */ |
| 3670 | while (max_part_id > min_part_id) |
| 3671 | { |
| 3672 | loc_part_id= (max_part_id + min_part_id) / 2; |
| 3673 | if (range_array[loc_part_id] < part_func_value) |
| 3674 | min_part_id= loc_part_id + 1; |
| 3675 | else |
| 3676 | max_part_id= loc_part_id; |
| 3677 | } |
| 3678 | loc_part_id= max_part_id; |
| 3679 | |
| 3680 | /* Adjust for endpoints */ |
| 3681 | part_end_val= range_array[loc_part_id]; |
| 3682 | if (left_endpoint) |
| 3683 | { |
| 3684 | DBUG_ASSERT(part_func_value > part_end_val ? |
| 3685 | (loc_part_id == max_partition && |
| 3686 | !part_info->defined_max_value) : |
| 3687 | 1); |
| 3688 | /* |
| 3689 | In case of PARTITION p VALUES LESS THAN MAXVALUE |
| 3690 | the maximum value is in the current (last) partition. |
| 3691 | If value is equal or greater than the endpoint, |
| 3692 | the range starts from the next partition. |
| 3693 | */ |
| 3694 | if (part_func_value >= part_end_val && |
| 3695 | (loc_part_id < max_partition || !part_info->defined_max_value)) |
| 3696 | loc_part_id++; |
| 3697 | } |
| 3698 | else |
| 3699 | { |
| 3700 | /* if 'WHERE <= X' and partition is LESS THAN (X) include next partition */ |
| 3701 | if (include_endpoint && loc_part_id < max_partition && |
| 3702 | part_func_value == part_end_val) |
| 3703 | loc_part_id++; |
| 3704 | |
| 3705 | /* Right endpoint, set end after correct partition */ |
| 3706 | loc_part_id++; |
| 3707 | } |
| 3708 | DBUG_RETURN(loc_part_id); |
| 3709 | } |
| 3710 | |
| 3711 | |
| 3712 | int get_partition_id_hash_nosub(partition_info *part_info, |
| 3713 | uint32 *part_id, |
| 3714 | longlong *func_value) |
| 3715 | { |
| 3716 | return get_part_id_hash(part_info->num_parts, part_info->part_expr, |
| 3717 | part_id, func_value); |
| 3718 | } |
| 3719 | |
| 3720 | |
| 3721 | int get_partition_id_linear_hash_nosub(partition_info *part_info, |
| 3722 | uint32 *part_id, |
| 3723 | longlong *func_value) |
| 3724 | { |
| 3725 | return get_part_id_linear_hash(part_info, part_info->num_parts, |
| 3726 | part_info->part_expr, part_id, func_value); |
| 3727 | } |
| 3728 | |
| 3729 | |
| 3730 | int get_partition_id_key_nosub(partition_info *part_info, |
| 3731 | uint32 *part_id, |
| 3732 | longlong *func_value) |
| 3733 | { |
| 3734 | *part_id= get_part_id_key(part_info->table->file, |
| 3735 | part_info->part_field_array, |
| 3736 | part_info->num_parts, func_value); |
| 3737 | return 0; |
| 3738 | } |
| 3739 | |
| 3740 | |
| 3741 | int get_partition_id_linear_key_nosub(partition_info *part_info, |
| 3742 | uint32 *part_id, |
| 3743 | longlong *func_value) |
| 3744 | { |
| 3745 | *part_id= get_part_id_linear_key(part_info, |
| 3746 | part_info->part_field_array, |
| 3747 | part_info->num_parts, func_value); |
| 3748 | return 0; |
| 3749 | } |
| 3750 | |
| 3751 | |
| 3752 | int get_partition_id_with_sub(partition_info *part_info, |
| 3753 | uint32 *part_id, |
| 3754 | longlong *func_value) |
| 3755 | { |
| 3756 | uint32 loc_part_id, sub_part_id; |
| 3757 | uint num_subparts; |
| 3758 | int error; |
| 3759 | DBUG_ENTER("get_partition_id_with_sub" ); |
| 3760 | |
| 3761 | if (unlikely((error= part_info->get_part_partition_id(part_info, |
| 3762 | &loc_part_id, |
| 3763 | func_value)))) |
| 3764 | { |
| 3765 | DBUG_RETURN(error); |
| 3766 | } |
| 3767 | num_subparts= part_info->num_subparts; |
| 3768 | if (unlikely((error= part_info->get_subpartition_id(part_info, |
| 3769 | &sub_part_id)))) |
| 3770 | { |
| 3771 | DBUG_RETURN(error); |
| 3772 | } |
| 3773 | *part_id= get_part_id_for_sub(loc_part_id, sub_part_id, num_subparts); |
| 3774 | DBUG_RETURN(0); |
| 3775 | } |
| 3776 | |
| 3777 | |
| 3778 | /* |
| 3779 | This function is used to calculate the subpartition id |
| 3780 | |
| 3781 | SYNOPSIS |
| 3782 | get_subpartition_id() |
| 3783 | part_info A reference to the partition_info struct where all the |
| 3784 | desired information is given |
| 3785 | |
| 3786 | RETURN VALUE |
| 3787 | part_id The subpartition identity |
| 3788 | |
| 3789 | DESCRIPTION |
| 3790 | A routine used in some SELECT's when only partial knowledge of the |
| 3791 | partitions is known. |
| 3792 | |
| 3793 | It is actually 4 different variants of this function which are called |
| 3794 | through a function pointer. |
| 3795 | |
| 3796 | get_partition_id_hash_sub |
| 3797 | get_partition_id_key_sub |
| 3798 | get_partition_id_linear_hash_sub |
| 3799 | get_partition_id_linear_key_sub |
| 3800 | */ |
| 3801 | |
| 3802 | int get_partition_id_hash_sub(partition_info *part_info, |
| 3803 | uint32 *part_id) |
| 3804 | { |
| 3805 | longlong func_value; |
| 3806 | return get_part_id_hash(part_info->num_subparts, part_info->subpart_expr, |
| 3807 | part_id, &func_value); |
| 3808 | } |
| 3809 | |
| 3810 | |
| 3811 | int get_partition_id_linear_hash_sub(partition_info *part_info, |
| 3812 | uint32 *part_id) |
| 3813 | { |
| 3814 | longlong func_value; |
| 3815 | return get_part_id_linear_hash(part_info, part_info->num_subparts, |
| 3816 | part_info->subpart_expr, part_id, |
| 3817 | &func_value); |
| 3818 | } |
| 3819 | |
| 3820 | |
| 3821 | int get_partition_id_key_sub(partition_info *part_info, |
| 3822 | uint32 *part_id) |
| 3823 | { |
| 3824 | longlong func_value; |
| 3825 | *part_id= get_part_id_key(part_info->table->file, |
| 3826 | part_info->subpart_field_array, |
| 3827 | part_info->num_subparts, &func_value); |
| 3828 | return FALSE; |
| 3829 | } |
| 3830 | |
| 3831 | |
| 3832 | int get_partition_id_linear_key_sub(partition_info *part_info, |
| 3833 | uint32 *part_id) |
| 3834 | { |
| 3835 | longlong func_value; |
| 3836 | *part_id= get_part_id_linear_key(part_info, |
| 3837 | part_info->subpart_field_array, |
| 3838 | part_info->num_subparts, &func_value); |
| 3839 | return FALSE; |
| 3840 | } |
| 3841 | |
| 3842 | |
| 3843 | /* |
| 3844 | Set an indicator on all partition fields that are set by the key |
| 3845 | |
| 3846 | SYNOPSIS |
| 3847 | set_PF_fields_in_key() |
| 3848 | key_info Information about the index |
| 3849 | key_length Length of key |
| 3850 | |
| 3851 | RETURN VALUE |
| 3852 | TRUE Found partition field set by key |
| 3853 | FALSE No partition field set by key |
| 3854 | */ |
| 3855 | |
| 3856 | static bool set_PF_fields_in_key(KEY *key_info, uint key_length) |
| 3857 | { |
| 3858 | KEY_PART_INFO *key_part; |
| 3859 | bool found_part_field= FALSE; |
| 3860 | DBUG_ENTER("set_PF_fields_in_key" ); |
| 3861 | |
| 3862 | for (key_part= key_info->key_part; (int)key_length > 0; key_part++) |
| 3863 | { |
| 3864 | if (key_part->null_bit) |
| 3865 | key_length--; |
| 3866 | if (key_part->type == HA_KEYTYPE_BIT) |
| 3867 | { |
| 3868 | if (((Field_bit*)key_part->field)->bit_len) |
| 3869 | key_length--; |
| 3870 | } |
| 3871 | if (key_part->key_part_flag & (HA_BLOB_PART + HA_VAR_LENGTH_PART)) |
| 3872 | { |
| 3873 | key_length-= HA_KEY_BLOB_LENGTH; |
| 3874 | } |
| 3875 | if (key_length < key_part->length) |
| 3876 | break; |
| 3877 | key_length-= key_part->length; |
| 3878 | if (key_part->field->flags & FIELD_IN_PART_FUNC_FLAG) |
| 3879 | { |
| 3880 | found_part_field= TRUE; |
| 3881 | key_part->field->flags|= GET_FIXED_FIELDS_FLAG; |
| 3882 | } |
| 3883 | } |
| 3884 | DBUG_RETURN(found_part_field); |
| 3885 | } |
| 3886 | |
| 3887 | |
| 3888 | /* |
| 3889 | We have found that at least one partition field was set by a key, now |
| 3890 | check if a partition function has all its fields bound or not. |
| 3891 | |
| 3892 | SYNOPSIS |
| 3893 | check_part_func_bound() |
| 3894 | ptr Array of fields NULL terminated (partition fields) |
| 3895 | |
| 3896 | RETURN VALUE |
| 3897 | TRUE All fields in partition function are set |
| 3898 | FALSE Not all fields in partition function are set |
| 3899 | */ |
| 3900 | |
| 3901 | static bool check_part_func_bound(Field **ptr) |
| 3902 | { |
| 3903 | bool result= TRUE; |
| 3904 | DBUG_ENTER("check_part_func_bound" ); |
| 3905 | |
| 3906 | for (; *ptr; ptr++) |
| 3907 | { |
| 3908 | if (!((*ptr)->flags & GET_FIXED_FIELDS_FLAG)) |
| 3909 | { |
| 3910 | result= FALSE; |
| 3911 | break; |
| 3912 | } |
| 3913 | } |
| 3914 | DBUG_RETURN(result); |
| 3915 | } |
| 3916 | |
| 3917 | |
| 3918 | /* |
| 3919 | Get the id of the subpartitioning part by using the key buffer of the |
| 3920 | index scan. |
| 3921 | |
| 3922 | SYNOPSIS |
| 3923 | get_sub_part_id_from_key() |
| 3924 | table The table object |
| 3925 | buf A buffer that can be used to evaluate the partition function |
| 3926 | key_info The index object |
| 3927 | key_spec A key_range containing key and key length |
| 3928 | out:part_id The returned partition id |
| 3929 | |
| 3930 | RETURN VALUES |
| 3931 | TRUE All fields in partition function are set |
| 3932 | FALSE Not all fields in partition function are set |
| 3933 | |
| 3934 | DESCRIPTION |
| 3935 | Use key buffer to set-up record in buf, move field pointers and |
| 3936 | get the partition identity and restore field pointers afterwards. |
| 3937 | */ |
| 3938 | |
| 3939 | static int get_sub_part_id_from_key(const TABLE *table,uchar *buf, |
| 3940 | KEY *key_info, |
| 3941 | const key_range *key_spec, |
| 3942 | uint32 *part_id) |
| 3943 | { |
| 3944 | uchar *rec0= table->record[0]; |
| 3945 | partition_info *part_info= table->part_info; |
| 3946 | int res; |
| 3947 | DBUG_ENTER("get_sub_part_id_from_key" ); |
| 3948 | |
| 3949 | key_restore(buf, (uchar*)key_spec->key, key_info, key_spec->length); |
| 3950 | if (likely(rec0 == buf)) |
| 3951 | { |
| 3952 | res= part_info->get_subpartition_id(part_info, part_id); |
| 3953 | } |
| 3954 | else |
| 3955 | { |
| 3956 | Field **part_field_array= part_info->subpart_field_array; |
| 3957 | part_info->table->move_fields(part_field_array, buf, rec0); |
| 3958 | res= part_info->get_subpartition_id(part_info, part_id); |
| 3959 | part_info->table->move_fields(part_field_array, rec0, buf); |
| 3960 | } |
| 3961 | DBUG_RETURN(res); |
| 3962 | } |
| 3963 | |
| 3964 | /* |
| 3965 | Get the id of the partitioning part by using the key buffer of the |
| 3966 | index scan. |
| 3967 | |
| 3968 | SYNOPSIS |
| 3969 | get_part_id_from_key() |
| 3970 | table The table object |
| 3971 | buf A buffer that can be used to evaluate the partition function |
| 3972 | key_info The index object |
| 3973 | key_spec A key_range containing key and key length |
| 3974 | out:part_id Partition to use |
| 3975 | |
| 3976 | RETURN VALUES |
| 3977 | TRUE Partition to use not found |
| 3978 | FALSE Ok, part_id indicates partition to use |
| 3979 | |
| 3980 | DESCRIPTION |
| 3981 | Use key buffer to set-up record in buf, move field pointers and |
| 3982 | get the partition identity and restore field pointers afterwards. |
| 3983 | */ |
| 3984 | |
| 3985 | bool get_part_id_from_key(const TABLE *table, uchar *buf, KEY *key_info, |
| 3986 | const key_range *key_spec, uint32 *part_id) |
| 3987 | { |
| 3988 | bool result; |
| 3989 | uchar *rec0= table->record[0]; |
| 3990 | partition_info *part_info= table->part_info; |
| 3991 | longlong func_value; |
| 3992 | DBUG_ENTER("get_part_id_from_key" ); |
| 3993 | |
| 3994 | key_restore(buf, (uchar*)key_spec->key, key_info, key_spec->length); |
| 3995 | if (likely(rec0 == buf)) |
| 3996 | { |
| 3997 | result= part_info->get_part_partition_id(part_info, part_id, |
| 3998 | &func_value); |
| 3999 | } |
| 4000 | else |
| 4001 | { |
| 4002 | Field **part_field_array= part_info->part_field_array; |
| 4003 | part_info->table->move_fields(part_field_array, buf, rec0); |
| 4004 | result= part_info->get_part_partition_id(part_info, part_id, |
| 4005 | &func_value); |
| 4006 | part_info->table->move_fields(part_field_array, rec0, buf); |
| 4007 | } |
| 4008 | DBUG_RETURN(result); |
| 4009 | } |
| 4010 | |
| 4011 | /* |
| 4012 | Get the partitioning id of the full PF by using the key buffer of the |
| 4013 | index scan. |
| 4014 | |
| 4015 | SYNOPSIS |
| 4016 | get_full_part_id_from_key() |
| 4017 | table The table object |
| 4018 | buf A buffer that is used to evaluate the partition function |
| 4019 | key_info The index object |
| 4020 | key_spec A key_range containing key and key length |
| 4021 | out:part_spec A partition id containing start part and end part |
| 4022 | |
| 4023 | RETURN VALUES |
| 4024 | part_spec |
| 4025 | No partitions to scan is indicated by end_part > start_part when returning |
| 4026 | |
| 4027 | DESCRIPTION |
| 4028 | Use key buffer to set-up record in buf, move field pointers if needed and |
| 4029 | get the partition identity and restore field pointers afterwards. |
| 4030 | */ |
| 4031 | |
| 4032 | void get_full_part_id_from_key(const TABLE *table, uchar *buf, |
| 4033 | KEY *key_info, |
| 4034 | const key_range *key_spec, |
| 4035 | part_id_range *part_spec) |
| 4036 | { |
| 4037 | bool result; |
| 4038 | partition_info *part_info= table->part_info; |
| 4039 | uchar *rec0= table->record[0]; |
| 4040 | longlong func_value; |
| 4041 | DBUG_ENTER("get_full_part_id_from_key" ); |
| 4042 | |
| 4043 | key_restore(buf, (uchar*)key_spec->key, key_info, key_spec->length); |
| 4044 | if (likely(rec0 == buf)) |
| 4045 | { |
| 4046 | result= part_info->get_partition_id(part_info, &part_spec->start_part, |
| 4047 | &func_value); |
| 4048 | } |
| 4049 | else |
| 4050 | { |
| 4051 | Field **part_field_array= part_info->full_part_field_array; |
| 4052 | part_info->table->move_fields(part_field_array, buf, rec0); |
| 4053 | result= part_info->get_partition_id(part_info, &part_spec->start_part, |
| 4054 | &func_value); |
| 4055 | part_info->table->move_fields(part_field_array, rec0, buf); |
| 4056 | } |
| 4057 | part_spec->end_part= part_spec->start_part; |
| 4058 | if (unlikely(result)) |
| 4059 | part_spec->start_part++; |
| 4060 | DBUG_VOID_RETURN; |
| 4061 | } |
| 4062 | |
| 4063 | |
| 4064 | /** |
| 4065 | @brief Verify that all rows in a table is in the given partition |
| 4066 | |
| 4067 | @param table Table which contains the data that will be checked if |
| 4068 | it is matching the partition definition. |
| 4069 | @param part_table Partitioned table containing the partition to check. |
| 4070 | @param part_id Which partition to match with. |
| 4071 | |
| 4072 | @return Operation status |
| 4073 | @retval TRUE Not all rows match the given partition |
| 4074 | @retval FALSE OK |
| 4075 | */ |
| 4076 | bool verify_data_with_partition(TABLE *table, TABLE *part_table, |
| 4077 | uint32 part_id) |
| 4078 | { |
| 4079 | uint32 found_part_id; |
| 4080 | longlong func_value; /* Unused */ |
| 4081 | handler *file; |
| 4082 | int error; |
| 4083 | uchar *old_rec; |
| 4084 | partition_info *part_info; |
| 4085 | DBUG_ENTER("verify_data_with_partition" ); |
| 4086 | DBUG_ASSERT(table && table->file && part_table && part_table->part_info && |
| 4087 | part_table->file); |
| 4088 | |
| 4089 | /* |
| 4090 | Verify all table rows. |
| 4091 | First implementation uses full scan + evaluates partition functions for |
| 4092 | every row. TODO: add optimization to use index if possible, see WL#5397. |
| 4093 | |
| 4094 | 1) Open both tables (already done) and set the row buffers to use |
| 4095 | the same buffer (to avoid copy). |
| 4096 | 2) Init rnd on table. |
| 4097 | 3) loop over all rows. |
| 4098 | 3.1) verify that partition_id on the row is correct. Break if error. |
| 4099 | */ |
| 4100 | file= table->file; |
| 4101 | part_info= part_table->part_info; |
| 4102 | bitmap_union(table->read_set, &part_info->full_part_field_set); |
| 4103 | old_rec= part_table->record[0]; |
| 4104 | part_table->record[0]= table->record[0]; |
| 4105 | part_info->table->move_fields(part_info->full_part_field_array, table->record[0], old_rec); |
| 4106 | if (unlikely(error= file->ha_rnd_init_with_error(TRUE))) |
| 4107 | goto err; |
| 4108 | |
| 4109 | do |
| 4110 | { |
| 4111 | if (unlikely((error= file->ha_rnd_next(table->record[0])))) |
| 4112 | { |
| 4113 | if (error == HA_ERR_END_OF_FILE) |
| 4114 | error= 0; |
| 4115 | else |
| 4116 | file->print_error(error, MYF(0)); |
| 4117 | break; |
| 4118 | } |
| 4119 | if (unlikely((error= part_info->get_partition_id(part_info, &found_part_id, |
| 4120 | &func_value)))) |
| 4121 | { |
| 4122 | part_table->file->print_error(error, MYF(0)); |
| 4123 | break; |
| 4124 | } |
| 4125 | DEBUG_SYNC(current_thd, "swap_partition_first_row_read" ); |
| 4126 | if (found_part_id != part_id) |
| 4127 | { |
| 4128 | my_error(ER_ROW_DOES_NOT_MATCH_PARTITION, MYF(0)); |
| 4129 | error= 1; |
| 4130 | break; |
| 4131 | } |
| 4132 | } while (TRUE); |
| 4133 | (void) file->ha_rnd_end(); |
| 4134 | err: |
| 4135 | part_info->table->move_fields(part_info->full_part_field_array, old_rec, |
| 4136 | table->record[0]); |
| 4137 | part_table->record[0]= old_rec; |
| 4138 | DBUG_RETURN(unlikely(error) ? TRUE : FALSE); |
| 4139 | } |
| 4140 | |
| 4141 | |
| 4142 | /* |
| 4143 | Prune the set of partitions to use in query |
| 4144 | |
| 4145 | SYNOPSIS |
| 4146 | prune_partition_set() |
| 4147 | table The table object |
| 4148 | out:part_spec Contains start part, end part |
| 4149 | |
| 4150 | DESCRIPTION |
| 4151 | This function is called to prune the range of partitions to scan by |
| 4152 | checking the read_partitions bitmap. |
| 4153 | If start_part > end_part at return it means no partition needs to be |
| 4154 | scanned. If start_part == end_part it always means a single partition |
| 4155 | needs to be scanned. |
| 4156 | |
| 4157 | RETURN VALUE |
| 4158 | part_spec |
| 4159 | */ |
| 4160 | void prune_partition_set(const TABLE *table, part_id_range *part_spec) |
| 4161 | { |
| 4162 | int last_partition= -1; |
| 4163 | uint i; |
| 4164 | partition_info *part_info= table->part_info; |
| 4165 | |
| 4166 | DBUG_ENTER("prune_partition_set" ); |
| 4167 | for (i= part_spec->start_part; i <= part_spec->end_part; i++) |
| 4168 | { |
| 4169 | if (bitmap_is_set(&(part_info->read_partitions), i)) |
| 4170 | { |
| 4171 | DBUG_PRINT("info" , ("Partition %d is set" , i)); |
| 4172 | if (last_partition == -1) |
| 4173 | /* First partition found in set and pruned bitmap */ |
| 4174 | part_spec->start_part= i; |
| 4175 | last_partition= i; |
| 4176 | } |
| 4177 | } |
| 4178 | if (last_partition == -1) |
| 4179 | /* No partition found in pruned bitmap */ |
| 4180 | part_spec->start_part= part_spec->end_part + 1; |
| 4181 | else //if (last_partition != -1) |
| 4182 | part_spec->end_part= last_partition; |
| 4183 | |
| 4184 | DBUG_VOID_RETURN; |
| 4185 | } |
| 4186 | |
| 4187 | /* |
| 4188 | Get the set of partitions to use in query. |
| 4189 | |
| 4190 | SYNOPSIS |
| 4191 | get_partition_set() |
| 4192 | table The table object |
| 4193 | buf A buffer that can be used to evaluate the partition function |
| 4194 | index The index of the key used, if MAX_KEY no index used |
| 4195 | key_spec A key_range containing key and key length |
| 4196 | out:part_spec Contains start part, end part and indicator if bitmap is |
| 4197 | used for which partitions to scan |
| 4198 | |
| 4199 | DESCRIPTION |
| 4200 | This function is called to discover which partitions to use in an index |
| 4201 | scan or a full table scan. |
| 4202 | It returns a range of partitions to scan. If there are holes in this |
| 4203 | range with partitions that are not needed to scan a bit array is used |
| 4204 | to signal which partitions to use and which not to use. |
| 4205 | If start_part > end_part at return it means no partition needs to be |
| 4206 | scanned. If start_part == end_part it always means a single partition |
| 4207 | needs to be scanned. |
| 4208 | |
| 4209 | RETURN VALUE |
| 4210 | part_spec |
| 4211 | */ |
| 4212 | void get_partition_set(const TABLE *table, uchar *buf, const uint index, |
| 4213 | const key_range *key_spec, part_id_range *part_spec) |
| 4214 | { |
| 4215 | partition_info *part_info= table->part_info; |
| 4216 | uint num_parts= part_info->get_tot_partitions(); |
| 4217 | uint i, part_id; |
| 4218 | uint sub_part= num_parts; |
| 4219 | uint32 part_part= num_parts; |
| 4220 | KEY *key_info= NULL; |
| 4221 | bool found_part_field= FALSE; |
| 4222 | DBUG_ENTER("get_partition_set" ); |
| 4223 | |
| 4224 | part_spec->start_part= 0; |
| 4225 | part_spec->end_part= num_parts - 1; |
| 4226 | if ((index < MAX_KEY) && |
| 4227 | key_spec && key_spec->flag == (uint)HA_READ_KEY_EXACT && |
| 4228 | part_info->some_fields_in_PF.is_set(index)) |
| 4229 | { |
| 4230 | key_info= table->key_info+index; |
| 4231 | /* |
| 4232 | The index can potentially provide at least one PF-field (field in the |
| 4233 | partition function). Thus it is interesting to continue our probe. |
| 4234 | */ |
| 4235 | if (key_spec->length == key_info->key_length) |
| 4236 | { |
| 4237 | /* |
| 4238 | The entire key is set so we can check whether we can immediately |
| 4239 | derive either the complete PF or if we can derive either |
| 4240 | the top PF or the subpartitioning PF. This can be established by |
| 4241 | checking precalculated bits on each index. |
| 4242 | */ |
| 4243 | if (part_info->all_fields_in_PF.is_set(index)) |
| 4244 | { |
| 4245 | /* |
| 4246 | We can derive the exact partition to use, no more than this one |
| 4247 | is needed. |
| 4248 | */ |
| 4249 | get_full_part_id_from_key(table,buf,key_info,key_spec,part_spec); |
| 4250 | /* |
| 4251 | Check if range can be adjusted by looking in read_partitions |
| 4252 | */ |
| 4253 | prune_partition_set(table, part_spec); |
| 4254 | DBUG_VOID_RETURN; |
| 4255 | } |
| 4256 | else if (part_info->is_sub_partitioned()) |
| 4257 | { |
| 4258 | if (part_info->all_fields_in_SPF.is_set(index)) |
| 4259 | { |
| 4260 | if (get_sub_part_id_from_key(table, buf, key_info, key_spec, &sub_part)) |
| 4261 | { |
| 4262 | part_spec->start_part= num_parts; |
| 4263 | DBUG_VOID_RETURN; |
| 4264 | } |
| 4265 | } |
| 4266 | else if (part_info->all_fields_in_PPF.is_set(index)) |
| 4267 | { |
| 4268 | if (get_part_id_from_key(table,buf,key_info, |
| 4269 | key_spec,(uint32*)&part_part)) |
| 4270 | { |
| 4271 | /* |
| 4272 | The value of the RANGE or LIST partitioning was outside of |
| 4273 | allowed values. Thus it is certain that the result of this |
| 4274 | scan will be empty. |
| 4275 | */ |
| 4276 | part_spec->start_part= num_parts; |
| 4277 | DBUG_VOID_RETURN; |
| 4278 | } |
| 4279 | } |
| 4280 | } |
| 4281 | } |
| 4282 | else |
| 4283 | { |
| 4284 | /* |
| 4285 | Set an indicator on all partition fields that are bound. |
| 4286 | If at least one PF-field was bound it pays off to check whether |
| 4287 | the PF or PPF or SPF has been bound. |
| 4288 | (PF = Partition Function, SPF = Subpartition Function and |
| 4289 | PPF = Partition Function part of subpartitioning) |
| 4290 | */ |
| 4291 | if ((found_part_field= set_PF_fields_in_key(key_info, |
| 4292 | key_spec->length))) |
| 4293 | { |
| 4294 | if (check_part_func_bound(part_info->full_part_field_array)) |
| 4295 | { |
| 4296 | /* |
| 4297 | We were able to bind all fields in the partition function even |
| 4298 | by using only a part of the key. Calculate the partition to use. |
| 4299 | */ |
| 4300 | get_full_part_id_from_key(table,buf,key_info,key_spec,part_spec); |
| 4301 | clear_indicator_in_key_fields(key_info); |
| 4302 | /* |
| 4303 | Check if range can be adjusted by looking in read_partitions |
| 4304 | */ |
| 4305 | prune_partition_set(table, part_spec); |
| 4306 | DBUG_VOID_RETURN; |
| 4307 | } |
| 4308 | else if (part_info->is_sub_partitioned()) |
| 4309 | { |
| 4310 | if (check_part_func_bound(part_info->subpart_field_array)) |
| 4311 | { |
| 4312 | if (get_sub_part_id_from_key(table, buf, key_info, key_spec, &sub_part)) |
| 4313 | { |
| 4314 | part_spec->start_part= num_parts; |
| 4315 | clear_indicator_in_key_fields(key_info); |
| 4316 | DBUG_VOID_RETURN; |
| 4317 | } |
| 4318 | } |
| 4319 | else if (check_part_func_bound(part_info->part_field_array)) |
| 4320 | { |
| 4321 | if (get_part_id_from_key(table,buf,key_info,key_spec,&part_part)) |
| 4322 | { |
| 4323 | part_spec->start_part= num_parts; |
| 4324 | clear_indicator_in_key_fields(key_info); |
| 4325 | DBUG_VOID_RETURN; |
| 4326 | } |
| 4327 | } |
| 4328 | } |
| 4329 | } |
| 4330 | } |
| 4331 | } |
| 4332 | { |
| 4333 | /* |
| 4334 | The next step is to analyse the table condition to see whether any |
| 4335 | information about which partitions to scan can be derived from there. |
| 4336 | Currently not implemented. |
| 4337 | */ |
| 4338 | } |
| 4339 | /* |
| 4340 | If we come here we have found a range of sorts we have either discovered |
| 4341 | nothing or we have discovered a range of partitions with possible holes |
| 4342 | in it. We need a bitvector to further the work here. |
| 4343 | */ |
| 4344 | if (!(part_part == num_parts && sub_part == num_parts)) |
| 4345 | { |
| 4346 | /* |
| 4347 | We can only arrive here if we are using subpartitioning. |
| 4348 | */ |
| 4349 | if (part_part != num_parts) |
| 4350 | { |
| 4351 | /* |
| 4352 | We know the top partition and need to scan all underlying |
| 4353 | subpartitions. This is a range without holes. |
| 4354 | */ |
| 4355 | DBUG_ASSERT(sub_part == num_parts); |
| 4356 | part_spec->start_part= part_part * part_info->num_subparts; |
| 4357 | part_spec->end_part= part_spec->start_part+part_info->num_subparts - 1; |
| 4358 | } |
| 4359 | else |
| 4360 | { |
| 4361 | DBUG_ASSERT(sub_part != num_parts); |
| 4362 | part_spec->start_part= sub_part; |
| 4363 | part_spec->end_part=sub_part+ |
| 4364 | (part_info->num_subparts*(part_info->num_parts-1)); |
| 4365 | for (i= 0, part_id= sub_part; i < part_info->num_parts; |
| 4366 | i++, part_id+= part_info->num_subparts) |
| 4367 | ; //Set bit part_id in bit array |
| 4368 | } |
| 4369 | } |
| 4370 | if (found_part_field) |
| 4371 | clear_indicator_in_key_fields(key_info); |
| 4372 | /* |
| 4373 | Check if range can be adjusted by looking in read_partitions |
| 4374 | */ |
| 4375 | prune_partition_set(table, part_spec); |
| 4376 | DBUG_VOID_RETURN; |
| 4377 | } |
| 4378 | |
| 4379 | /* |
| 4380 | If the table is partitioned we will read the partition info into the |
| 4381 | .frm file here. |
| 4382 | ------------------------------- |
| 4383 | | Fileinfo 64 bytes | |
| 4384 | ------------------------------- |
| 4385 | | Formnames 7 bytes | |
| 4386 | ------------------------------- |
| 4387 | | Not used 4021 bytes | |
| 4388 | ------------------------------- |
| 4389 | | Keyinfo + record | |
| 4390 | ------------------------------- |
| 4391 | | Padded to next multiple | |
| 4392 | | of IO_SIZE | |
| 4393 | ------------------------------- |
| 4394 | | Forminfo 288 bytes | |
| 4395 | ------------------------------- |
| 4396 | | Screen buffer, to make | |
| 4397 | |field names readable | |
| 4398 | ------------------------------- |
| 4399 | | Packed field info | |
| 4400 | |17 + 1 + strlen(field_name) | |
| 4401 | | + 1 end of file character | |
| 4402 | ------------------------------- |
| 4403 | | Partition info | |
| 4404 | ------------------------------- |
| 4405 | We provide the length of partition length in Fileinfo[55-58]. |
| 4406 | |
| 4407 | Read the partition syntax from the frm file and parse it to get the |
| 4408 | data structures of the partitioning. |
| 4409 | |
| 4410 | SYNOPSIS |
| 4411 | mysql_unpack_partition() |
| 4412 | thd Thread object |
| 4413 | part_buf Partition info from frm file |
| 4414 | part_info_len Length of partition syntax |
| 4415 | table Table object of partitioned table |
| 4416 | create_table_ind Is it called from CREATE TABLE |
| 4417 | default_db_type What is the default engine of the table |
| 4418 | work_part_info_used Flag is raised if we don't create new |
| 4419 | part_info, but used thd->work_part_info |
| 4420 | |
| 4421 | RETURN VALUE |
| 4422 | TRUE Error |
| 4423 | FALSE Sucess |
| 4424 | |
| 4425 | DESCRIPTION |
| 4426 | Read the partition syntax from the current position in the frm file. |
| 4427 | Initiate a LEX object, save the list of item tree objects to free after |
| 4428 | the query is done. Set-up partition info object such that parser knows |
| 4429 | it is called from internally. Call parser to create data structures |
| 4430 | (best possible recreation of item trees and so forth since there is no |
| 4431 | serialisation of these objects other than in parseable text format). |
| 4432 | We need to save the text of the partition functions since it is not |
| 4433 | possible to retrace this given an item tree. |
| 4434 | */ |
| 4435 | |
| 4436 | bool mysql_unpack_partition(THD *thd, |
| 4437 | char *part_buf, uint part_info_len, |
| 4438 | TABLE* table, bool is_create_table_ind, |
| 4439 | handlerton *default_db_type, |
| 4440 | bool *work_part_info_used) |
| 4441 | { |
| 4442 | bool result= TRUE; |
| 4443 | partition_info *part_info; |
| 4444 | CHARSET_INFO *old_character_set_client= thd->variables.character_set_client; |
| 4445 | LEX *old_lex= thd->lex; |
| 4446 | LEX lex; |
| 4447 | PSI_statement_locker *parent_locker= thd->m_statement_psi; |
| 4448 | DBUG_ENTER("mysql_unpack_partition" ); |
| 4449 | |
| 4450 | thd->variables.character_set_client= system_charset_info; |
| 4451 | |
| 4452 | Parser_state parser_state; |
| 4453 | if (unlikely(parser_state.init(thd, part_buf, part_info_len))) |
| 4454 | goto end; |
| 4455 | |
| 4456 | if (unlikely(init_lex_with_single_table(thd, table, &lex))) |
| 4457 | goto end; |
| 4458 | |
| 4459 | *work_part_info_used= FALSE; |
| 4460 | |
| 4461 | if (unlikely(!(lex.part_info= new partition_info()))) |
| 4462 | goto end; |
| 4463 | |
| 4464 | lex.part_info->table= table; /* Indicates MYSQLparse from this place */ |
| 4465 | part_info= lex.part_info; |
| 4466 | DBUG_PRINT("info" , ("Parse: %s" , part_buf)); |
| 4467 | |
| 4468 | thd->m_statement_psi= NULL; |
| 4469 | if (unlikely(parse_sql(thd, & parser_state, NULL)) || |
| 4470 | unlikely(part_info->fix_parser_data(thd))) |
| 4471 | { |
| 4472 | thd->free_items(); |
| 4473 | thd->m_statement_psi= parent_locker; |
| 4474 | goto end; |
| 4475 | } |
| 4476 | thd->m_statement_psi= parent_locker; |
| 4477 | /* |
| 4478 | The parsed syntax residing in the frm file can still contain defaults. |
| 4479 | The reason is that the frm file is sometimes saved outside of this |
| 4480 | MySQL Server and used in backup and restore of clusters or partitioned |
| 4481 | tables. It is not certain that the restore will restore exactly the |
| 4482 | same default partitioning. |
| 4483 | |
| 4484 | The easiest manner of handling this is to simply continue using the |
| 4485 | part_info we already built up during mysql_create_table if we are |
| 4486 | in the process of creating a table. If the table already exists we |
| 4487 | need to discover the number of partitions for the default parts. Since |
| 4488 | the handler object hasn't been created here yet we need to postpone this |
| 4489 | to the fix_partition_func method. |
| 4490 | */ |
| 4491 | |
| 4492 | DBUG_PRINT("info" , ("Successful parse" )); |
| 4493 | DBUG_PRINT("info" , ("default engine = %s, default_db_type = %s" , |
| 4494 | ha_resolve_storage_engine_name(part_info->default_engine_type), |
| 4495 | ha_resolve_storage_engine_name(default_db_type))); |
| 4496 | if (is_create_table_ind && old_lex->sql_command == SQLCOM_CREATE_TABLE) |
| 4497 | { |
| 4498 | /* |
| 4499 | When we come here we are doing a create table. In this case we |
| 4500 | have already done some preparatory work on the old part_info |
| 4501 | object. We don't really need this new partition_info object. |
| 4502 | Thus we go back to the old partition info object. |
| 4503 | We need to free any memory objects allocated on item_free_list |
| 4504 | by the parser since we are keeping the old info from the first |
| 4505 | parser call in CREATE TABLE. |
| 4506 | |
| 4507 | This table object can not be used any more. However, since |
| 4508 | this is CREATE TABLE, we know that it will be destroyed by the |
| 4509 | caller, and rely on that. |
| 4510 | */ |
| 4511 | thd->free_items(); |
| 4512 | part_info= thd->work_part_info; |
| 4513 | *work_part_info_used= true; |
| 4514 | } |
| 4515 | table->part_info= part_info; |
| 4516 | part_info->table= table; |
| 4517 | table->file->set_part_info(part_info); |
| 4518 | if (!part_info->default_engine_type) |
| 4519 | part_info->default_engine_type= default_db_type; |
| 4520 | DBUG_ASSERT(part_info->default_engine_type == default_db_type); |
| 4521 | DBUG_ASSERT(part_info->default_engine_type->db_type != DB_TYPE_UNKNOWN); |
| 4522 | DBUG_ASSERT(part_info->default_engine_type != partition_hton); |
| 4523 | result= FALSE; |
| 4524 | end: |
| 4525 | end_lex_with_single_table(thd, table, old_lex); |
| 4526 | thd->variables.character_set_client= old_character_set_client; |
| 4527 | DBUG_RETURN(result); |
| 4528 | } |
| 4529 | |
| 4530 | |
| 4531 | /* |
| 4532 | Set engine type on all partition element objects |
| 4533 | SYNOPSIS |
| 4534 | set_engine_all_partitions() |
| 4535 | part_info Partition info |
| 4536 | engine_type Handlerton reference of engine |
| 4537 | RETURN VALUES |
| 4538 | NONE |
| 4539 | */ |
| 4540 | |
| 4541 | static |
| 4542 | void |
| 4543 | set_engine_all_partitions(partition_info *part_info, |
| 4544 | handlerton *engine_type) |
| 4545 | { |
| 4546 | uint i= 0; |
| 4547 | List_iterator<partition_element> part_it(part_info->partitions); |
| 4548 | do |
| 4549 | { |
| 4550 | partition_element *part_elem= part_it++; |
| 4551 | |
| 4552 | part_elem->engine_type= engine_type; |
| 4553 | if (part_info->is_sub_partitioned()) |
| 4554 | { |
| 4555 | List_iterator<partition_element> sub_it(part_elem->subpartitions); |
| 4556 | uint j= 0; |
| 4557 | |
| 4558 | do |
| 4559 | { |
| 4560 | partition_element *sub_elem= sub_it++; |
| 4561 | |
| 4562 | sub_elem->engine_type= engine_type; |
| 4563 | } while (++j < part_info->num_subparts); |
| 4564 | } |
| 4565 | } while (++i < part_info->num_parts); |
| 4566 | } |
| 4567 | |
| 4568 | |
| 4569 | /** |
| 4570 | Support routine to handle the successful cases for partition management. |
| 4571 | |
| 4572 | @param thd Thread object |
| 4573 | @param copied Number of records copied |
| 4574 | @param deleted Number of records deleted |
| 4575 | @param table_list Table list with the one table in it |
| 4576 | |
| 4577 | @return Operation status |
| 4578 | @retval FALSE Success |
| 4579 | @retval TRUE Failure |
| 4580 | */ |
| 4581 | |
| 4582 | static int fast_end_partition(THD *thd, ulonglong copied, |
| 4583 | ulonglong deleted, |
| 4584 | TABLE_LIST *table_list) |
| 4585 | { |
| 4586 | char tmp_name[80]; |
| 4587 | DBUG_ENTER("fast_end_partition" ); |
| 4588 | |
| 4589 | thd->proc_info="end" ; |
| 4590 | |
| 4591 | query_cache_invalidate3(thd, table_list, 0); |
| 4592 | |
| 4593 | my_snprintf(tmp_name, sizeof(tmp_name), ER_THD(thd, ER_INSERT_INFO), |
| 4594 | (ulong) (copied + deleted), |
| 4595 | (ulong) deleted, |
| 4596 | (ulong) 0); |
| 4597 | my_ok(thd, (ha_rows) (copied+deleted),0L, tmp_name); |
| 4598 | DBUG_RETURN(FALSE); |
| 4599 | } |
| 4600 | |
| 4601 | |
| 4602 | /* |
| 4603 | We need to check if engine used by all partitions can handle |
| 4604 | partitioning natively. |
| 4605 | |
| 4606 | SYNOPSIS |
| 4607 | check_native_partitioned() |
| 4608 | create_info Create info in CREATE TABLE |
| 4609 | out:ret_val Return value |
| 4610 | part_info Partition info |
| 4611 | thd Thread object |
| 4612 | |
| 4613 | RETURN VALUES |
| 4614 | Value returned in bool ret_value |
| 4615 | TRUE Native partitioning supported by engine |
| 4616 | FALSE Need to use partition handler |
| 4617 | |
| 4618 | Return value from function |
| 4619 | TRUE Error |
| 4620 | FALSE Success |
| 4621 | */ |
| 4622 | |
| 4623 | static bool check_native_partitioned(HA_CREATE_INFO *create_info,bool *ret_val, |
| 4624 | partition_info *part_info, THD *thd) |
| 4625 | { |
| 4626 | bool table_engine_set; |
| 4627 | handlerton *engine_type= part_info->default_engine_type; |
| 4628 | handlerton *old_engine_type= engine_type; |
| 4629 | DBUG_ENTER("check_native_partitioned" ); |
| 4630 | |
| 4631 | if (create_info->used_fields & HA_CREATE_USED_ENGINE) |
| 4632 | { |
| 4633 | table_engine_set= TRUE; |
| 4634 | engine_type= create_info->db_type; |
| 4635 | } |
| 4636 | else |
| 4637 | { |
| 4638 | table_engine_set= FALSE; |
| 4639 | if (thd->lex->sql_command != SQLCOM_CREATE_TABLE) |
| 4640 | { |
| 4641 | table_engine_set= TRUE; |
| 4642 | DBUG_ASSERT(engine_type && engine_type != partition_hton); |
| 4643 | } |
| 4644 | } |
| 4645 | DBUG_PRINT("info" , ("engine_type = %s, table_engine_set = %u" , |
| 4646 | ha_resolve_storage_engine_name(engine_type), |
| 4647 | table_engine_set)); |
| 4648 | if (part_info->check_engine_mix(engine_type, table_engine_set)) |
| 4649 | goto error; |
| 4650 | |
| 4651 | /* |
| 4652 | All engines are of the same type. Check if this engine supports |
| 4653 | native partitioning. |
| 4654 | */ |
| 4655 | |
| 4656 | if (!engine_type) |
| 4657 | engine_type= old_engine_type; |
| 4658 | DBUG_PRINT("info" , ("engine_type = %s" , |
| 4659 | ha_resolve_storage_engine_name(engine_type))); |
| 4660 | if (engine_type->partition_flags && |
| 4661 | (engine_type->partition_flags() & HA_CAN_PARTITION)) |
| 4662 | { |
| 4663 | create_info->db_type= engine_type; |
| 4664 | DBUG_PRINT("info" , ("Changed to native partitioning" )); |
| 4665 | *ret_val= TRUE; |
| 4666 | } |
| 4667 | DBUG_RETURN(FALSE); |
| 4668 | error: |
| 4669 | /* |
| 4670 | Mixed engines not yet supported but when supported it will need |
| 4671 | the partition handler |
| 4672 | */ |
| 4673 | my_error(ER_MIX_HANDLER_ERROR, MYF(0)); |
| 4674 | *ret_val= FALSE; |
| 4675 | DBUG_RETURN(TRUE); |
| 4676 | } |
| 4677 | |
| 4678 | |
| 4679 | /** |
| 4680 | Sets which partitions to be used in the command. |
| 4681 | |
| 4682 | @param alter_info Alter_info pointer holding partition names and flags. |
| 4683 | @param tab_part_info partition_info holding all partitions. |
| 4684 | @param part_state Which state to set for the named partitions. |
| 4685 | |
| 4686 | @return Operation status |
| 4687 | @retval false Success |
| 4688 | @retval true Failure |
| 4689 | */ |
| 4690 | |
| 4691 | bool set_part_state(Alter_info *alter_info, partition_info *tab_part_info, |
| 4692 | enum partition_state part_state) |
| 4693 | { |
| 4694 | uint part_count= 0; |
| 4695 | uint num_parts_found= 0; |
| 4696 | List_iterator<partition_element> part_it(tab_part_info->partitions); |
| 4697 | |
| 4698 | do |
| 4699 | { |
| 4700 | partition_element *part_elem= part_it++; |
| 4701 | if ((alter_info->partition_flags & ALTER_PARTITION_ALL) || |
| 4702 | (is_name_in_list(part_elem->partition_name, |
| 4703 | alter_info->partition_names))) |
| 4704 | { |
| 4705 | /* |
| 4706 | Mark the partition. |
| 4707 | I.e mark the partition as a partition to be "changed" by |
| 4708 | analyzing/optimizing/rebuilding/checking/repairing/... |
| 4709 | */ |
| 4710 | num_parts_found++; |
| 4711 | part_elem->part_state= part_state; |
| 4712 | DBUG_PRINT("info" , ("Setting part_state to %u for partition %s" , |
| 4713 | part_state, part_elem->partition_name)); |
| 4714 | } |
| 4715 | else |
| 4716 | part_elem->part_state= PART_NORMAL; |
| 4717 | } while (++part_count < tab_part_info->num_parts); |
| 4718 | |
| 4719 | if (num_parts_found != alter_info->partition_names.elements && |
| 4720 | !(alter_info->partition_flags & ALTER_PARTITION_ALL)) |
| 4721 | { |
| 4722 | /* Not all given partitions found, revert and return failure */ |
| 4723 | part_it.rewind(); |
| 4724 | part_count= 0; |
| 4725 | do |
| 4726 | { |
| 4727 | partition_element *part_elem= part_it++; |
| 4728 | part_elem->part_state= PART_NORMAL; |
| 4729 | } while (++part_count < tab_part_info->num_parts); |
| 4730 | return true; |
| 4731 | } |
| 4732 | return false; |
| 4733 | } |
| 4734 | |
| 4735 | |
| 4736 | /** |
| 4737 | @brief Check if partition is exchangable with table by checking table options |
| 4738 | |
| 4739 | @param table_create_info Table options from table. |
| 4740 | @param part_elem All the info of the partition. |
| 4741 | |
| 4742 | @retval FALSE if they are equal, otherwise TRUE. |
| 4743 | |
| 4744 | @note Any differens that would cause a change in the frm file is prohibited. |
| 4745 | Such options as data_file_name, index_file_name, min_rows, max_rows etc. are |
| 4746 | not allowed to differ. But comment is allowed to differ. |
| 4747 | */ |
| 4748 | bool compare_partition_options(HA_CREATE_INFO *table_create_info, |
| 4749 | partition_element *part_elem) |
| 4750 | { |
| 4751 | #define MAX_COMPARE_PARTITION_OPTION_ERRORS 5 |
| 4752 | const char *option_diffs[MAX_COMPARE_PARTITION_OPTION_ERRORS + 1]; |
| 4753 | int i, errors= 0; |
| 4754 | DBUG_ENTER("compare_partition_options" ); |
| 4755 | |
| 4756 | /* |
| 4757 | Note that there are not yet any engine supporting tablespace together |
| 4758 | with partitioning. TODO: when there are, add compare. |
| 4759 | */ |
| 4760 | if (part_elem->tablespace_name || table_create_info->tablespace) |
| 4761 | option_diffs[errors++]= "TABLESPACE" ; |
| 4762 | if (part_elem->part_max_rows != table_create_info->max_rows) |
| 4763 | option_diffs[errors++]= "MAX_ROWS" ; |
| 4764 | if (part_elem->part_min_rows != table_create_info->min_rows) |
| 4765 | option_diffs[errors++]= "MIN_ROWS" ; |
| 4766 | |
| 4767 | for (i= 0; i < errors; i++) |
| 4768 | my_error(ER_PARTITION_EXCHANGE_DIFFERENT_OPTION, MYF(0), |
| 4769 | option_diffs[i]); |
| 4770 | DBUG_RETURN(errors != 0); |
| 4771 | } |
| 4772 | |
| 4773 | |
| 4774 | /* |
| 4775 | Prepare for ALTER TABLE of partition structure |
| 4776 | |
| 4777 | @param[in] thd Thread object |
| 4778 | @param[in] table Table object |
| 4779 | @param[in,out] alter_info Alter information |
| 4780 | @param[in,out] create_info Create info for CREATE TABLE |
| 4781 | @param[in] alter_ctx ALTER TABLE runtime context |
| 4782 | @param[out] partition_changed Boolean indicating whether partition changed |
| 4783 | @param[out] fast_alter_table Boolean indicating if fast partition alter is |
| 4784 | possible. |
| 4785 | |
| 4786 | @return Operation status |
| 4787 | @retval TRUE Error |
| 4788 | @retval FALSE Success |
| 4789 | |
| 4790 | @note |
| 4791 | This method handles all preparations for ALTER TABLE for partitioned |
| 4792 | tables. |
| 4793 | We need to handle both partition management command such as Add Partition |
| 4794 | and others here as well as an ALTER TABLE that completely changes the |
| 4795 | partitioning and yet others that don't change anything at all. We start |
| 4796 | by checking the partition management variants and then check the general |
| 4797 | change patterns. |
| 4798 | */ |
| 4799 | |
| 4800 | uint prep_alter_part_table(THD *thd, TABLE *table, Alter_info *alter_info, |
| 4801 | HA_CREATE_INFO *create_info, |
| 4802 | Alter_table_ctx *alter_ctx, |
| 4803 | bool *partition_changed, |
| 4804 | bool *fast_alter_table) |
| 4805 | { |
| 4806 | DBUG_ENTER("prep_alter_part_table" ); |
| 4807 | |
| 4808 | /* Foreign keys on partitioned tables are not supported, waits for WL#148 */ |
| 4809 | if (table->part_info && (alter_info->flags & (ALTER_ADD_FOREIGN_KEY | |
| 4810 | ALTER_DROP_FOREIGN_KEY))) |
| 4811 | { |
| 4812 | my_error(ER_FOREIGN_KEY_ON_PARTITIONED, MYF(0)); |
| 4813 | DBUG_RETURN(TRUE); |
| 4814 | } |
| 4815 | /* Remove partitioning on a not partitioned table is not possible */ |
| 4816 | if (!table->part_info && (alter_info->partition_flags & |
| 4817 | ALTER_PARTITION_REMOVE)) |
| 4818 | { |
| 4819 | my_error(ER_PARTITION_MGMT_ON_NONPARTITIONED, MYF(0)); |
| 4820 | DBUG_RETURN(TRUE); |
| 4821 | } |
| 4822 | |
| 4823 | partition_info *alt_part_info= thd->lex->part_info; |
| 4824 | /* |
| 4825 | This variable is TRUE in very special case when we add only DEFAULT |
| 4826 | partition to the existing table |
| 4827 | */ |
| 4828 | bool only_default_value_added= |
| 4829 | (alt_part_info && |
| 4830 | alt_part_info->current_partition && |
| 4831 | alt_part_info->current_partition->list_val_list.elements == 1 && |
| 4832 | alt_part_info->current_partition->list_val_list.head()-> |
| 4833 | added_items >= 1 && |
| 4834 | alt_part_info->current_partition->list_val_list.head()-> |
| 4835 | col_val_array[0].max_value) && |
| 4836 | alt_part_info->part_type == LIST_PARTITION && |
| 4837 | (alter_info->partition_flags & ALTER_PARTITION_ADD); |
| 4838 | if (only_default_value_added && |
| 4839 | !thd->lex->part_info->num_columns) |
| 4840 | thd->lex->part_info->num_columns= 1; // to make correct clone |
| 4841 | |
| 4842 | /* |
| 4843 | One of these is done in handle_if_exists_option(): |
| 4844 | thd->work_part_info= thd->lex->part_info; |
| 4845 | or |
| 4846 | thd->work_part_info= NULL; |
| 4847 | */ |
| 4848 | if (thd->work_part_info && |
| 4849 | !(thd->work_part_info= thd->work_part_info->get_clone(thd))) |
| 4850 | DBUG_RETURN(TRUE); |
| 4851 | |
| 4852 | /* ALTER_PARTITION_ADMIN is handled in mysql_admin_table */ |
| 4853 | DBUG_ASSERT(!(alter_info->partition_flags & ALTER_PARTITION_ADMIN)); |
| 4854 | |
| 4855 | partition_info *saved_part_info= NULL; |
| 4856 | |
| 4857 | if (alter_info->partition_flags & |
| 4858 | (ALTER_PARTITION_ADD | |
| 4859 | ALTER_PARTITION_DROP | |
| 4860 | ALTER_PARTITION_COALESCE | |
| 4861 | ALTER_PARTITION_REORGANIZE | |
| 4862 | ALTER_PARTITION_TABLE_REORG | |
| 4863 | ALTER_PARTITION_REBUILD)) |
| 4864 | { |
| 4865 | /* |
| 4866 | You can't add column when we are doing alter related to partition |
| 4867 | */ |
| 4868 | DBUG_EXECUTE_IF("test_pseudo_invisible" , { |
| 4869 | my_error(ER_INTERNAL_ERROR, MYF(0), "Don't to it with test_pseudo_invisible" ); |
| 4870 | DBUG_RETURN(1); |
| 4871 | }); |
| 4872 | DBUG_EXECUTE_IF("test_completely_invisible" , { |
| 4873 | my_error(ER_INTERNAL_ERROR, MYF(0), "Don't to it with test_completely_invisible" ); |
| 4874 | DBUG_RETURN(1); |
| 4875 | }); |
| 4876 | partition_info *tab_part_info; |
| 4877 | ulonglong flags= 0; |
| 4878 | bool is_last_partition_reorged= FALSE; |
| 4879 | part_elem_value *tab_max_elem_val= NULL; |
| 4880 | part_elem_value *alt_max_elem_val= NULL; |
| 4881 | longlong tab_max_range= 0, alt_max_range= 0; |
| 4882 | alt_part_info= thd->work_part_info; |
| 4883 | |
| 4884 | if (!table->part_info) |
| 4885 | { |
| 4886 | my_error(ER_PARTITION_MGMT_ON_NONPARTITIONED, MYF(0)); |
| 4887 | DBUG_RETURN(TRUE); |
| 4888 | } |
| 4889 | |
| 4890 | /* |
| 4891 | Open our intermediate table, we will operate on a temporary instance |
| 4892 | of the original table, to be able to skip copying all partitions. |
| 4893 | Open it as a copy of the original table, and modify its partition_info |
| 4894 | object to allow fast_alter_partition_table to perform the changes. |
| 4895 | */ |
| 4896 | DBUG_ASSERT(thd->mdl_context.is_lock_owner(MDL_key::TABLE, |
| 4897 | alter_ctx->db.str, |
| 4898 | alter_ctx->table_name.str, |
| 4899 | MDL_INTENTION_EXCLUSIVE)); |
| 4900 | |
| 4901 | tab_part_info= table->part_info; |
| 4902 | |
| 4903 | if (alter_info->partition_flags & ALTER_PARTITION_TABLE_REORG) |
| 4904 | { |
| 4905 | uint new_part_no, curr_part_no; |
| 4906 | /* |
| 4907 | 'ALTER TABLE t REORG PARTITION' only allowed with auto partition |
| 4908 | if default partitioning is used. |
| 4909 | */ |
| 4910 | |
| 4911 | if (tab_part_info->part_type != HASH_PARTITION || |
| 4912 | ((table->s->db_type()->partition_flags() & HA_USE_AUTO_PARTITION) && |
| 4913 | !tab_part_info->use_default_num_partitions) || |
| 4914 | ((!(table->s->db_type()->partition_flags() & HA_USE_AUTO_PARTITION)) && |
| 4915 | tab_part_info->use_default_num_partitions)) |
| 4916 | { |
| 4917 | my_error(ER_REORG_NO_PARAM_ERROR, MYF(0)); |
| 4918 | goto err; |
| 4919 | } |
| 4920 | new_part_no= table->file->get_default_no_partitions(create_info); |
| 4921 | curr_part_no= tab_part_info->num_parts; |
| 4922 | if (new_part_no == curr_part_no) |
| 4923 | { |
| 4924 | /* |
| 4925 | No change is needed, we will have the same number of partitions |
| 4926 | after the change as before. Thus we can reply ok immediately |
| 4927 | without any changes at all. |
| 4928 | */ |
| 4929 | flags= table->file->alter_table_flags(alter_info->flags); |
| 4930 | if (flags & (HA_FAST_CHANGE_PARTITION | HA_PARTITION_ONE_PHASE)) |
| 4931 | { |
| 4932 | *fast_alter_table= true; |
| 4933 | /* Force table re-open for consistency with the main case. */ |
| 4934 | table->m_needs_reopen= true; |
| 4935 | } |
| 4936 | else |
| 4937 | { |
| 4938 | /* |
| 4939 | Create copy of partition_info to avoid modifying original |
| 4940 | TABLE::part_info, to keep it safe for later use. |
| 4941 | */ |
| 4942 | if (!(tab_part_info= tab_part_info->get_clone(thd))) |
| 4943 | DBUG_RETURN(TRUE); |
| 4944 | } |
| 4945 | |
| 4946 | thd->work_part_info= tab_part_info; |
| 4947 | DBUG_RETURN(FALSE); |
| 4948 | } |
| 4949 | else if (new_part_no > curr_part_no) |
| 4950 | { |
| 4951 | /* |
| 4952 | We will add more partitions, we use the ADD PARTITION without |
| 4953 | setting the flag for no default number of partitions |
| 4954 | */ |
| 4955 | alter_info->partition_flags|= ALTER_PARTITION_ADD; |
| 4956 | thd->work_part_info->num_parts= new_part_no - curr_part_no; |
| 4957 | } |
| 4958 | else |
| 4959 | { |
| 4960 | /* |
| 4961 | We will remove hash partitions, we use the COALESCE PARTITION |
| 4962 | without setting the flag for no default number of partitions |
| 4963 | */ |
| 4964 | alter_info->partition_flags|= ALTER_PARTITION_COALESCE; |
| 4965 | alter_info->num_parts= curr_part_no - new_part_no; |
| 4966 | } |
| 4967 | } |
| 4968 | if (!(flags= table->file->alter_table_flags(alter_info->flags))) |
| 4969 | { |
| 4970 | my_error(ER_PARTITION_FUNCTION_FAILURE, MYF(0)); |
| 4971 | goto err; |
| 4972 | } |
| 4973 | if ((flags & (HA_FAST_CHANGE_PARTITION | HA_PARTITION_ONE_PHASE)) != 0) |
| 4974 | { |
| 4975 | /* |
| 4976 | "Fast" change of partitioning is supported in this case. |
| 4977 | We will change TABLE::part_info (as this is how we pass |
| 4978 | information to storage engine in this case), so the table |
| 4979 | must be reopened. |
| 4980 | */ |
| 4981 | *fast_alter_table= true; |
| 4982 | table->m_needs_reopen= true; |
| 4983 | } |
| 4984 | else |
| 4985 | { |
| 4986 | /* |
| 4987 | "Fast" changing of partitioning is not supported. Create |
| 4988 | a copy of TABLE::part_info object, so we can modify it safely. |
| 4989 | Modifying original TABLE::part_info will cause problems when |
| 4990 | we read data from old version of table using this TABLE object |
| 4991 | while copying them to new version of table. |
| 4992 | */ |
| 4993 | if (!(tab_part_info= tab_part_info->get_clone(thd))) |
| 4994 | DBUG_RETURN(TRUE); |
| 4995 | } |
| 4996 | DBUG_PRINT("info" , ("*fast_alter_table flags: 0x%llx" , flags)); |
| 4997 | if ((alter_info->partition_flags & ALTER_PARTITION_ADD) || |
| 4998 | (alter_info->partition_flags & ALTER_PARTITION_REORGANIZE)) |
| 4999 | { |
| 5000 | if (thd->work_part_info->part_type != tab_part_info->part_type) |
| 5001 | { |
| 5002 | if (thd->work_part_info->part_type == NOT_A_PARTITION) |
| 5003 | { |
| 5004 | if (tab_part_info->part_type == RANGE_PARTITION) |
| 5005 | { |
| 5006 | my_error(ER_PARTITIONS_MUST_BE_DEFINED_ERROR, MYF(0), "RANGE" ); |
| 5007 | goto err; |
| 5008 | } |
| 5009 | else if (tab_part_info->part_type == LIST_PARTITION) |
| 5010 | { |
| 5011 | my_error(ER_PARTITIONS_MUST_BE_DEFINED_ERROR, MYF(0), "LIST" ); |
| 5012 | goto err; |
| 5013 | } |
| 5014 | /* |
| 5015 | Hash partitions can be altered without parser finds out about |
| 5016 | that it is HASH partitioned. So no error here. |
| 5017 | */ |
| 5018 | } |
| 5019 | else |
| 5020 | { |
| 5021 | if (thd->work_part_info->part_type == RANGE_PARTITION) |
| 5022 | { |
| 5023 | my_error(ER_PARTITION_WRONG_VALUES_ERROR, MYF(0), |
| 5024 | "RANGE" , "LESS THAN" ); |
| 5025 | } |
| 5026 | else if (thd->work_part_info->part_type == LIST_PARTITION) |
| 5027 | { |
| 5028 | DBUG_ASSERT(thd->work_part_info->part_type == LIST_PARTITION); |
| 5029 | my_error(ER_PARTITION_WRONG_VALUES_ERROR, MYF(0), |
| 5030 | "LIST" , "IN" ); |
| 5031 | } |
| 5032 | else if (thd->work_part_info->part_type == VERSIONING_PARTITION) |
| 5033 | { |
| 5034 | my_error(ER_PARTITION_WRONG_TYPE, MYF(0), "SYSTEM_TIME" ); |
| 5035 | } |
| 5036 | else |
| 5037 | { |
| 5038 | DBUG_ASSERT(tab_part_info->part_type == RANGE_PARTITION || |
| 5039 | tab_part_info->part_type == LIST_PARTITION); |
| 5040 | (void) tab_part_info->error_if_requires_values(); |
| 5041 | } |
| 5042 | goto err; |
| 5043 | } |
| 5044 | } |
| 5045 | if ((tab_part_info->column_list && |
| 5046 | alt_part_info->num_columns != tab_part_info->num_columns && |
| 5047 | !only_default_value_added) || |
| 5048 | (!tab_part_info->column_list && |
| 5049 | (tab_part_info->part_type == RANGE_PARTITION || |
| 5050 | tab_part_info->part_type == LIST_PARTITION) && |
| 5051 | alt_part_info->num_columns != 1U && |
| 5052 | !only_default_value_added) || |
| 5053 | (!tab_part_info->column_list && |
| 5054 | tab_part_info->part_type == HASH_PARTITION && |
| 5055 | (alt_part_info->num_columns != 0))) |
| 5056 | { |
| 5057 | my_error(ER_PARTITION_COLUMN_LIST_ERROR, MYF(0)); |
| 5058 | goto err; |
| 5059 | } |
| 5060 | alt_part_info->column_list= tab_part_info->column_list; |
| 5061 | if (alt_part_info->fix_parser_data(thd)) |
| 5062 | { |
| 5063 | goto err; |
| 5064 | } |
| 5065 | } |
| 5066 | if (alter_info->partition_flags & ALTER_PARTITION_ADD) |
| 5067 | { |
| 5068 | if (*fast_alter_table && thd->locked_tables_mode) |
| 5069 | { |
| 5070 | MEM_ROOT *old_root= thd->mem_root; |
| 5071 | thd->mem_root= &thd->locked_tables_list.m_locked_tables_root; |
| 5072 | saved_part_info= tab_part_info->get_clone(thd); |
| 5073 | thd->mem_root= old_root; |
| 5074 | saved_part_info->read_partitions= tab_part_info->read_partitions; |
| 5075 | saved_part_info->lock_partitions= tab_part_info->lock_partitions; |
| 5076 | saved_part_info->bitmaps_are_initialized= tab_part_info->bitmaps_are_initialized; |
| 5077 | } |
| 5078 | /* |
| 5079 | We start by moving the new partitions to the list of temporary |
| 5080 | partitions. We will then check that the new partitions fit in the |
| 5081 | partitioning scheme as currently set-up. |
| 5082 | Partitions are always added at the end in ADD PARTITION. |
| 5083 | */ |
| 5084 | uint num_new_partitions= alt_part_info->num_parts; |
| 5085 | uint num_orig_partitions= tab_part_info->num_parts; |
| 5086 | uint check_total_partitions= num_new_partitions + num_orig_partitions; |
| 5087 | uint new_total_partitions= check_total_partitions; |
| 5088 | /* |
| 5089 | We allow quite a lot of values to be supplied by defaults, however we |
| 5090 | must know the number of new partitions in this case. |
| 5091 | */ |
| 5092 | if (thd->lex->no_write_to_binlog && |
| 5093 | tab_part_info->part_type != HASH_PARTITION && |
| 5094 | tab_part_info->part_type != VERSIONING_PARTITION) |
| 5095 | { |
| 5096 | my_error(ER_NO_BINLOG_ERROR, MYF(0)); |
| 5097 | goto err; |
| 5098 | } |
| 5099 | if (tab_part_info->defined_max_value && |
| 5100 | (tab_part_info->part_type == RANGE_PARTITION || |
| 5101 | alt_part_info->defined_max_value)) |
| 5102 | { |
| 5103 | my_error((tab_part_info->part_type == RANGE_PARTITION ? |
| 5104 | ER_PARTITION_MAXVALUE_ERROR : |
| 5105 | ER_PARTITION_DEFAULT_ERROR), MYF(0)); |
| 5106 | goto err; |
| 5107 | } |
| 5108 | if (num_new_partitions == 0) |
| 5109 | { |
| 5110 | my_error(ER_ADD_PARTITION_NO_NEW_PARTITION, MYF(0)); |
| 5111 | goto err; |
| 5112 | } |
| 5113 | if (tab_part_info->is_sub_partitioned()) |
| 5114 | { |
| 5115 | if (alt_part_info->num_subparts == 0) |
| 5116 | alt_part_info->num_subparts= tab_part_info->num_subparts; |
| 5117 | else if (alt_part_info->num_subparts != tab_part_info->num_subparts) |
| 5118 | { |
| 5119 | my_error(ER_ADD_PARTITION_SUBPART_ERROR, MYF(0)); |
| 5120 | goto err; |
| 5121 | } |
| 5122 | check_total_partitions= new_total_partitions* |
| 5123 | alt_part_info->num_subparts; |
| 5124 | } |
| 5125 | if (check_total_partitions > MAX_PARTITIONS) |
| 5126 | { |
| 5127 | my_error(ER_TOO_MANY_PARTITIONS_ERROR, MYF(0)); |
| 5128 | goto err; |
| 5129 | } |
| 5130 | alt_part_info->part_type= tab_part_info->part_type; |
| 5131 | alt_part_info->subpart_type= tab_part_info->subpart_type; |
| 5132 | if (alt_part_info->set_up_defaults_for_partitioning(thd, table->file, 0, |
| 5133 | tab_part_info->num_parts)) |
| 5134 | { |
| 5135 | goto err; |
| 5136 | } |
| 5137 | /* |
| 5138 | Handling of on-line cases: |
| 5139 | |
| 5140 | ADD PARTITION for RANGE/LIST PARTITIONING: |
| 5141 | ------------------------------------------ |
| 5142 | For range and list partitions add partition is simply adding a |
| 5143 | new empty partition to the table. If the handler support this we |
| 5144 | will use the simple method of doing this. The figure below shows |
| 5145 | an example of this and the states involved in making this change. |
| 5146 | |
| 5147 | Existing partitions New added partitions |
| 5148 | ------ ------ ------ ------ | ------ ------ |
| 5149 | | | | | | | | | | | | | | |
| 5150 | | p0 | | p1 | | p2 | | p3 | | | p4 | | p5 | |
| 5151 | ------ ------ ------ ------ | ------ ------ |
| 5152 | PART_NORMAL PART_NORMAL PART_NORMAL PART_NORMAL PART_TO_BE_ADDED*2 |
| 5153 | PART_NORMAL PART_NORMAL PART_NORMAL PART_NORMAL PART_IS_ADDED*2 |
| 5154 | |
| 5155 | The first line is the states before adding the new partitions and the |
| 5156 | second line is after the new partitions are added. All the partitions are |
| 5157 | in the partitions list, no partitions are placed in the temp_partitions |
| 5158 | list. |
| 5159 | |
| 5160 | ADD PARTITION for HASH PARTITIONING |
| 5161 | ----------------------------------- |
| 5162 | This little figure tries to show the various partitions involved when |
| 5163 | adding two new partitions to a linear hash based partitioned table with |
| 5164 | four partitions to start with, which lists are used and the states they |
| 5165 | pass through. Adding partitions to a normal hash based is similar except |
| 5166 | that it is always all the existing partitions that are reorganised not |
| 5167 | only a subset of them. |
| 5168 | |
| 5169 | Existing partitions New added partitions |
| 5170 | ------ ------ ------ ------ | ------ ------ |
| 5171 | | | | | | | | | | | | | | |
| 5172 | | p0 | | p1 | | p2 | | p3 | | | p4 | | p5 | |
| 5173 | ------ ------ ------ ------ | ------ ------ |
| 5174 | PART_CHANGED PART_CHANGED PART_NORMAL PART_NORMAL PART_TO_BE_ADDED |
| 5175 | PART_IS_CHANGED*2 PART_NORMAL PART_NORMAL PART_IS_ADDED |
| 5176 | PART_NORMAL PART_NORMAL PART_NORMAL PART_NORMAL PART_IS_ADDED |
| 5177 | |
| 5178 | Reorganised existing partitions |
| 5179 | ------ ------ |
| 5180 | | | | | |
| 5181 | | p0'| | p1'| |
| 5182 | ------ ------ |
| 5183 | |
| 5184 | p0 - p5 will be in the partitions list of partitions. |
| 5185 | p0' and p1' will actually not exist as separate objects, there presence can |
| 5186 | be deduced from the state of the partition and also the names of those |
| 5187 | partitions can be deduced this way. |
| 5188 | |
| 5189 | After adding the partitions and copying the partition data to p0', p1', |
| 5190 | p4 and p5 from p0 and p1 the states change to adapt for the new situation |
| 5191 | where p0 and p1 is dropped and replaced by p0' and p1' and the new p4 and |
| 5192 | p5 are in the table again. |
| 5193 | |
| 5194 | The first line above shows the states of the partitions before we start |
| 5195 | adding and copying partitions, the second after completing the adding |
| 5196 | and copying and finally the third line after also dropping the partitions |
| 5197 | that are reorganised. |
| 5198 | */ |
| 5199 | if (*fast_alter_table && tab_part_info->part_type == HASH_PARTITION) |
| 5200 | { |
| 5201 | uint part_no= 0, start_part= 1, start_sec_part= 1; |
| 5202 | uint end_part= 0, end_sec_part= 0; |
| 5203 | uint upper_2n= tab_part_info->linear_hash_mask + 1; |
| 5204 | uint lower_2n= upper_2n >> 1; |
| 5205 | bool all_parts= TRUE; |
| 5206 | if (tab_part_info->linear_hash_ind && num_new_partitions < upper_2n) |
| 5207 | { |
| 5208 | /* |
| 5209 | An analysis of which parts needs reorganisation shows that it is |
| 5210 | divided into two intervals. The first interval is those parts |
| 5211 | that are reorganised up until upper_2n - 1. From upper_2n and |
| 5212 | onwards it starts again from partition 0 and goes on until |
| 5213 | it reaches p(upper_2n - 1). If the last new partition reaches |
| 5214 | beyond upper_2n - 1 then the first interval will end with |
| 5215 | p(lower_2n - 1) and start with p(num_orig_partitions - lower_2n). |
| 5216 | If lower_2n partitions are added then p0 to p(lower_2n - 1) will |
| 5217 | be reorganised which means that the two interval becomes one |
| 5218 | interval at this point. Thus only when adding less than |
| 5219 | lower_2n partitions and going beyond a total of upper_2n we |
| 5220 | actually get two intervals. |
| 5221 | |
| 5222 | To exemplify this assume we have 6 partitions to start with and |
| 5223 | add 1, 2, 3, 5, 6, 7, 8, 9 partitions. |
| 5224 | The first to add after p5 is p6 = 110 in bit numbers. Thus we |
| 5225 | can see that 10 = p2 will be partition to reorganise if only one |
| 5226 | partition. |
| 5227 | If 2 partitions are added we reorganise [p2, p3]. Those two |
| 5228 | cases are covered by the second if part below. |
| 5229 | If 3 partitions are added we reorganise [p2, p3] U [p0,p0]. This |
| 5230 | part is covered by the else part below. |
| 5231 | If 5 partitions are added we get [p2,p3] U [p0, p2] = [p0, p3]. |
| 5232 | This is covered by the first if part where we need the max check |
| 5233 | to here use lower_2n - 1. |
| 5234 | If 7 partitions are added we get [p2,p3] U [p0, p4] = [p0, p4]. |
| 5235 | This is covered by the first if part but here we use the first |
| 5236 | calculated end_part. |
| 5237 | Finally with 9 new partitions we would also reorganise p6 if we |
| 5238 | used the method below but we cannot reorganise more partitions |
| 5239 | than what we had from the start and thus we simply set all_parts |
| 5240 | to TRUE. In this case we don't get into this if-part at all. |
| 5241 | */ |
| 5242 | all_parts= FALSE; |
| 5243 | if (num_new_partitions >= lower_2n) |
| 5244 | { |
| 5245 | /* |
| 5246 | In this case there is only one interval since the two intervals |
| 5247 | overlap and this starts from zero to last_part_no - upper_2n |
| 5248 | */ |
| 5249 | start_part= 0; |
| 5250 | end_part= new_total_partitions - (upper_2n + 1); |
| 5251 | end_part= max(lower_2n - 1, end_part); |
| 5252 | } |
| 5253 | else if (new_total_partitions <= upper_2n) |
| 5254 | { |
| 5255 | /* |
| 5256 | Also in this case there is only one interval since we are not |
| 5257 | going over a 2**n boundary |
| 5258 | */ |
| 5259 | start_part= num_orig_partitions - lower_2n; |
| 5260 | end_part= start_part + (num_new_partitions - 1); |
| 5261 | } |
| 5262 | else |
| 5263 | { |
| 5264 | /* We have two non-overlapping intervals since we are not |
| 5265 | passing a 2**n border and we have not at least lower_2n |
| 5266 | new parts that would ensure that the intervals become |
| 5267 | overlapping. |
| 5268 | */ |
| 5269 | start_part= num_orig_partitions - lower_2n; |
| 5270 | end_part= upper_2n - 1; |
| 5271 | start_sec_part= 0; |
| 5272 | end_sec_part= new_total_partitions - (upper_2n + 1); |
| 5273 | } |
| 5274 | } |
| 5275 | List_iterator<partition_element> tab_it(tab_part_info->partitions); |
| 5276 | part_no= 0; |
| 5277 | do |
| 5278 | { |
| 5279 | partition_element *p_elem= tab_it++; |
| 5280 | if (all_parts || |
| 5281 | (part_no >= start_part && part_no <= end_part) || |
| 5282 | (part_no >= start_sec_part && part_no <= end_sec_part)) |
| 5283 | { |
| 5284 | p_elem->part_state= PART_CHANGED; |
| 5285 | } |
| 5286 | } while (++part_no < num_orig_partitions); |
| 5287 | } |
| 5288 | /* |
| 5289 | Need to concatenate the lists here to make it possible to check the |
| 5290 | partition info for correctness using check_partition_info. |
| 5291 | For on-line add partition we set the state of this partition to |
| 5292 | PART_TO_BE_ADDED to ensure that it is known that it is not yet |
| 5293 | usable (becomes usable when partition is created and the switch of |
| 5294 | partition configuration is made. |
| 5295 | */ |
| 5296 | { |
| 5297 | partition_element *now_part= NULL; |
| 5298 | if (tab_part_info->part_type == VERSIONING_PARTITION) |
| 5299 | { |
| 5300 | List_iterator<partition_element> it(tab_part_info->partitions); |
| 5301 | partition_element *el; |
| 5302 | while ((el= it++)) |
| 5303 | { |
| 5304 | if (el->type() == partition_element::CURRENT) |
| 5305 | { |
| 5306 | it.remove(); |
| 5307 | now_part= el; |
| 5308 | } |
| 5309 | } |
| 5310 | if (*fast_alter_table && tab_part_info->vers_info->interval.is_set()) |
| 5311 | { |
| 5312 | partition_element *hist_part= tab_part_info->vers_info->hist_part; |
| 5313 | if (hist_part->range_value <= thd->query_start()) |
| 5314 | hist_part->part_state= PART_CHANGED; |
| 5315 | } |
| 5316 | } |
| 5317 | List_iterator<partition_element> alt_it(alt_part_info->partitions); |
| 5318 | uint part_count= 0; |
| 5319 | do |
| 5320 | { |
| 5321 | partition_element *part_elem= alt_it++; |
| 5322 | if (*fast_alter_table) |
| 5323 | part_elem->part_state= PART_TO_BE_ADDED; |
| 5324 | if (unlikely(tab_part_info->partitions.push_back(part_elem, |
| 5325 | thd->mem_root))) |
| 5326 | goto err; |
| 5327 | } while (++part_count < num_new_partitions); |
| 5328 | tab_part_info->num_parts+= num_new_partitions; |
| 5329 | if (tab_part_info->part_type == VERSIONING_PARTITION) |
| 5330 | { |
| 5331 | DBUG_ASSERT(now_part); |
| 5332 | if (unlikely(tab_part_info->partitions.push_back(now_part, |
| 5333 | thd->mem_root))) |
| 5334 | goto err; |
| 5335 | } |
| 5336 | } |
| 5337 | /* |
| 5338 | If we specify partitions explicitly we don't use defaults anymore. |
| 5339 | Using ADD PARTITION also means that we don't have the default number |
| 5340 | of partitions anymore. We use this code also for Table reorganisations |
| 5341 | and here we don't set any default flags to FALSE. |
| 5342 | */ |
| 5343 | if (!(alter_info->partition_flags & ALTER_PARTITION_TABLE_REORG)) |
| 5344 | { |
| 5345 | if (!alt_part_info->use_default_partitions) |
| 5346 | { |
| 5347 | DBUG_PRINT("info" , ("part_info: %p" , tab_part_info)); |
| 5348 | tab_part_info->use_default_partitions= FALSE; |
| 5349 | } |
| 5350 | tab_part_info->use_default_num_partitions= FALSE; |
| 5351 | tab_part_info->is_auto_partitioned= FALSE; |
| 5352 | } |
| 5353 | } |
| 5354 | else if (alter_info->partition_flags & ALTER_PARTITION_DROP) |
| 5355 | { |
| 5356 | /* |
| 5357 | Drop a partition from a range partition and list partitioning is |
| 5358 | always safe and can be made more or less immediate. It is necessary |
| 5359 | however to ensure that the partition to be removed is safely removed |
| 5360 | and that REPAIR TABLE can remove the partition if for some reason the |
| 5361 | command to drop the partition failed in the middle. |
| 5362 | */ |
| 5363 | uint part_count= 0; |
| 5364 | uint num_parts_dropped= alter_info->partition_names.elements; |
| 5365 | uint num_parts_found= 0; |
| 5366 | List_iterator<partition_element> part_it(tab_part_info->partitions); |
| 5367 | |
| 5368 | tab_part_info->is_auto_partitioned= FALSE; |
| 5369 | if (tab_part_info->part_type == VERSIONING_PARTITION) |
| 5370 | { |
| 5371 | if (num_parts_dropped >= tab_part_info->num_parts - 1) |
| 5372 | { |
| 5373 | DBUG_ASSERT(table && table->s && table->s->table_name.str); |
| 5374 | my_error(ER_VERS_WRONG_PARTS, MYF(0), table->s->table_name.str); |
| 5375 | goto err; |
| 5376 | } |
| 5377 | } |
| 5378 | else |
| 5379 | { |
| 5380 | if (!(tab_part_info->part_type == RANGE_PARTITION || |
| 5381 | tab_part_info->part_type == LIST_PARTITION)) |
| 5382 | { |
| 5383 | my_error(ER_ONLY_ON_RANGE_LIST_PARTITION, MYF(0), "DROP" ); |
| 5384 | goto err; |
| 5385 | } |
| 5386 | if (num_parts_dropped >= tab_part_info->num_parts) |
| 5387 | { |
| 5388 | my_error(ER_DROP_LAST_PARTITION, MYF(0)); |
| 5389 | goto err; |
| 5390 | } |
| 5391 | } |
| 5392 | do |
| 5393 | { |
| 5394 | partition_element *part_elem= part_it++; |
| 5395 | if (is_name_in_list(part_elem->partition_name, |
| 5396 | alter_info->partition_names)) |
| 5397 | { |
| 5398 | if (tab_part_info->part_type == VERSIONING_PARTITION) |
| 5399 | { |
| 5400 | if (part_elem->type() == partition_element::CURRENT) |
| 5401 | { |
| 5402 | my_error(ER_VERS_WRONG_PARTS, MYF(0), table->s->table_name.str); |
| 5403 | goto err; |
| 5404 | } |
| 5405 | if (tab_part_info->vers_info->interval.is_set()) |
| 5406 | { |
| 5407 | if (num_parts_found < part_count) |
| 5408 | { |
| 5409 | my_error(ER_VERS_DROP_PARTITION_INTERVAL, MYF(0)); |
| 5410 | goto err; |
| 5411 | } |
| 5412 | tab_part_info->vers_info->interval.start= |
| 5413 | (my_time_t)part_elem->range_value; |
| 5414 | } |
| 5415 | } |
| 5416 | /* |
| 5417 | Set state to indicate that the partition is to be dropped. |
| 5418 | */ |
| 5419 | num_parts_found++; |
| 5420 | part_elem->part_state= PART_TO_BE_DROPPED; |
| 5421 | } |
| 5422 | } while (++part_count < tab_part_info->num_parts); |
| 5423 | if (num_parts_found != num_parts_dropped) |
| 5424 | { |
| 5425 | my_error(ER_DROP_PARTITION_NON_EXISTENT, MYF(0), "DROP" ); |
| 5426 | goto err; |
| 5427 | } |
| 5428 | if (table->file->is_fk_defined_on_table_or_index(MAX_KEY)) |
| 5429 | { |
| 5430 | my_error(ER_ROW_IS_REFERENCED, MYF(0)); |
| 5431 | goto err; |
| 5432 | } |
| 5433 | tab_part_info->num_parts-= num_parts_dropped; |
| 5434 | } |
| 5435 | else if (alter_info->partition_flags & ALTER_PARTITION_REBUILD) |
| 5436 | { |
| 5437 | set_engine_all_partitions(tab_part_info, |
| 5438 | tab_part_info->default_engine_type); |
| 5439 | if (set_part_state(alter_info, tab_part_info, PART_CHANGED)) |
| 5440 | { |
| 5441 | my_error(ER_DROP_PARTITION_NON_EXISTENT, MYF(0), "REBUILD" ); |
| 5442 | goto err; |
| 5443 | } |
| 5444 | if (!(*fast_alter_table)) |
| 5445 | { |
| 5446 | table->file->print_error(HA_ERR_WRONG_COMMAND, MYF(0)); |
| 5447 | goto err; |
| 5448 | } |
| 5449 | } |
| 5450 | else if (alter_info->partition_flags & ALTER_PARTITION_COALESCE) |
| 5451 | { |
| 5452 | uint num_parts_coalesced= alter_info->num_parts; |
| 5453 | uint num_parts_remain= tab_part_info->num_parts - num_parts_coalesced; |
| 5454 | List_iterator<partition_element> part_it(tab_part_info->partitions); |
| 5455 | if (tab_part_info->part_type != HASH_PARTITION) |
| 5456 | { |
| 5457 | my_error(ER_COALESCE_ONLY_ON_HASH_PARTITION, MYF(0)); |
| 5458 | goto err; |
| 5459 | } |
| 5460 | if (num_parts_coalesced == 0) |
| 5461 | { |
| 5462 | my_error(ER_COALESCE_PARTITION_NO_PARTITION, MYF(0)); |
| 5463 | goto err; |
| 5464 | } |
| 5465 | if (num_parts_coalesced >= tab_part_info->num_parts) |
| 5466 | { |
| 5467 | my_error(ER_DROP_LAST_PARTITION, MYF(0)); |
| 5468 | goto err; |
| 5469 | } |
| 5470 | /* |
| 5471 | Online handling: |
| 5472 | COALESCE PARTITION: |
| 5473 | ------------------- |
| 5474 | The figure below shows the manner in which partitions are handled when |
| 5475 | performing an on-line coalesce partition and which states they go through |
| 5476 | at start, after adding and copying partitions and finally after dropping |
| 5477 | the partitions to drop. The figure shows an example using four partitions |
| 5478 | to start with, using linear hash and coalescing one partition (always the |
| 5479 | last partition). |
| 5480 | |
| 5481 | Using linear hash then all remaining partitions will have a new reorganised |
| 5482 | part. |
| 5483 | |
| 5484 | Existing partitions Coalesced partition |
| 5485 | ------ ------ ------ | ------ |
| 5486 | | | | | | | | | | |
| 5487 | | p0 | | p1 | | p2 | | | p3 | |
| 5488 | ------ ------ ------ | ------ |
| 5489 | PART_NORMAL PART_CHANGED PART_NORMAL PART_REORGED_DROPPED |
| 5490 | PART_NORMAL PART_IS_CHANGED PART_NORMAL PART_TO_BE_DROPPED |
| 5491 | PART_NORMAL PART_NORMAL PART_NORMAL PART_IS_DROPPED |
| 5492 | |
| 5493 | Reorganised existing partitions |
| 5494 | ------ |
| 5495 | | | |
| 5496 | | p1'| |
| 5497 | ------ |
| 5498 | |
| 5499 | p0 - p3 is in the partitions list. |
| 5500 | The p1' partition will actually not be in any list it is deduced from the |
| 5501 | state of p1. |
| 5502 | */ |
| 5503 | { |
| 5504 | uint part_count= 0, start_part= 1, start_sec_part= 1; |
| 5505 | uint end_part= 0, end_sec_part= 0; |
| 5506 | bool all_parts= TRUE; |
| 5507 | if (*fast_alter_table && |
| 5508 | tab_part_info->linear_hash_ind) |
| 5509 | { |
| 5510 | uint upper_2n= tab_part_info->linear_hash_mask + 1; |
| 5511 | uint lower_2n= upper_2n >> 1; |
| 5512 | all_parts= FALSE; |
| 5513 | if (num_parts_coalesced >= lower_2n) |
| 5514 | { |
| 5515 | all_parts= TRUE; |
| 5516 | } |
| 5517 | else if (num_parts_remain >= lower_2n) |
| 5518 | { |
| 5519 | end_part= tab_part_info->num_parts - (lower_2n + 1); |
| 5520 | start_part= num_parts_remain - lower_2n; |
| 5521 | } |
| 5522 | else |
| 5523 | { |
| 5524 | start_part= 0; |
| 5525 | end_part= tab_part_info->num_parts - (lower_2n + 1); |
| 5526 | end_sec_part= (lower_2n >> 1) - 1; |
| 5527 | start_sec_part= end_sec_part - (lower_2n - (num_parts_remain + 1)); |
| 5528 | } |
| 5529 | } |
| 5530 | do |
| 5531 | { |
| 5532 | partition_element *p_elem= part_it++; |
| 5533 | if (*fast_alter_table && |
| 5534 | (all_parts || |
| 5535 | (part_count >= start_part && part_count <= end_part) || |
| 5536 | (part_count >= start_sec_part && part_count <= end_sec_part))) |
| 5537 | p_elem->part_state= PART_CHANGED; |
| 5538 | if (++part_count > num_parts_remain) |
| 5539 | { |
| 5540 | if (*fast_alter_table) |
| 5541 | p_elem->part_state= PART_REORGED_DROPPED; |
| 5542 | else |
| 5543 | part_it.remove(); |
| 5544 | } |
| 5545 | } while (part_count < tab_part_info->num_parts); |
| 5546 | tab_part_info->num_parts= num_parts_remain; |
| 5547 | } |
| 5548 | if (!(alter_info->partition_flags & ALTER_PARTITION_TABLE_REORG)) |
| 5549 | { |
| 5550 | tab_part_info->use_default_num_partitions= FALSE; |
| 5551 | tab_part_info->is_auto_partitioned= FALSE; |
| 5552 | } |
| 5553 | } |
| 5554 | else if (alter_info->partition_flags & ALTER_PARTITION_REORGANIZE) |
| 5555 | { |
| 5556 | /* |
| 5557 | Reorganise partitions takes a number of partitions that are next |
| 5558 | to each other (at least for RANGE PARTITIONS) and then uses those |
| 5559 | to create a set of new partitions. So data is copied from those |
| 5560 | partitions into the new set of partitions. Those new partitions |
| 5561 | can have more values in the LIST value specifications or less both |
| 5562 | are allowed. The ranges can be different but since they are |
| 5563 | changing a set of consecutive partitions they must cover the same |
| 5564 | range as those changed from. |
| 5565 | This command can be used on RANGE and LIST partitions. |
| 5566 | */ |
| 5567 | uint num_parts_reorged= alter_info->partition_names.elements; |
| 5568 | uint num_parts_new= thd->work_part_info->partitions.elements; |
| 5569 | uint check_total_partitions; |
| 5570 | |
| 5571 | tab_part_info->is_auto_partitioned= FALSE; |
| 5572 | if (num_parts_reorged > tab_part_info->num_parts) |
| 5573 | { |
| 5574 | my_error(ER_REORG_PARTITION_NOT_EXIST, MYF(0)); |
| 5575 | goto err; |
| 5576 | } |
| 5577 | if (!(tab_part_info->part_type == RANGE_PARTITION || |
| 5578 | tab_part_info->part_type == LIST_PARTITION) && |
| 5579 | (num_parts_new != num_parts_reorged)) |
| 5580 | { |
| 5581 | my_error(ER_REORG_HASH_ONLY_ON_SAME_NO, MYF(0)); |
| 5582 | goto err; |
| 5583 | } |
| 5584 | if (tab_part_info->is_sub_partitioned() && |
| 5585 | alt_part_info->num_subparts && |
| 5586 | alt_part_info->num_subparts != tab_part_info->num_subparts) |
| 5587 | { |
| 5588 | my_error(ER_PARTITION_WRONG_NO_SUBPART_ERROR, MYF(0)); |
| 5589 | goto err; |
| 5590 | } |
| 5591 | check_total_partitions= tab_part_info->num_parts + num_parts_new; |
| 5592 | check_total_partitions-= num_parts_reorged; |
| 5593 | if (check_total_partitions > MAX_PARTITIONS) |
| 5594 | { |
| 5595 | my_error(ER_TOO_MANY_PARTITIONS_ERROR, MYF(0)); |
| 5596 | goto err; |
| 5597 | } |
| 5598 | alt_part_info->part_type= tab_part_info->part_type; |
| 5599 | alt_part_info->subpart_type= tab_part_info->subpart_type; |
| 5600 | alt_part_info->num_subparts= tab_part_info->num_subparts; |
| 5601 | DBUG_ASSERT(!alt_part_info->use_default_partitions); |
| 5602 | /* We specified partitions explicitly so don't use defaults anymore. */ |
| 5603 | tab_part_info->use_default_partitions= FALSE; |
| 5604 | if (alt_part_info->set_up_defaults_for_partitioning(thd, table->file, 0, |
| 5605 | 0)) |
| 5606 | { |
| 5607 | goto err; |
| 5608 | } |
| 5609 | /* |
| 5610 | Online handling: |
| 5611 | REORGANIZE PARTITION: |
| 5612 | --------------------- |
| 5613 | The figure exemplifies the handling of partitions, their state changes and |
| 5614 | how they are organised. It exemplifies four partitions where two of the |
| 5615 | partitions are reorganised (p1 and p2) into two new partitions (p4 and p5). |
| 5616 | The reason of this change could be to change range limits, change list |
| 5617 | values or for hash partitions simply reorganise the partition which could |
| 5618 | also involve moving them to new disks or new node groups (MySQL Cluster). |
| 5619 | |
| 5620 | Existing partitions |
| 5621 | ------ ------ ------ ------ |
| 5622 | | | | | | | | | |
| 5623 | | p0 | | p1 | | p2 | | p3 | |
| 5624 | ------ ------ ------ ------ |
| 5625 | PART_NORMAL PART_TO_BE_REORGED PART_NORMAL |
| 5626 | PART_NORMAL PART_TO_BE_DROPPED PART_NORMAL |
| 5627 | PART_NORMAL PART_IS_DROPPED PART_NORMAL |
| 5628 | |
| 5629 | Reorganised new partitions (replacing p1 and p2) |
| 5630 | ------ ------ |
| 5631 | | | | | |
| 5632 | | p4 | | p5 | |
| 5633 | ------ ------ |
| 5634 | PART_TO_BE_ADDED |
| 5635 | PART_IS_ADDED |
| 5636 | PART_IS_ADDED |
| 5637 | |
| 5638 | All unchanged partitions and the new partitions are in the partitions list |
| 5639 | in the order they will have when the change is completed. The reorganised |
| 5640 | partitions are placed in the temp_partitions list. PART_IS_ADDED is only a |
| 5641 | temporary state not written in the frm file. It is used to ensure we write |
| 5642 | the generated partition syntax in a correct manner. |
| 5643 | */ |
| 5644 | { |
| 5645 | List_iterator<partition_element> tab_it(tab_part_info->partitions); |
| 5646 | uint part_count= 0; |
| 5647 | bool found_first= FALSE; |
| 5648 | bool found_last= FALSE; |
| 5649 | uint drop_count= 0; |
| 5650 | do |
| 5651 | { |
| 5652 | partition_element *part_elem= tab_it++; |
| 5653 | is_last_partition_reorged= FALSE; |
| 5654 | if (is_name_in_list(part_elem->partition_name, |
| 5655 | alter_info->partition_names)) |
| 5656 | { |
| 5657 | is_last_partition_reorged= TRUE; |
| 5658 | drop_count++; |
| 5659 | if (tab_part_info->column_list) |
| 5660 | { |
| 5661 | List_iterator<part_elem_value> p(part_elem->list_val_list); |
| 5662 | tab_max_elem_val= p++; |
| 5663 | } |
| 5664 | else |
| 5665 | tab_max_range= part_elem->range_value; |
| 5666 | if (*fast_alter_table && |
| 5667 | unlikely(tab_part_info->temp_partitions. |
| 5668 | push_back(part_elem, thd->mem_root))) |
| 5669 | goto err; |
| 5670 | |
| 5671 | if (*fast_alter_table) |
| 5672 | part_elem->part_state= PART_TO_BE_REORGED; |
| 5673 | if (!found_first) |
| 5674 | { |
| 5675 | uint alt_part_count= 0; |
| 5676 | partition_element *alt_part_elem; |
| 5677 | List_iterator<partition_element> |
| 5678 | alt_it(alt_part_info->partitions); |
| 5679 | found_first= TRUE; |
| 5680 | do |
| 5681 | { |
| 5682 | alt_part_elem= alt_it++; |
| 5683 | if (tab_part_info->column_list) |
| 5684 | { |
| 5685 | List_iterator<part_elem_value> p(alt_part_elem->list_val_list); |
| 5686 | alt_max_elem_val= p++; |
| 5687 | } |
| 5688 | else |
| 5689 | alt_max_range= alt_part_elem->range_value; |
| 5690 | |
| 5691 | if (*fast_alter_table) |
| 5692 | alt_part_elem->part_state= PART_TO_BE_ADDED; |
| 5693 | if (alt_part_count == 0) |
| 5694 | tab_it.replace(alt_part_elem); |
| 5695 | else |
| 5696 | tab_it.after(alt_part_elem); |
| 5697 | } while (++alt_part_count < num_parts_new); |
| 5698 | } |
| 5699 | else if (found_last) |
| 5700 | { |
| 5701 | my_error(ER_CONSECUTIVE_REORG_PARTITIONS, MYF(0)); |
| 5702 | goto err; |
| 5703 | } |
| 5704 | else |
| 5705 | tab_it.remove(); |
| 5706 | } |
| 5707 | else |
| 5708 | { |
| 5709 | if (found_first) |
| 5710 | found_last= TRUE; |
| 5711 | } |
| 5712 | } while (++part_count < tab_part_info->num_parts); |
| 5713 | if (drop_count != num_parts_reorged) |
| 5714 | { |
| 5715 | my_error(ER_DROP_PARTITION_NON_EXISTENT, MYF(0), "REORGANIZE" ); |
| 5716 | goto err; |
| 5717 | } |
| 5718 | tab_part_info->num_parts= check_total_partitions; |
| 5719 | } |
| 5720 | } |
| 5721 | else |
| 5722 | { |
| 5723 | DBUG_ASSERT(FALSE); |
| 5724 | } |
| 5725 | *partition_changed= TRUE; |
| 5726 | thd->work_part_info= tab_part_info; |
| 5727 | if (alter_info->partition_flags & (ALTER_PARTITION_ADD | |
| 5728 | ALTER_PARTITION_REORGANIZE)) |
| 5729 | { |
| 5730 | if (tab_part_info->use_default_subpartitions && |
| 5731 | !alt_part_info->use_default_subpartitions) |
| 5732 | { |
| 5733 | tab_part_info->use_default_subpartitions= FALSE; |
| 5734 | tab_part_info->use_default_num_subpartitions= FALSE; |
| 5735 | } |
| 5736 | |
| 5737 | if (tab_part_info->check_partition_info(thd, (handlerton**)NULL, |
| 5738 | table->file, 0, alt_part_info)) |
| 5739 | { |
| 5740 | goto err; |
| 5741 | } |
| 5742 | /* |
| 5743 | The check below needs to be performed after check_partition_info |
| 5744 | since this function "fixes" the item trees of the new partitions |
| 5745 | to reorganize into |
| 5746 | */ |
| 5747 | if (alter_info->partition_flags == ALTER_PARTITION_REORGANIZE && |
| 5748 | tab_part_info->part_type == RANGE_PARTITION && |
| 5749 | ((is_last_partition_reorged && |
| 5750 | (tab_part_info->column_list ? |
| 5751 | (partition_info_compare_column_values( |
| 5752 | alt_max_elem_val->col_val_array, |
| 5753 | tab_max_elem_val->col_val_array) < 0) : |
| 5754 | alt_max_range < tab_max_range)) || |
| 5755 | (!is_last_partition_reorged && |
| 5756 | (tab_part_info->column_list ? |
| 5757 | (partition_info_compare_column_values( |
| 5758 | alt_max_elem_val->col_val_array, |
| 5759 | tab_max_elem_val->col_val_array) != 0) : |
| 5760 | alt_max_range != tab_max_range)))) |
| 5761 | { |
| 5762 | /* |
| 5763 | For range partitioning the total resulting range before and |
| 5764 | after the change must be the same except in one case. This is |
| 5765 | when the last partition is reorganised, in this case it is |
| 5766 | acceptable to increase the total range. |
| 5767 | The reason is that it is not allowed to have "holes" in the |
| 5768 | middle of the ranges and thus we should not allow to reorganise |
| 5769 | to create "holes". |
| 5770 | */ |
| 5771 | my_error(ER_REORG_OUTSIDE_RANGE, MYF(0)); |
| 5772 | goto err; |
| 5773 | } |
| 5774 | } |
| 5775 | } // ADD, DROP, COALESCE, REORGANIZE, TABLE_REORG, REBUILD |
| 5776 | else |
| 5777 | { |
| 5778 | /* |
| 5779 | When thd->lex->part_info has a reference to a partition_info the |
| 5780 | ALTER TABLE contained a definition of a partitioning. |
| 5781 | |
| 5782 | Case I: |
| 5783 | If there was a partition before and there is a new one defined. |
| 5784 | We use the new partitioning. The new partitioning is already |
| 5785 | defined in the correct variable so no work is needed to |
| 5786 | accomplish this. |
| 5787 | We do however need to update partition_changed to ensure that not |
| 5788 | only the frm file is changed in the ALTER TABLE command. |
| 5789 | |
| 5790 | Case IIa: |
| 5791 | There was a partitioning before and there is no new one defined. |
| 5792 | Also the user has not specified to remove partitioning explicitly. |
| 5793 | |
| 5794 | We use the old partitioning also for the new table. We do this |
| 5795 | by assigning the partition_info from the table loaded in |
| 5796 | open_table to the partition_info struct used by mysql_create_table |
| 5797 | later in this method. |
| 5798 | |
| 5799 | Case IIb: |
| 5800 | There was a partitioning before and there is no new one defined. |
| 5801 | The user has specified explicitly to remove partitioning |
| 5802 | |
| 5803 | Since the user has specified explicitly to remove partitioning |
| 5804 | we override the old partitioning info and create a new table using |
| 5805 | the specified engine. |
| 5806 | In this case the partition also is changed. |
| 5807 | |
| 5808 | Case III: |
| 5809 | There was no partitioning before altering the table, there is |
| 5810 | partitioning defined in the altered table. Use the new partitioning. |
| 5811 | No work needed since the partitioning info is already in the |
| 5812 | correct variable. |
| 5813 | |
| 5814 | In this case we discover one case where the new partitioning is using |
| 5815 | the same partition function as the default (PARTITION BY KEY or |
| 5816 | PARTITION BY LINEAR KEY with the list of fields equal to the primary |
| 5817 | key fields OR PARTITION BY [LINEAR] KEY() for tables without primary |
| 5818 | key) |
| 5819 | Also here partition has changed and thus a new table must be |
| 5820 | created. |
| 5821 | |
| 5822 | Case IV: |
| 5823 | There was no partitioning before and no partitioning defined. |
| 5824 | Obviously no work needed. |
| 5825 | */ |
| 5826 | partition_info *tab_part_info= table->part_info; |
| 5827 | |
| 5828 | if (tab_part_info) |
| 5829 | { |
| 5830 | if (alter_info->partition_flags & ALTER_PARTITION_REMOVE) |
| 5831 | { |
| 5832 | DBUG_PRINT("info" , ("Remove partitioning" )); |
| 5833 | if (!(create_info->used_fields & HA_CREATE_USED_ENGINE)) |
| 5834 | { |
| 5835 | DBUG_PRINT("info" , ("No explicit engine used" )); |
| 5836 | create_info->db_type= tab_part_info->default_engine_type; |
| 5837 | } |
| 5838 | DBUG_PRINT("info" , ("New engine type: %s" , |
| 5839 | ha_resolve_storage_engine_name(create_info->db_type))); |
| 5840 | thd->work_part_info= NULL; |
| 5841 | *partition_changed= TRUE; |
| 5842 | } |
| 5843 | else if (!thd->work_part_info) |
| 5844 | { |
| 5845 | /* |
| 5846 | Retain partitioning but possibly with a new storage engine |
| 5847 | beneath. |
| 5848 | |
| 5849 | Create a copy of TABLE::part_info to be able to modify it freely. |
| 5850 | */ |
| 5851 | if (!(tab_part_info= tab_part_info->get_clone(thd))) |
| 5852 | DBUG_RETURN(TRUE); |
| 5853 | thd->work_part_info= tab_part_info; |
| 5854 | if (create_info->used_fields & HA_CREATE_USED_ENGINE && |
| 5855 | create_info->db_type != tab_part_info->default_engine_type) |
| 5856 | { |
| 5857 | /* |
| 5858 | Make sure change of engine happens to all partitions. |
| 5859 | */ |
| 5860 | DBUG_PRINT("info" , ("partition changed" )); |
| 5861 | if (tab_part_info->is_auto_partitioned) |
| 5862 | { |
| 5863 | /* |
| 5864 | If the user originally didn't specify partitioning to be |
| 5865 | used we can remove it now. |
| 5866 | */ |
| 5867 | thd->work_part_info= NULL; |
| 5868 | } |
| 5869 | else |
| 5870 | { |
| 5871 | /* |
| 5872 | Ensure that all partitions have the proper engine set-up |
| 5873 | */ |
| 5874 | set_engine_all_partitions(thd->work_part_info, |
| 5875 | create_info->db_type); |
| 5876 | } |
| 5877 | *partition_changed= TRUE; |
| 5878 | } |
| 5879 | } |
| 5880 | } |
| 5881 | if (thd->work_part_info) |
| 5882 | { |
| 5883 | partition_info *part_info= thd->work_part_info; |
| 5884 | bool is_native_partitioned= FALSE; |
| 5885 | /* |
| 5886 | Need to cater for engine types that can handle partition without |
| 5887 | using the partition handler. |
| 5888 | */ |
| 5889 | if (part_info != tab_part_info) |
| 5890 | { |
| 5891 | if (part_info->fix_parser_data(thd)) |
| 5892 | { |
| 5893 | goto err; |
| 5894 | } |
| 5895 | /* |
| 5896 | Compare the old and new part_info. If only key_algorithm |
| 5897 | change is done, don't consider it as changed partitioning (to avoid |
| 5898 | rebuild). This is to handle KEY (numeric_cols) partitioned tables |
| 5899 | created in 5.1. For more info, see bug#14521864. |
| 5900 | */ |
| 5901 | if (alter_info->partition_flags != ALTER_PARTITION_INFO || |
| 5902 | !table->part_info || |
| 5903 | alter_info->requested_algorithm != |
| 5904 | Alter_info::ALTER_TABLE_ALGORITHM_INPLACE || |
| 5905 | !table->part_info->has_same_partitioning(part_info)) |
| 5906 | { |
| 5907 | DBUG_PRINT("info" , ("partition changed" )); |
| 5908 | *partition_changed= true; |
| 5909 | } |
| 5910 | } |
| 5911 | /* |
| 5912 | Set up partition default_engine_type either from the create_info |
| 5913 | or from the previus table |
| 5914 | */ |
| 5915 | if (create_info->used_fields & HA_CREATE_USED_ENGINE) |
| 5916 | part_info->default_engine_type= create_info->db_type; |
| 5917 | else |
| 5918 | { |
| 5919 | if (tab_part_info) |
| 5920 | part_info->default_engine_type= tab_part_info->default_engine_type; |
| 5921 | else |
| 5922 | part_info->default_engine_type= create_info->db_type; |
| 5923 | } |
| 5924 | DBUG_ASSERT(part_info->default_engine_type && |
| 5925 | part_info->default_engine_type != partition_hton); |
| 5926 | if (check_native_partitioned(create_info, &is_native_partitioned, |
| 5927 | part_info, thd)) |
| 5928 | { |
| 5929 | goto err; |
| 5930 | } |
| 5931 | if (!is_native_partitioned) |
| 5932 | { |
| 5933 | DBUG_ASSERT(create_info->db_type); |
| 5934 | create_info->db_type= partition_hton; |
| 5935 | } |
| 5936 | } |
| 5937 | } |
| 5938 | DBUG_RETURN(FALSE); |
| 5939 | err: |
| 5940 | *fast_alter_table= false; |
| 5941 | if (saved_part_info) |
| 5942 | table->part_info= saved_part_info; |
| 5943 | DBUG_RETURN(TRUE); |
| 5944 | } |
| 5945 | |
| 5946 | |
| 5947 | /* |
| 5948 | Change partitions, used to implement ALTER TABLE ADD/REORGANIZE/COALESCE |
| 5949 | partitions. This method is used to implement both single-phase and multi- |
| 5950 | phase implementations of ADD/REORGANIZE/COALESCE partitions. |
| 5951 | |
| 5952 | SYNOPSIS |
| 5953 | mysql_change_partitions() |
| 5954 | lpt Struct containing parameters |
| 5955 | |
| 5956 | RETURN VALUES |
| 5957 | TRUE Failure |
| 5958 | FALSE Success |
| 5959 | |
| 5960 | DESCRIPTION |
| 5961 | Request handler to add partitions as set in states of the partition |
| 5962 | |
| 5963 | Elements of the lpt parameters used: |
| 5964 | create_info Create information used to create partitions |
| 5965 | db Database name |
| 5966 | table_name Table name |
| 5967 | copied Output parameter where number of copied |
| 5968 | records are added |
| 5969 | deleted Output parameter where number of deleted |
| 5970 | records are added |
| 5971 | */ |
| 5972 | |
| 5973 | static bool mysql_change_partitions(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 5974 | { |
| 5975 | char path[FN_REFLEN+1]; |
| 5976 | int error; |
| 5977 | handler *file= lpt->table->file; |
| 5978 | THD *thd= lpt->thd; |
| 5979 | DBUG_ENTER("mysql_change_partitions" ); |
| 5980 | |
| 5981 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 5982 | |
| 5983 | if(mysql_trans_prepare_alter_copy_data(thd)) |
| 5984 | DBUG_RETURN(TRUE); |
| 5985 | |
| 5986 | /* TODO: test if bulk_insert would increase the performance */ |
| 5987 | |
| 5988 | if (unlikely((error= file->ha_change_partitions(lpt->create_info, path, |
| 5989 | &lpt->copied, |
| 5990 | &lpt->deleted, |
| 5991 | lpt->pack_frm_data, |
| 5992 | lpt->pack_frm_len)))) |
| 5993 | { |
| 5994 | file->print_error(error, MYF(error != ER_OUTOFMEMORY ? 0 : ME_FATALERROR)); |
| 5995 | } |
| 5996 | |
| 5997 | if (mysql_trans_commit_alter_copy_data(thd)) |
| 5998 | error= 1; /* The error has been reported */ |
| 5999 | |
| 6000 | DBUG_RETURN(MY_TEST(error)); |
| 6001 | } |
| 6002 | |
| 6003 | |
| 6004 | /* |
| 6005 | Rename partitions in an ALTER TABLE of partitions |
| 6006 | |
| 6007 | SYNOPSIS |
| 6008 | mysql_rename_partitions() |
| 6009 | lpt Struct containing parameters |
| 6010 | |
| 6011 | RETURN VALUES |
| 6012 | TRUE Failure |
| 6013 | FALSE Success |
| 6014 | |
| 6015 | DESCRIPTION |
| 6016 | Request handler to rename partitions as set in states of the partition |
| 6017 | |
| 6018 | Parameters used: |
| 6019 | db Database name |
| 6020 | table_name Table name |
| 6021 | */ |
| 6022 | |
| 6023 | static bool mysql_rename_partitions(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6024 | { |
| 6025 | char path[FN_REFLEN+1]; |
| 6026 | int error; |
| 6027 | DBUG_ENTER("mysql_rename_partitions" ); |
| 6028 | |
| 6029 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 6030 | if (unlikely((error= lpt->table->file->ha_rename_partitions(path)))) |
| 6031 | { |
| 6032 | if (error != 1) |
| 6033 | lpt->table->file->print_error(error, MYF(0)); |
| 6034 | DBUG_RETURN(TRUE); |
| 6035 | } |
| 6036 | DBUG_RETURN(FALSE); |
| 6037 | } |
| 6038 | |
| 6039 | |
| 6040 | /* |
| 6041 | Drop partitions in an ALTER TABLE of partitions |
| 6042 | |
| 6043 | SYNOPSIS |
| 6044 | mysql_drop_partitions() |
| 6045 | lpt Struct containing parameters |
| 6046 | |
| 6047 | RETURN VALUES |
| 6048 | TRUE Failure |
| 6049 | FALSE Success |
| 6050 | DESCRIPTION |
| 6051 | Drop the partitions marked with PART_TO_BE_DROPPED state and remove |
| 6052 | those partitions from the list. |
| 6053 | |
| 6054 | Parameters used: |
| 6055 | table Table object |
| 6056 | db Database name |
| 6057 | table_name Table name |
| 6058 | */ |
| 6059 | |
| 6060 | static bool mysql_drop_partitions(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6061 | { |
| 6062 | char path[FN_REFLEN+1]; |
| 6063 | partition_info *part_info= lpt->table->part_info; |
| 6064 | List_iterator<partition_element> part_it(part_info->partitions); |
| 6065 | uint i= 0; |
| 6066 | uint remove_count= 0; |
| 6067 | int error; |
| 6068 | DBUG_ENTER("mysql_drop_partitions" ); |
| 6069 | |
| 6070 | DBUG_ASSERT(lpt->thd->mdl_context.is_lock_owner(MDL_key::TABLE, |
| 6071 | lpt->table->s->db.str, |
| 6072 | lpt->table->s->table_name.str, |
| 6073 | MDL_EXCLUSIVE)); |
| 6074 | |
| 6075 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 6076 | if ((error= lpt->table->file->ha_drop_partitions(path))) |
| 6077 | { |
| 6078 | lpt->table->file->print_error(error, MYF(0)); |
| 6079 | DBUG_RETURN(TRUE); |
| 6080 | } |
| 6081 | do |
| 6082 | { |
| 6083 | partition_element *part_elem= part_it++; |
| 6084 | if (part_elem->part_state == PART_IS_DROPPED) |
| 6085 | { |
| 6086 | part_it.remove(); |
| 6087 | remove_count++; |
| 6088 | } |
| 6089 | } while (++i < part_info->num_parts); |
| 6090 | part_info->num_parts-= remove_count; |
| 6091 | DBUG_RETURN(FALSE); |
| 6092 | } |
| 6093 | |
| 6094 | |
| 6095 | /* |
| 6096 | Insert log entry into list |
| 6097 | SYNOPSIS |
| 6098 | insert_part_info_log_entry_list() |
| 6099 | log_entry |
| 6100 | RETURN VALUES |
| 6101 | NONE |
| 6102 | */ |
| 6103 | |
| 6104 | static void insert_part_info_log_entry_list(partition_info *part_info, |
| 6105 | DDL_LOG_MEMORY_ENTRY *log_entry) |
| 6106 | { |
| 6107 | log_entry->next_active_log_entry= part_info->first_log_entry; |
| 6108 | part_info->first_log_entry= log_entry; |
| 6109 | } |
| 6110 | |
| 6111 | |
| 6112 | /* |
| 6113 | Release all log entries for this partition info struct |
| 6114 | SYNOPSIS |
| 6115 | release_part_info_log_entries() |
| 6116 | first_log_entry First log entry in list to release |
| 6117 | RETURN VALUES |
| 6118 | NONE |
| 6119 | */ |
| 6120 | |
| 6121 | static void release_part_info_log_entries(DDL_LOG_MEMORY_ENTRY *log_entry) |
| 6122 | { |
| 6123 | DBUG_ENTER("release_part_info_log_entries" ); |
| 6124 | |
| 6125 | while (log_entry) |
| 6126 | { |
| 6127 | release_ddl_log_memory_entry(log_entry); |
| 6128 | log_entry= log_entry->next_active_log_entry; |
| 6129 | } |
| 6130 | DBUG_VOID_RETURN; |
| 6131 | } |
| 6132 | |
| 6133 | |
| 6134 | /* |
| 6135 | Log an delete/rename frm file |
| 6136 | SYNOPSIS |
| 6137 | write_log_replace_delete_frm() |
| 6138 | lpt Struct for parameters |
| 6139 | next_entry Next reference to use in log record |
| 6140 | from_path Name to rename from |
| 6141 | to_path Name to rename to |
| 6142 | replace_flag TRUE if replace, else delete |
| 6143 | RETURN VALUES |
| 6144 | TRUE Error |
| 6145 | FALSE Success |
| 6146 | DESCRIPTION |
| 6147 | Support routine that writes a replace or delete of an frm file into the |
| 6148 | ddl log. It also inserts an entry that keeps track of used space into |
| 6149 | the partition info object |
| 6150 | */ |
| 6151 | |
| 6152 | static bool write_log_replace_delete_frm(ALTER_PARTITION_PARAM_TYPE *lpt, |
| 6153 | uint next_entry, |
| 6154 | const char *from_path, |
| 6155 | const char *to_path, |
| 6156 | bool replace_flag) |
| 6157 | { |
| 6158 | DDL_LOG_ENTRY ddl_log_entry; |
| 6159 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6160 | DBUG_ENTER("write_log_replace_delete_frm" ); |
| 6161 | |
| 6162 | if (replace_flag) |
| 6163 | ddl_log_entry.action_type= DDL_LOG_REPLACE_ACTION; |
| 6164 | else |
| 6165 | ddl_log_entry.action_type= DDL_LOG_DELETE_ACTION; |
| 6166 | ddl_log_entry.next_entry= next_entry; |
| 6167 | ddl_log_entry.handler_name= reg_ext; |
| 6168 | ddl_log_entry.name= to_path; |
| 6169 | if (replace_flag) |
| 6170 | ddl_log_entry.from_name= from_path; |
| 6171 | if (write_ddl_log_entry(&ddl_log_entry, &log_entry)) |
| 6172 | { |
| 6173 | DBUG_RETURN(TRUE); |
| 6174 | } |
| 6175 | insert_part_info_log_entry_list(lpt->part_info, log_entry); |
| 6176 | DBUG_RETURN(FALSE); |
| 6177 | } |
| 6178 | |
| 6179 | |
| 6180 | /* |
| 6181 | Log final partition changes in change partition |
| 6182 | SYNOPSIS |
| 6183 | write_log_changed_partitions() |
| 6184 | lpt Struct containing parameters |
| 6185 | RETURN VALUES |
| 6186 | TRUE Error |
| 6187 | FALSE Success |
| 6188 | DESCRIPTION |
| 6189 | This code is used to perform safe ADD PARTITION for HASH partitions |
| 6190 | and COALESCE for HASH partitions and REORGANIZE for any type of |
| 6191 | partitions. |
| 6192 | We prepare entries for all partitions except the reorganised partitions |
| 6193 | in REORGANIZE partition, those are handled by |
| 6194 | write_log_dropped_partitions. For those partitions that are replaced |
| 6195 | special care is needed to ensure that this is performed correctly and |
| 6196 | this requires a two-phased approach with this log as a helper for this. |
| 6197 | |
| 6198 | This code is closely intertwined with the code in rename_partitions in |
| 6199 | the partition handler. |
| 6200 | */ |
| 6201 | |
| 6202 | static bool write_log_changed_partitions(ALTER_PARTITION_PARAM_TYPE *lpt, |
| 6203 | uint *next_entry, const char *path) |
| 6204 | { |
| 6205 | DDL_LOG_ENTRY ddl_log_entry; |
| 6206 | partition_info *part_info= lpt->part_info; |
| 6207 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6208 | char tmp_path[FN_REFLEN + 1]; |
| 6209 | char normal_path[FN_REFLEN + 1]; |
| 6210 | List_iterator<partition_element> part_it(part_info->partitions); |
| 6211 | uint temp_partitions= part_info->temp_partitions.elements; |
| 6212 | uint num_elements= part_info->partitions.elements; |
| 6213 | uint i= 0; |
| 6214 | DBUG_ENTER("write_log_changed_partitions" ); |
| 6215 | |
| 6216 | do |
| 6217 | { |
| 6218 | partition_element *part_elem= part_it++; |
| 6219 | if (part_elem->part_state == PART_IS_CHANGED || |
| 6220 | (part_elem->part_state == PART_IS_ADDED && temp_partitions)) |
| 6221 | { |
| 6222 | if (part_info->is_sub_partitioned()) |
| 6223 | { |
| 6224 | List_iterator<partition_element> sub_it(part_elem->subpartitions); |
| 6225 | uint num_subparts= part_info->num_subparts; |
| 6226 | uint j= 0; |
| 6227 | do |
| 6228 | { |
| 6229 | partition_element *sub_elem= sub_it++; |
| 6230 | ddl_log_entry.next_entry= *next_entry; |
| 6231 | ddl_log_entry.handler_name= |
| 6232 | ha_resolve_storage_engine_name(sub_elem->engine_type); |
| 6233 | if (create_subpartition_name(tmp_path, sizeof(tmp_path), path, |
| 6234 | part_elem->partition_name, |
| 6235 | sub_elem->partition_name, |
| 6236 | TEMP_PART_NAME) || |
| 6237 | create_subpartition_name(normal_path, sizeof(normal_path), path, |
| 6238 | part_elem->partition_name, |
| 6239 | sub_elem->partition_name, |
| 6240 | NORMAL_PART_NAME)) |
| 6241 | DBUG_RETURN(TRUE); |
| 6242 | ddl_log_entry.name= normal_path; |
| 6243 | ddl_log_entry.from_name= tmp_path; |
| 6244 | if (part_elem->part_state == PART_IS_CHANGED) |
| 6245 | ddl_log_entry.action_type= DDL_LOG_REPLACE_ACTION; |
| 6246 | else |
| 6247 | ddl_log_entry.action_type= DDL_LOG_RENAME_ACTION; |
| 6248 | if (write_ddl_log_entry(&ddl_log_entry, &log_entry)) |
| 6249 | { |
| 6250 | DBUG_RETURN(TRUE); |
| 6251 | } |
| 6252 | *next_entry= log_entry->entry_pos; |
| 6253 | sub_elem->log_entry= log_entry; |
| 6254 | insert_part_info_log_entry_list(part_info, log_entry); |
| 6255 | } while (++j < num_subparts); |
| 6256 | } |
| 6257 | else |
| 6258 | { |
| 6259 | ddl_log_entry.next_entry= *next_entry; |
| 6260 | ddl_log_entry.handler_name= |
| 6261 | ha_resolve_storage_engine_name(part_elem->engine_type); |
| 6262 | if (create_partition_name(tmp_path, sizeof(tmp_path), path, |
| 6263 | part_elem->partition_name, TEMP_PART_NAME, |
| 6264 | TRUE) || |
| 6265 | create_partition_name(normal_path, sizeof(normal_path), path, |
| 6266 | part_elem->partition_name, NORMAL_PART_NAME, |
| 6267 | TRUE)) |
| 6268 | DBUG_RETURN(TRUE); |
| 6269 | ddl_log_entry.name= normal_path; |
| 6270 | ddl_log_entry.from_name= tmp_path; |
| 6271 | if (part_elem->part_state == PART_IS_CHANGED) |
| 6272 | ddl_log_entry.action_type= DDL_LOG_REPLACE_ACTION; |
| 6273 | else |
| 6274 | ddl_log_entry.action_type= DDL_LOG_RENAME_ACTION; |
| 6275 | if (write_ddl_log_entry(&ddl_log_entry, &log_entry)) |
| 6276 | { |
| 6277 | DBUG_RETURN(TRUE); |
| 6278 | } |
| 6279 | *next_entry= log_entry->entry_pos; |
| 6280 | part_elem->log_entry= log_entry; |
| 6281 | insert_part_info_log_entry_list(part_info, log_entry); |
| 6282 | } |
| 6283 | } |
| 6284 | } while (++i < num_elements); |
| 6285 | DBUG_RETURN(FALSE); |
| 6286 | } |
| 6287 | |
| 6288 | |
| 6289 | /* |
| 6290 | Log dropped partitions |
| 6291 | SYNOPSIS |
| 6292 | write_log_dropped_partitions() |
| 6293 | lpt Struct containing parameters |
| 6294 | RETURN VALUES |
| 6295 | TRUE Error |
| 6296 | FALSE Success |
| 6297 | */ |
| 6298 | |
| 6299 | static bool write_log_dropped_partitions(ALTER_PARTITION_PARAM_TYPE *lpt, |
| 6300 | uint *next_entry, |
| 6301 | const char *path, |
| 6302 | bool temp_list) |
| 6303 | { |
| 6304 | DDL_LOG_ENTRY ddl_log_entry; |
| 6305 | partition_info *part_info= lpt->part_info; |
| 6306 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6307 | char tmp_path[FN_REFLEN + 1]; |
| 6308 | List_iterator<partition_element> part_it(part_info->partitions); |
| 6309 | List_iterator<partition_element> temp_it(part_info->temp_partitions); |
| 6310 | uint num_temp_partitions= part_info->temp_partitions.elements; |
| 6311 | uint num_elements= part_info->partitions.elements; |
| 6312 | DBUG_ENTER("write_log_dropped_partitions" ); |
| 6313 | |
| 6314 | ddl_log_entry.action_type= DDL_LOG_DELETE_ACTION; |
| 6315 | if (temp_list) |
| 6316 | num_elements= num_temp_partitions; |
| 6317 | while (num_elements--) |
| 6318 | { |
| 6319 | partition_element *part_elem; |
| 6320 | if (temp_list) |
| 6321 | part_elem= temp_it++; |
| 6322 | else |
| 6323 | part_elem= part_it++; |
| 6324 | if (part_elem->part_state == PART_TO_BE_DROPPED || |
| 6325 | part_elem->part_state == PART_TO_BE_ADDED || |
| 6326 | part_elem->part_state == PART_CHANGED) |
| 6327 | { |
| 6328 | uint name_variant; |
| 6329 | if (part_elem->part_state == PART_CHANGED || |
| 6330 | (part_elem->part_state == PART_TO_BE_ADDED && |
| 6331 | num_temp_partitions)) |
| 6332 | name_variant= TEMP_PART_NAME; |
| 6333 | else |
| 6334 | name_variant= NORMAL_PART_NAME; |
| 6335 | if (part_info->is_sub_partitioned()) |
| 6336 | { |
| 6337 | List_iterator<partition_element> sub_it(part_elem->subpartitions); |
| 6338 | uint num_subparts= part_info->num_subparts; |
| 6339 | uint j= 0; |
| 6340 | do |
| 6341 | { |
| 6342 | partition_element *sub_elem= sub_it++; |
| 6343 | ddl_log_entry.next_entry= *next_entry; |
| 6344 | ddl_log_entry.handler_name= |
| 6345 | ha_resolve_storage_engine_name(sub_elem->engine_type); |
| 6346 | if (create_subpartition_name(tmp_path, sizeof(tmp_path), path, |
| 6347 | part_elem->partition_name, |
| 6348 | sub_elem->partition_name, name_variant)) |
| 6349 | DBUG_RETURN(TRUE); |
| 6350 | ddl_log_entry.name= tmp_path; |
| 6351 | if (write_ddl_log_entry(&ddl_log_entry, &log_entry)) |
| 6352 | { |
| 6353 | DBUG_RETURN(TRUE); |
| 6354 | } |
| 6355 | *next_entry= log_entry->entry_pos; |
| 6356 | sub_elem->log_entry= log_entry; |
| 6357 | insert_part_info_log_entry_list(part_info, log_entry); |
| 6358 | } while (++j < num_subparts); |
| 6359 | } |
| 6360 | else |
| 6361 | { |
| 6362 | ddl_log_entry.next_entry= *next_entry; |
| 6363 | ddl_log_entry.handler_name= |
| 6364 | ha_resolve_storage_engine_name(part_elem->engine_type); |
| 6365 | if (create_partition_name(tmp_path, sizeof(tmp_path), path, |
| 6366 | part_elem->partition_name, name_variant, |
| 6367 | TRUE)) |
| 6368 | DBUG_RETURN(TRUE); |
| 6369 | ddl_log_entry.name= tmp_path; |
| 6370 | if (write_ddl_log_entry(&ddl_log_entry, &log_entry)) |
| 6371 | { |
| 6372 | DBUG_RETURN(TRUE); |
| 6373 | } |
| 6374 | *next_entry= log_entry->entry_pos; |
| 6375 | part_elem->log_entry= log_entry; |
| 6376 | insert_part_info_log_entry_list(part_info, log_entry); |
| 6377 | } |
| 6378 | } |
| 6379 | } |
| 6380 | DBUG_RETURN(FALSE); |
| 6381 | } |
| 6382 | |
| 6383 | |
| 6384 | /* |
| 6385 | Set execute log entry in ddl log for this partitioned table |
| 6386 | SYNOPSIS |
| 6387 | set_part_info_exec_log_entry() |
| 6388 | part_info Partition info object |
| 6389 | exec_log_entry Log entry |
| 6390 | RETURN VALUES |
| 6391 | NONE |
| 6392 | */ |
| 6393 | |
| 6394 | static void set_part_info_exec_log_entry(partition_info *part_info, |
| 6395 | DDL_LOG_MEMORY_ENTRY *exec_log_entry) |
| 6396 | { |
| 6397 | part_info->exec_log_entry= exec_log_entry; |
| 6398 | exec_log_entry->next_active_log_entry= NULL; |
| 6399 | } |
| 6400 | |
| 6401 | |
| 6402 | /* |
| 6403 | Write the log entry to ensure that the shadow frm file is removed at |
| 6404 | crash. |
| 6405 | SYNOPSIS |
| 6406 | write_log_drop_shadow_frm() |
| 6407 | lpt Struct containing parameters |
| 6408 | install_frm Should we log action to install shadow frm or should |
| 6409 | the action be to remove the shadow frm file. |
| 6410 | RETURN VALUES |
| 6411 | TRUE Error |
| 6412 | FALSE Success |
| 6413 | DESCRIPTION |
| 6414 | Prepare an entry to the ddl log indicating a drop/install of the shadow frm |
| 6415 | file and its corresponding handler file. |
| 6416 | */ |
| 6417 | |
| 6418 | static bool write_log_drop_shadow_frm(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6419 | { |
| 6420 | partition_info *part_info= lpt->part_info; |
| 6421 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6422 | DDL_LOG_MEMORY_ENTRY *exec_log_entry= NULL; |
| 6423 | char shadow_path[FN_REFLEN + 1]; |
| 6424 | DBUG_ENTER("write_log_drop_shadow_frm" ); |
| 6425 | |
| 6426 | build_table_shadow_filename(shadow_path, sizeof(shadow_path) - 1, lpt); |
| 6427 | mysql_mutex_lock(&LOCK_gdl); |
| 6428 | if (write_log_replace_delete_frm(lpt, 0UL, NULL, |
| 6429 | (const char*)shadow_path, FALSE)) |
| 6430 | goto error; |
| 6431 | log_entry= part_info->first_log_entry; |
| 6432 | if (write_execute_ddl_log_entry(log_entry->entry_pos, |
| 6433 | FALSE, &exec_log_entry)) |
| 6434 | goto error; |
| 6435 | mysql_mutex_unlock(&LOCK_gdl); |
| 6436 | set_part_info_exec_log_entry(part_info, exec_log_entry); |
| 6437 | DBUG_RETURN(FALSE); |
| 6438 | |
| 6439 | error: |
| 6440 | release_part_info_log_entries(part_info->first_log_entry); |
| 6441 | mysql_mutex_unlock(&LOCK_gdl); |
| 6442 | part_info->first_log_entry= NULL; |
| 6443 | my_error(ER_DDL_LOG_ERROR, MYF(0)); |
| 6444 | DBUG_RETURN(TRUE); |
| 6445 | } |
| 6446 | |
| 6447 | |
| 6448 | /* |
| 6449 | Log renaming of shadow frm to real frm name and dropping of old frm |
| 6450 | SYNOPSIS |
| 6451 | write_log_rename_frm() |
| 6452 | lpt Struct containing parameters |
| 6453 | RETURN VALUES |
| 6454 | TRUE Error |
| 6455 | FALSE Success |
| 6456 | DESCRIPTION |
| 6457 | Prepare an entry to ensure that we complete the renaming of the frm |
| 6458 | file if failure occurs in the middle of the rename process. |
| 6459 | */ |
| 6460 | |
| 6461 | static bool write_log_rename_frm(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6462 | { |
| 6463 | partition_info *part_info= lpt->part_info; |
| 6464 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6465 | DDL_LOG_MEMORY_ENTRY *exec_log_entry= part_info->exec_log_entry; |
| 6466 | char path[FN_REFLEN + 1]; |
| 6467 | char shadow_path[FN_REFLEN + 1]; |
| 6468 | DDL_LOG_MEMORY_ENTRY *old_first_log_entry= part_info->first_log_entry; |
| 6469 | DBUG_ENTER("write_log_rename_frm" ); |
| 6470 | |
| 6471 | part_info->first_log_entry= NULL; |
| 6472 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 6473 | build_table_shadow_filename(shadow_path, sizeof(shadow_path) - 1, lpt); |
| 6474 | mysql_mutex_lock(&LOCK_gdl); |
| 6475 | if (write_log_replace_delete_frm(lpt, 0UL, shadow_path, path, TRUE)) |
| 6476 | goto error; |
| 6477 | log_entry= part_info->first_log_entry; |
| 6478 | part_info->frm_log_entry= log_entry; |
| 6479 | if (write_execute_ddl_log_entry(log_entry->entry_pos, |
| 6480 | FALSE, &exec_log_entry)) |
| 6481 | goto error; |
| 6482 | release_part_info_log_entries(old_first_log_entry); |
| 6483 | mysql_mutex_unlock(&LOCK_gdl); |
| 6484 | DBUG_RETURN(FALSE); |
| 6485 | |
| 6486 | error: |
| 6487 | release_part_info_log_entries(part_info->first_log_entry); |
| 6488 | mysql_mutex_unlock(&LOCK_gdl); |
| 6489 | part_info->first_log_entry= old_first_log_entry; |
| 6490 | part_info->frm_log_entry= NULL; |
| 6491 | my_error(ER_DDL_LOG_ERROR, MYF(0)); |
| 6492 | DBUG_RETURN(TRUE); |
| 6493 | } |
| 6494 | |
| 6495 | |
| 6496 | /* |
| 6497 | Write the log entries to ensure that the drop partition command is completed |
| 6498 | even in the presence of a crash. |
| 6499 | |
| 6500 | SYNOPSIS |
| 6501 | write_log_drop_partition() |
| 6502 | lpt Struct containing parameters |
| 6503 | RETURN VALUES |
| 6504 | TRUE Error |
| 6505 | FALSE Success |
| 6506 | DESCRIPTION |
| 6507 | Prepare entries to the ddl log indicating all partitions to drop and to |
| 6508 | install the shadow frm file and remove the old frm file. |
| 6509 | */ |
| 6510 | |
| 6511 | static bool write_log_drop_partition(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6512 | { |
| 6513 | partition_info *part_info= lpt->part_info; |
| 6514 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6515 | DDL_LOG_MEMORY_ENTRY *exec_log_entry= part_info->exec_log_entry; |
| 6516 | char tmp_path[FN_REFLEN + 1]; |
| 6517 | char path[FN_REFLEN + 1]; |
| 6518 | uint next_entry= 0; |
| 6519 | DDL_LOG_MEMORY_ENTRY *old_first_log_entry= part_info->first_log_entry; |
| 6520 | DBUG_ENTER("write_log_drop_partition" ); |
| 6521 | |
| 6522 | part_info->first_log_entry= NULL; |
| 6523 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 6524 | build_table_shadow_filename(tmp_path, sizeof(tmp_path) - 1, lpt); |
| 6525 | mysql_mutex_lock(&LOCK_gdl); |
| 6526 | if (write_log_dropped_partitions(lpt, &next_entry, (const char*)path, |
| 6527 | FALSE)) |
| 6528 | goto error; |
| 6529 | if (write_log_replace_delete_frm(lpt, next_entry, (const char*)tmp_path, |
| 6530 | (const char*)path, TRUE)) |
| 6531 | goto error; |
| 6532 | log_entry= part_info->first_log_entry; |
| 6533 | part_info->frm_log_entry= log_entry; |
| 6534 | if (write_execute_ddl_log_entry(log_entry->entry_pos, |
| 6535 | FALSE, &exec_log_entry)) |
| 6536 | goto error; |
| 6537 | release_part_info_log_entries(old_first_log_entry); |
| 6538 | mysql_mutex_unlock(&LOCK_gdl); |
| 6539 | DBUG_RETURN(FALSE); |
| 6540 | |
| 6541 | error: |
| 6542 | release_part_info_log_entries(part_info->first_log_entry); |
| 6543 | mysql_mutex_unlock(&LOCK_gdl); |
| 6544 | part_info->first_log_entry= old_first_log_entry; |
| 6545 | part_info->frm_log_entry= NULL; |
| 6546 | my_error(ER_DDL_LOG_ERROR, MYF(0)); |
| 6547 | DBUG_RETURN(TRUE); |
| 6548 | } |
| 6549 | |
| 6550 | |
| 6551 | /* |
| 6552 | Write the log entries to ensure that the add partition command is not |
| 6553 | executed at all if a crash before it has completed |
| 6554 | |
| 6555 | SYNOPSIS |
| 6556 | write_log_add_change_partition() |
| 6557 | lpt Struct containing parameters |
| 6558 | RETURN VALUES |
| 6559 | TRUE Error |
| 6560 | FALSE Success |
| 6561 | DESCRIPTION |
| 6562 | Prepare entries to the ddl log indicating all partitions to drop and to |
| 6563 | remove the shadow frm file. |
| 6564 | We always inject entries backwards in the list in the ddl log since we |
| 6565 | don't know the entry position until we have written it. |
| 6566 | */ |
| 6567 | |
| 6568 | static bool write_log_add_change_partition(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6569 | { |
| 6570 | partition_info *part_info= lpt->part_info; |
| 6571 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6572 | DDL_LOG_MEMORY_ENTRY *exec_log_entry= part_info->exec_log_entry; |
| 6573 | char tmp_path[FN_REFLEN + 1]; |
| 6574 | char path[FN_REFLEN + 1]; |
| 6575 | uint next_entry= 0; |
| 6576 | DDL_LOG_MEMORY_ENTRY *old_first_log_entry= part_info->first_log_entry; |
| 6577 | /* write_log_drop_shadow_frm(lpt) must have been run first */ |
| 6578 | DBUG_ASSERT(old_first_log_entry); |
| 6579 | DBUG_ENTER("write_log_add_change_partition" ); |
| 6580 | |
| 6581 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 6582 | build_table_shadow_filename(tmp_path, sizeof(tmp_path) - 1, lpt); |
| 6583 | mysql_mutex_lock(&LOCK_gdl); |
| 6584 | |
| 6585 | /* Relink the previous drop shadow frm entry */ |
| 6586 | if (old_first_log_entry) |
| 6587 | next_entry= old_first_log_entry->entry_pos; |
| 6588 | if (write_log_dropped_partitions(lpt, &next_entry, (const char*)path, |
| 6589 | FALSE)) |
| 6590 | goto error; |
| 6591 | log_entry= part_info->first_log_entry; |
| 6592 | |
| 6593 | if (write_execute_ddl_log_entry(log_entry->entry_pos, |
| 6594 | FALSE, |
| 6595 | /* Reuse the old execute ddl_log_entry */ |
| 6596 | &exec_log_entry)) |
| 6597 | goto error; |
| 6598 | mysql_mutex_unlock(&LOCK_gdl); |
| 6599 | set_part_info_exec_log_entry(part_info, exec_log_entry); |
| 6600 | DBUG_RETURN(FALSE); |
| 6601 | |
| 6602 | error: |
| 6603 | release_part_info_log_entries(part_info->first_log_entry); |
| 6604 | mysql_mutex_unlock(&LOCK_gdl); |
| 6605 | part_info->first_log_entry= old_first_log_entry; |
| 6606 | my_error(ER_DDL_LOG_ERROR, MYF(0)); |
| 6607 | DBUG_RETURN(TRUE); |
| 6608 | } |
| 6609 | |
| 6610 | |
| 6611 | /* |
| 6612 | Write description of how to complete the operation after first phase of |
| 6613 | change partitions. |
| 6614 | |
| 6615 | SYNOPSIS |
| 6616 | write_log_final_change_partition() |
| 6617 | lpt Struct containing parameters |
| 6618 | RETURN VALUES |
| 6619 | TRUE Error |
| 6620 | FALSE Success |
| 6621 | DESCRIPTION |
| 6622 | We will write log entries that specify to |
| 6623 | 1) Install the shadow frm file. |
| 6624 | 2) Remove all partitions reorganized. (To be able to reorganize a partition |
| 6625 | to the same name. Like in REORGANIZE p0 INTO (p0, p1), |
| 6626 | so that the later rename from the new p0-temporary name to p0 don't |
| 6627 | fail because the partition already exists. |
| 6628 | 3) Rename others to reflect the new naming scheme. |
| 6629 | |
| 6630 | Note that it is written in the ddl log in reverse. |
| 6631 | */ |
| 6632 | |
| 6633 | static bool write_log_final_change_partition(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6634 | { |
| 6635 | partition_info *part_info= lpt->part_info; |
| 6636 | DDL_LOG_MEMORY_ENTRY *log_entry; |
| 6637 | DDL_LOG_MEMORY_ENTRY *exec_log_entry= part_info->exec_log_entry; |
| 6638 | char path[FN_REFLEN + 1]; |
| 6639 | char shadow_path[FN_REFLEN + 1]; |
| 6640 | DDL_LOG_MEMORY_ENTRY *old_first_log_entry= part_info->first_log_entry; |
| 6641 | uint next_entry= 0; |
| 6642 | DBUG_ENTER("write_log_final_change_partition" ); |
| 6643 | |
| 6644 | /* |
| 6645 | Do not link any previous log entry. |
| 6646 | Replace the revert operations with forced retry operations. |
| 6647 | */ |
| 6648 | part_info->first_log_entry= NULL; |
| 6649 | build_table_filename(path, sizeof(path) - 1, lpt->db.str, lpt->table_name.str, "" , 0); |
| 6650 | build_table_shadow_filename(shadow_path, sizeof(shadow_path) - 1, lpt); |
| 6651 | mysql_mutex_lock(&LOCK_gdl); |
| 6652 | if (write_log_changed_partitions(lpt, &next_entry, (const char*)path)) |
| 6653 | goto error; |
| 6654 | if (write_log_dropped_partitions(lpt, &next_entry, (const char*)path, |
| 6655 | lpt->alter_info->partition_flags & |
| 6656 | ALTER_PARTITION_REORGANIZE)) |
| 6657 | goto error; |
| 6658 | if (write_log_replace_delete_frm(lpt, next_entry, shadow_path, path, TRUE)) |
| 6659 | goto error; |
| 6660 | log_entry= part_info->first_log_entry; |
| 6661 | part_info->frm_log_entry= log_entry; |
| 6662 | /* Overwrite the revert execute log entry with this retry execute entry */ |
| 6663 | if (write_execute_ddl_log_entry(log_entry->entry_pos, |
| 6664 | FALSE, &exec_log_entry)) |
| 6665 | goto error; |
| 6666 | release_part_info_log_entries(old_first_log_entry); |
| 6667 | mysql_mutex_unlock(&LOCK_gdl); |
| 6668 | DBUG_RETURN(FALSE); |
| 6669 | |
| 6670 | error: |
| 6671 | release_part_info_log_entries(part_info->first_log_entry); |
| 6672 | mysql_mutex_unlock(&LOCK_gdl); |
| 6673 | part_info->first_log_entry= old_first_log_entry; |
| 6674 | part_info->frm_log_entry= NULL; |
| 6675 | my_error(ER_DDL_LOG_ERROR, MYF(0)); |
| 6676 | DBUG_RETURN(TRUE); |
| 6677 | } |
| 6678 | |
| 6679 | |
| 6680 | /* |
| 6681 | Remove entry from ddl log and release resources for others to use |
| 6682 | |
| 6683 | SYNOPSIS |
| 6684 | write_log_completed() |
| 6685 | lpt Struct containing parameters |
| 6686 | RETURN VALUES |
| 6687 | TRUE Error |
| 6688 | FALSE Success |
| 6689 | */ |
| 6690 | |
| 6691 | static void write_log_completed(ALTER_PARTITION_PARAM_TYPE *lpt, |
| 6692 | bool dont_crash) |
| 6693 | { |
| 6694 | partition_info *part_info= lpt->part_info; |
| 6695 | DDL_LOG_MEMORY_ENTRY *log_entry= part_info->exec_log_entry; |
| 6696 | DBUG_ENTER("write_log_completed" ); |
| 6697 | |
| 6698 | DBUG_ASSERT(log_entry); |
| 6699 | mysql_mutex_lock(&LOCK_gdl); |
| 6700 | if (write_execute_ddl_log_entry(0UL, TRUE, &log_entry)) |
| 6701 | { |
| 6702 | /* |
| 6703 | Failed to write, Bad... |
| 6704 | We have completed the operation but have log records to REMOVE |
| 6705 | stuff that shouldn't be removed. What clever things could one do |
| 6706 | here? An error output was written to the error output by the |
| 6707 | above method so we don't do anything here. |
| 6708 | */ |
| 6709 | ; |
| 6710 | } |
| 6711 | release_part_info_log_entries(part_info->first_log_entry); |
| 6712 | release_part_info_log_entries(part_info->exec_log_entry); |
| 6713 | mysql_mutex_unlock(&LOCK_gdl); |
| 6714 | part_info->exec_log_entry= NULL; |
| 6715 | part_info->first_log_entry= NULL; |
| 6716 | DBUG_VOID_RETURN; |
| 6717 | } |
| 6718 | |
| 6719 | |
| 6720 | /* |
| 6721 | Release all log entries |
| 6722 | SYNOPSIS |
| 6723 | release_log_entries() |
| 6724 | part_info Partition info struct |
| 6725 | RETURN VALUES |
| 6726 | NONE |
| 6727 | */ |
| 6728 | |
| 6729 | static void release_log_entries(partition_info *part_info) |
| 6730 | { |
| 6731 | mysql_mutex_lock(&LOCK_gdl); |
| 6732 | release_part_info_log_entries(part_info->first_log_entry); |
| 6733 | release_part_info_log_entries(part_info->exec_log_entry); |
| 6734 | mysql_mutex_unlock(&LOCK_gdl); |
| 6735 | part_info->first_log_entry= NULL; |
| 6736 | part_info->exec_log_entry= NULL; |
| 6737 | } |
| 6738 | |
| 6739 | |
| 6740 | /* |
| 6741 | Final part of partition changes to handle things when under |
| 6742 | LOCK TABLES. |
| 6743 | SYNPOSIS |
| 6744 | alter_partition_lock_handling() |
| 6745 | lpt Struct carrying parameters |
| 6746 | RETURN VALUES |
| 6747 | NONE |
| 6748 | */ |
| 6749 | static void alter_partition_lock_handling(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6750 | { |
| 6751 | THD *thd= lpt->thd; |
| 6752 | |
| 6753 | if (lpt->table) |
| 6754 | { |
| 6755 | /* |
| 6756 | Remove all instances of the table and its locks and other resources. |
| 6757 | */ |
| 6758 | close_all_tables_for_name(thd, lpt->table->s, HA_EXTRA_NOT_USED, NULL); |
| 6759 | } |
| 6760 | lpt->table= 0; |
| 6761 | lpt->table_list->table= 0; |
| 6762 | if (thd->locked_tables_mode) |
| 6763 | { |
| 6764 | Diagnostics_area *stmt_da= NULL; |
| 6765 | Diagnostics_area tmp_stmt_da(true); |
| 6766 | |
| 6767 | if (unlikely(thd->is_error())) |
| 6768 | { |
| 6769 | /* reopen might fail if we have a previous error, use a temporary da. */ |
| 6770 | stmt_da= thd->get_stmt_da(); |
| 6771 | thd->set_stmt_da(&tmp_stmt_da); |
| 6772 | } |
| 6773 | |
| 6774 | if (unlikely(thd->locked_tables_list.reopen_tables(thd, false))) |
| 6775 | sql_print_warning("We failed to reacquire LOCKs in ALTER TABLE" ); |
| 6776 | |
| 6777 | if (stmt_da) |
| 6778 | thd->set_stmt_da(stmt_da); |
| 6779 | } |
| 6780 | } |
| 6781 | |
| 6782 | |
| 6783 | /** |
| 6784 | Unlock and close table before renaming and dropping partitions. |
| 6785 | |
| 6786 | @param lpt Struct carrying parameters |
| 6787 | |
| 6788 | @return Always 0. |
| 6789 | */ |
| 6790 | |
| 6791 | static int alter_close_table(ALTER_PARTITION_PARAM_TYPE *lpt) |
| 6792 | { |
| 6793 | DBUG_ENTER("alter_close_table" ); |
| 6794 | |
| 6795 | if (lpt->table->db_stat) |
| 6796 | { |
| 6797 | mysql_lock_remove(lpt->thd, lpt->thd->lock, lpt->table); |
| 6798 | lpt->table->file->ha_close(); |
| 6799 | lpt->table->db_stat= 0; // Mark file closed |
| 6800 | } |
| 6801 | DBUG_RETURN(0); |
| 6802 | } |
| 6803 | |
| 6804 | |
| 6805 | /** |
| 6806 | Handle errors for ALTER TABLE for partitioning. |
| 6807 | |
| 6808 | @param lpt Struct carrying parameters |
| 6809 | @param action_completed The action must be completed, NOT reverted |
| 6810 | @param drop_partition Partitions has not been dropped yet |
| 6811 | @param frm_install The shadow frm-file has not yet been installed |
| 6812 | @param close_table Table is still open, close it before reverting |
| 6813 | */ |
| 6814 | |
| 6815 | void handle_alter_part_error(ALTER_PARTITION_PARAM_TYPE *lpt, |
| 6816 | bool action_completed, |
| 6817 | bool drop_partition, |
| 6818 | bool frm_install, |
| 6819 | bool close_table) |
| 6820 | { |
| 6821 | partition_info *part_info= lpt->part_info; |
| 6822 | THD *thd= lpt->thd; |
| 6823 | TABLE *table= lpt->table; |
| 6824 | DBUG_ENTER("handle_alter_part_error" ); |
| 6825 | DBUG_ASSERT(table->m_needs_reopen); |
| 6826 | |
| 6827 | if (close_table) |
| 6828 | { |
| 6829 | /* |
| 6830 | All instances of this table needs to be closed. |
| 6831 | Better to do that here, than leave the cleaning up to others. |
| 6832 | Aquire EXCLUSIVE mdl lock if not already aquired. |
| 6833 | */ |
| 6834 | if (!thd->mdl_context.is_lock_owner(MDL_key::TABLE, lpt->db.str, |
| 6835 | lpt->table_name.str, |
| 6836 | MDL_EXCLUSIVE)) |
| 6837 | { |
| 6838 | if (wait_while_table_is_used(thd, table, HA_EXTRA_FORCE_REOPEN)) |
| 6839 | { |
| 6840 | /* At least remove this instance on failure */ |
| 6841 | goto err_exclusive_lock; |
| 6842 | } |
| 6843 | } |
| 6844 | /* Ensure the share is destroyed and reopened. */ |
| 6845 | if (part_info) |
| 6846 | part_info= part_info->get_clone(thd); |
| 6847 | close_all_tables_for_name(thd, table->s, HA_EXTRA_NOT_USED, NULL); |
| 6848 | } |
| 6849 | else |
| 6850 | { |
| 6851 | err_exclusive_lock: |
| 6852 | /* |
| 6853 | Temporarily remove it from the locked table list, so that it will get |
| 6854 | reopened. |
| 6855 | */ |
| 6856 | thd->locked_tables_list.unlink_from_list(thd, |
| 6857 | table->pos_in_locked_tables, |
| 6858 | false); |
| 6859 | /* |
| 6860 | Make sure that the table is unlocked, closed and removed from |
| 6861 | the table cache. |
| 6862 | */ |
| 6863 | mysql_lock_remove(thd, thd->lock, table); |
| 6864 | if (part_info) |
| 6865 | part_info= part_info->get_clone(thd); |
| 6866 | close_thread_table(thd, &thd->open_tables); |
| 6867 | lpt->table_list->table= NULL; |
| 6868 | } |
| 6869 | |
| 6870 | if (part_info->first_log_entry && |
| 6871 | execute_ddl_log_entry(thd, part_info->first_log_entry->entry_pos)) |
| 6872 | { |
| 6873 | /* |
| 6874 | We couldn't recover from error, most likely manual interaction |
| 6875 | is required. |
| 6876 | */ |
| 6877 | write_log_completed(lpt, FALSE); |
| 6878 | release_log_entries(part_info); |
| 6879 | if (!action_completed) |
| 6880 | { |
| 6881 | if (drop_partition) |
| 6882 | { |
| 6883 | /* Table is still ok, but we left a shadow frm file behind. */ |
| 6884 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, 1, |
| 6885 | "%s %s" , |
| 6886 | "Operation was unsuccessful, table is still intact," , |
| 6887 | "but it is possible that a shadow frm file was left behind" ); |
| 6888 | } |
| 6889 | else |
| 6890 | { |
| 6891 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, 1, |
| 6892 | "%s %s %s %s" , |
| 6893 | "Operation was unsuccessful, table is still intact," , |
| 6894 | "but it is possible that a shadow frm file was left behind." , |
| 6895 | "It is also possible that temporary partitions are left behind," , |
| 6896 | "these could be empty or more or less filled with records" ); |
| 6897 | } |
| 6898 | } |
| 6899 | else |
| 6900 | { |
| 6901 | if (frm_install) |
| 6902 | { |
| 6903 | /* |
| 6904 | Failed during install of shadow frm file, table isn't intact |
| 6905 | and dropped partitions are still there |
| 6906 | */ |
| 6907 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, 1, |
| 6908 | "%s %s %s" , |
| 6909 | "Failed during alter of partitions, table is no longer intact." , |
| 6910 | "The frm file is in an unknown state, and a backup" , |
| 6911 | "is required." ); |
| 6912 | } |
| 6913 | else if (drop_partition) |
| 6914 | { |
| 6915 | /* |
| 6916 | Table is ok, we have switched to new table but left dropped |
| 6917 | partitions still in their places. We remove the log records and |
| 6918 | ask the user to perform the action manually. We remove the log |
| 6919 | records and ask the user to perform the action manually. |
| 6920 | */ |
| 6921 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, 1, |
| 6922 | "%s %s" , |
| 6923 | "Failed during drop of partitions, table is intact." , |
| 6924 | "Manual drop of remaining partitions is required" ); |
| 6925 | } |
| 6926 | else |
| 6927 | { |
| 6928 | /* |
| 6929 | We failed during renaming of partitions. The table is most |
| 6930 | certainly in a very bad state so we give user warning and disable |
| 6931 | the table by writing an ancient frm version into it. |
| 6932 | */ |
| 6933 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, 1, |
| 6934 | "%s %s %s" , |
| 6935 | "Failed during renaming of partitions. We are now in a position" , |
| 6936 | "where table is not reusable" , |
| 6937 | "Table is disabled by writing ancient frm file version into it" ); |
| 6938 | } |
| 6939 | } |
| 6940 | } |
| 6941 | else |
| 6942 | { |
| 6943 | release_log_entries(part_info); |
| 6944 | if (!action_completed) |
| 6945 | { |
| 6946 | /* |
| 6947 | We hit an error before things were completed but managed |
| 6948 | to recover from the error. An error occurred and we have |
| 6949 | restored things to original so no need for further action. |
| 6950 | */ |
| 6951 | ; |
| 6952 | } |
| 6953 | else |
| 6954 | { |
| 6955 | /* |
| 6956 | We hit an error after we had completed most of the operation |
| 6957 | and were successful in a second attempt so the operation |
| 6958 | actually is successful now. We need to issue a warning that |
| 6959 | even though we reported an error the operation was successfully |
| 6960 | completed. |
| 6961 | */ |
| 6962 | push_warning_printf(thd, Sql_condition::WARN_LEVEL_WARN, 1,"%s %s" , |
| 6963 | "Operation was successfully completed by failure handling," , |
| 6964 | "after failure of normal operation" ); |
| 6965 | } |
| 6966 | } |
| 6967 | |
| 6968 | if (thd->locked_tables_mode) |
| 6969 | { |
| 6970 | Diagnostics_area *stmt_da= NULL; |
| 6971 | Diagnostics_area tmp_stmt_da(true); |
| 6972 | |
| 6973 | if (unlikely(thd->is_error())) |
| 6974 | { |
| 6975 | /* reopen might fail if we have a previous error, use a temporary da. */ |
| 6976 | stmt_da= thd->get_stmt_da(); |
| 6977 | thd->set_stmt_da(&tmp_stmt_da); |
| 6978 | } |
| 6979 | |
| 6980 | if (unlikely(thd->locked_tables_list.reopen_tables(thd, false))) |
| 6981 | sql_print_warning("We failed to reacquire LOCKs in ALTER TABLE" ); |
| 6982 | |
| 6983 | if (stmt_da) |
| 6984 | thd->set_stmt_da(stmt_da); |
| 6985 | } |
| 6986 | |
| 6987 | DBUG_VOID_RETURN; |
| 6988 | } |
| 6989 | |
| 6990 | |
| 6991 | /** |
| 6992 | Downgrade an exclusive MDL lock if under LOCK TABLE. |
| 6993 | |
| 6994 | If we don't downgrade the lock, it will not be downgraded or released |
| 6995 | until the table is unlocked, resulting in blocking other threads using |
| 6996 | the table. |
| 6997 | */ |
| 6998 | |
| 6999 | static void downgrade_mdl_if_lock_tables_mode(THD *thd, MDL_ticket *ticket, |
| 7000 | enum_mdl_type type) |
| 7001 | { |
| 7002 | if (thd->locked_tables_mode) |
| 7003 | ticket->downgrade_lock(type); |
| 7004 | } |
| 7005 | |
| 7006 | |
| 7007 | /** |
| 7008 | Actually perform the change requested by ALTER TABLE of partitions |
| 7009 | previously prepared. |
| 7010 | |
| 7011 | @param thd Thread object |
| 7012 | @param table Original table object with new part_info |
| 7013 | @param alter_info ALTER TABLE info |
| 7014 | @param create_info Create info for CREATE TABLE |
| 7015 | @param table_list List of the table involved |
| 7016 | @param db Database name of new table |
| 7017 | @param table_name Table name of new table |
| 7018 | |
| 7019 | @return Operation status |
| 7020 | @retval TRUE Error |
| 7021 | @retval FALSE Success |
| 7022 | |
| 7023 | @note |
| 7024 | Perform all ALTER TABLE operations for partitioned tables that can be |
| 7025 | performed fast without a full copy of the original table. |
| 7026 | */ |
| 7027 | |
| 7028 | uint fast_alter_partition_table(THD *thd, TABLE *table, |
| 7029 | Alter_info *alter_info, |
| 7030 | HA_CREATE_INFO *create_info, |
| 7031 | TABLE_LIST *table_list, |
| 7032 | const LEX_CSTRING *db, |
| 7033 | const LEX_CSTRING *table_name) |
| 7034 | { |
| 7035 | /* Set-up struct used to write frm files */ |
| 7036 | partition_info *part_info; |
| 7037 | ALTER_PARTITION_PARAM_TYPE lpt_obj; |
| 7038 | ALTER_PARTITION_PARAM_TYPE *lpt= &lpt_obj; |
| 7039 | bool action_completed= FALSE; |
| 7040 | bool close_table_on_failure= FALSE; |
| 7041 | bool frm_install= FALSE; |
| 7042 | MDL_ticket *mdl_ticket= table->mdl_ticket; |
| 7043 | DBUG_ENTER("fast_alter_partition_table" ); |
| 7044 | DBUG_ASSERT(table->m_needs_reopen); |
| 7045 | |
| 7046 | part_info= table->part_info; |
| 7047 | lpt->thd= thd; |
| 7048 | lpt->table_list= table_list; |
| 7049 | lpt->part_info= part_info; |
| 7050 | lpt->alter_info= alter_info; |
| 7051 | lpt->create_info= create_info; |
| 7052 | lpt->db_options= create_info->table_options_with_row_type(); |
| 7053 | lpt->table= table; |
| 7054 | lpt->key_info_buffer= 0; |
| 7055 | lpt->key_count= 0; |
| 7056 | lpt->db= *db; |
| 7057 | lpt->table_name= *table_name; |
| 7058 | lpt->copied= 0; |
| 7059 | lpt->deleted= 0; |
| 7060 | lpt->pack_frm_data= NULL; |
| 7061 | lpt->pack_frm_len= 0; |
| 7062 | |
| 7063 | if (table->file->alter_table_flags(alter_info->flags) & |
| 7064 | HA_PARTITION_ONE_PHASE) |
| 7065 | { |
| 7066 | /* |
| 7067 | In the case where the engine supports one phase online partition |
| 7068 | changes it is not necessary to have any exclusive locks. The |
| 7069 | correctness is upheld instead by transactions being aborted if they |
| 7070 | access the table after its partition definition has changed (if they |
| 7071 | are still using the old partition definition). |
| 7072 | |
| 7073 | The handler is in this case responsible to ensure that all users |
| 7074 | start using the new frm file after it has changed. To implement |
| 7075 | one phase it is necessary for the handler to have the master copy |
| 7076 | of the frm file and use discovery mechanisms to renew it. Thus |
| 7077 | write frm will write the frm, pack the new frm and finally |
| 7078 | the frm is deleted and the discovery mechanisms will either restore |
| 7079 | back to the old or installing the new after the change is activated. |
| 7080 | |
| 7081 | Thus all open tables will be discovered that they are old, if not |
| 7082 | earlier as soon as they try an operation using the old table. One |
| 7083 | should ensure that this is checked already when opening a table, |
| 7084 | even if it is found in the cache of open tables. |
| 7085 | |
| 7086 | change_partitions will perform all operations and it is the duty of |
| 7087 | the handler to ensure that the frm files in the system gets updated |
| 7088 | in synch with the changes made and if an error occurs that a proper |
| 7089 | error handling is done. |
| 7090 | |
| 7091 | If the MySQL Server crashes at this moment but the handler succeeds |
| 7092 | in performing the change then the binlog is not written for the |
| 7093 | change. There is no way to solve this as long as the binlog is not |
| 7094 | transactional and even then it is hard to solve it completely. |
| 7095 | |
| 7096 | The first approach here was to downgrade locks. Now a different approach |
| 7097 | is decided upon. The idea is that the handler will have access to the |
| 7098 | Alter_info when store_lock arrives with TL_WRITE_ALLOW_READ. So if the |
| 7099 | handler knows that this functionality can be handled with a lower lock |
| 7100 | level it will set the lock level to TL_WRITE_ALLOW_WRITE immediately. |
| 7101 | Thus the need to downgrade the lock disappears. |
| 7102 | 1) Write the new frm, pack it and then delete it |
| 7103 | 2) Perform the change within the handler |
| 7104 | */ |
| 7105 | if (mysql_write_frm(lpt, WFRM_WRITE_SHADOW) || |
| 7106 | mysql_change_partitions(lpt)) |
| 7107 | { |
| 7108 | goto err; |
| 7109 | } |
| 7110 | } |
| 7111 | else if (alter_info->partition_flags & ALTER_PARTITION_DROP) |
| 7112 | { |
| 7113 | /* |
| 7114 | Now after all checks and setting state on dropped partitions we can |
| 7115 | start the actual dropping of the partitions. |
| 7116 | |
| 7117 | Drop partition is actually two things happening. The first is that |
| 7118 | a lot of records are deleted. The second is that the behaviour of |
| 7119 | subsequent updates and writes and deletes will change. The delete |
| 7120 | part can be handled without any particular high lock level by |
| 7121 | transactional engines whereas non-transactional engines need to |
| 7122 | ensure that this change is done with an exclusive lock on the table. |
| 7123 | The second part, the change of partitioning does however require |
| 7124 | an exclusive lock to install the new partitioning as one atomic |
| 7125 | operation. If this is not the case, it is possible for two |
| 7126 | transactions to see the change in a different order than their |
| 7127 | serialisation order. Thus we need an exclusive lock for both |
| 7128 | transactional and non-transactional engines. |
| 7129 | |
| 7130 | For LIST partitions it could be possible to avoid the exclusive lock |
| 7131 | (and for RANGE partitions if they didn't rearrange range definitions |
| 7132 | after a DROP PARTITION) if one ensured that failed accesses to the |
| 7133 | dropped partitions was aborted for sure (thus only possible for |
| 7134 | transactional engines). |
| 7135 | |
| 7136 | 0) Write an entry that removes the shadow frm file if crash occurs |
| 7137 | 1) Write the new frm file as a shadow frm |
| 7138 | 2) Get an exclusive metadata lock on the table (waits for all active |
| 7139 | transactions using this table). This ensures that we |
| 7140 | can release all other locks on the table and since no one can open |
| 7141 | the table, there can be no new threads accessing the table. They |
| 7142 | will be hanging on this exclusive lock. |
| 7143 | 3) Write the ddl log to ensure that the operation is completed |
| 7144 | even in the presence of a MySQL Server crash (the log is executed |
| 7145 | before any other threads are started, so there are no locking issues). |
| 7146 | 4) Close the table that have already been opened but didn't stumble on |
| 7147 | the abort locked previously. This is done as part of the |
| 7148 | alter_close_table call. |
| 7149 | 5) Write the bin log |
| 7150 | Unfortunately the writing of the binlog is not synchronised with |
| 7151 | other logging activities. So no matter in which order the binlog |
| 7152 | is written compared to other activities there will always be cases |
| 7153 | where crashes make strange things occur. In this placement it can |
| 7154 | happen that the ALTER TABLE DROP PARTITION gets performed in the |
| 7155 | master but not in the slaves if we have a crash, after writing the |
| 7156 | ddl log but before writing the binlog. A solution to this would |
| 7157 | require writing the statement first in the ddl log and then |
| 7158 | when recovering from the crash read the binlog and insert it into |
| 7159 | the binlog if not written already. |
| 7160 | 6) Install the previously written shadow frm file |
| 7161 | 7) Prepare handlers for drop of partitions |
| 7162 | 8) Drop the partitions |
| 7163 | 9) Remove entries from ddl log |
| 7164 | 10) Reopen table if under lock tables |
| 7165 | 11) Complete query |
| 7166 | |
| 7167 | We insert Error injections at all places where it could be interesting |
| 7168 | to test if recovery is properly done. |
| 7169 | */ |
| 7170 | if (write_log_drop_shadow_frm(lpt) || |
| 7171 | ERROR_INJECT_CRASH("crash_drop_partition_1" ) || |
| 7172 | ERROR_INJECT_ERROR("fail_drop_partition_1" ) || |
| 7173 | mysql_write_frm(lpt, WFRM_WRITE_SHADOW) || |
| 7174 | ERROR_INJECT_CRASH("crash_drop_partition_2" ) || |
| 7175 | ERROR_INJECT_ERROR("fail_drop_partition_2" ) || |
| 7176 | wait_while_table_is_used(thd, table, HA_EXTRA_NOT_USED) || |
| 7177 | ERROR_INJECT_CRASH("crash_drop_partition_3" ) || |
| 7178 | ERROR_INJECT_ERROR("fail_drop_partition_3" ) || |
| 7179 | (close_table_on_failure= TRUE, FALSE) || |
| 7180 | write_log_drop_partition(lpt) || |
| 7181 | (action_completed= TRUE, FALSE) || |
| 7182 | ERROR_INJECT_CRASH("crash_drop_partition_4" ) || |
| 7183 | ERROR_INJECT_ERROR("fail_drop_partition_4" ) || |
| 7184 | alter_close_table(lpt) || |
| 7185 | (close_table_on_failure= FALSE, FALSE) || |
| 7186 | ERROR_INJECT_CRASH("crash_drop_partition_5" ) || |
| 7187 | ERROR_INJECT_ERROR("fail_drop_partition_5" ) || |
| 7188 | ((!thd->lex->no_write_to_binlog) && |
| 7189 | (write_bin_log(thd, FALSE, |
| 7190 | thd->query(), thd->query_length()), FALSE)) || |
| 7191 | ERROR_INJECT_CRASH("crash_drop_partition_6" ) || |
| 7192 | ERROR_INJECT_ERROR("fail_drop_partition_6" ) || |
| 7193 | (frm_install= TRUE, FALSE) || |
| 7194 | mysql_write_frm(lpt, WFRM_INSTALL_SHADOW) || |
| 7195 | (frm_install= FALSE, FALSE) || |
| 7196 | ERROR_INJECT_CRASH("crash_drop_partition_7" ) || |
| 7197 | ERROR_INJECT_ERROR("fail_drop_partition_7" ) || |
| 7198 | mysql_drop_partitions(lpt) || |
| 7199 | ERROR_INJECT_CRASH("crash_drop_partition_8" ) || |
| 7200 | ERROR_INJECT_ERROR("fail_drop_partition_8" ) || |
| 7201 | (write_log_completed(lpt, FALSE), FALSE) || |
| 7202 | ERROR_INJECT_CRASH("crash_drop_partition_9" ) || |
| 7203 | ERROR_INJECT_ERROR("fail_drop_partition_9" ) || |
| 7204 | (alter_partition_lock_handling(lpt), FALSE)) |
| 7205 | { |
| 7206 | handle_alter_part_error(lpt, action_completed, TRUE, frm_install, |
| 7207 | close_table_on_failure); |
| 7208 | goto err; |
| 7209 | } |
| 7210 | } |
| 7211 | else if ((alter_info->partition_flags & ALTER_PARTITION_ADD) && |
| 7212 | (part_info->part_type == RANGE_PARTITION || |
| 7213 | part_info->part_type == LIST_PARTITION)) |
| 7214 | { |
| 7215 | /* |
| 7216 | ADD RANGE/LIST PARTITIONS |
| 7217 | In this case there are no tuples removed and no tuples are added. |
| 7218 | Thus the operation is merely adding a new partition. Thus it is |
| 7219 | necessary to perform the change as an atomic operation. Otherwise |
| 7220 | someone reading without seeing the new partition could potentially |
| 7221 | miss updates made by a transaction serialised before it that are |
| 7222 | inserted into the new partition. |
| 7223 | |
| 7224 | 0) Write an entry that removes the shadow frm file if crash occurs |
| 7225 | 1) Write the new frm file as a shadow frm file |
| 7226 | 2) Get an exclusive metadata lock on the table (waits for all active |
| 7227 | transactions using this table). This ensures that we |
| 7228 | can release all other locks on the table and since no one can open |
| 7229 | the table, there can be no new threads accessing the table. They |
| 7230 | will be hanging on this exclusive lock. |
| 7231 | 3) Write an entry to remove the new parttions if crash occurs |
| 7232 | 4) Add the new partitions. |
| 7233 | 5) Close all instances of the table and remove them from the table cache. |
| 7234 | 6) Write binlog |
| 7235 | 7) Now the change is completed except for the installation of the |
| 7236 | new frm file. We thus write an action in the log to change to |
| 7237 | the shadow frm file |
| 7238 | 8) Install the new frm file of the table where the partitions are |
| 7239 | added to the table. |
| 7240 | 9) Remove entries from ddl log |
| 7241 | 10)Reopen tables if under lock tables |
| 7242 | 11)Complete query |
| 7243 | */ |
| 7244 | if (write_log_drop_shadow_frm(lpt) || |
| 7245 | ERROR_INJECT_CRASH("crash_add_partition_1" ) || |
| 7246 | ERROR_INJECT_ERROR("fail_add_partition_1" ) || |
| 7247 | mysql_write_frm(lpt, WFRM_WRITE_SHADOW) || |
| 7248 | ERROR_INJECT_CRASH("crash_add_partition_2" ) || |
| 7249 | ERROR_INJECT_ERROR("fail_add_partition_2" ) || |
| 7250 | wait_while_table_is_used(thd, table, HA_EXTRA_NOT_USED) || |
| 7251 | ERROR_INJECT_CRASH("crash_add_partition_3" ) || |
| 7252 | ERROR_INJECT_ERROR("fail_add_partition_3" ) || |
| 7253 | (close_table_on_failure= TRUE, FALSE) || |
| 7254 | write_log_add_change_partition(lpt) || |
| 7255 | ERROR_INJECT_CRASH("crash_add_partition_4" ) || |
| 7256 | ERROR_INJECT_ERROR("fail_add_partition_4" ) || |
| 7257 | mysql_change_partitions(lpt) || |
| 7258 | ERROR_INJECT_CRASH("crash_add_partition_5" ) || |
| 7259 | ERROR_INJECT_ERROR("fail_add_partition_5" ) || |
| 7260 | (close_table_on_failure= FALSE, FALSE) || |
| 7261 | alter_close_table(lpt) || |
| 7262 | ERROR_INJECT_CRASH("crash_add_partition_6" ) || |
| 7263 | ERROR_INJECT_ERROR("fail_add_partition_6" ) || |
| 7264 | ((!thd->lex->no_write_to_binlog) && |
| 7265 | (write_bin_log(thd, FALSE, |
| 7266 | thd->query(), thd->query_length()), FALSE)) || |
| 7267 | ERROR_INJECT_CRASH("crash_add_partition_7" ) || |
| 7268 | ERROR_INJECT_ERROR("fail_add_partition_7" ) || |
| 7269 | write_log_rename_frm(lpt) || |
| 7270 | (action_completed= TRUE, FALSE) || |
| 7271 | ERROR_INJECT_CRASH("crash_add_partition_8" ) || |
| 7272 | ERROR_INJECT_ERROR("fail_add_partition_8" ) || |
| 7273 | (frm_install= TRUE, FALSE) || |
| 7274 | mysql_write_frm(lpt, WFRM_INSTALL_SHADOW) || |
| 7275 | (frm_install= FALSE, FALSE) || |
| 7276 | ERROR_INJECT_CRASH("crash_add_partition_9" ) || |
| 7277 | ERROR_INJECT_ERROR("fail_add_partition_9" ) || |
| 7278 | (write_log_completed(lpt, FALSE), FALSE) || |
| 7279 | ERROR_INJECT_CRASH("crash_add_partition_10" ) || |
| 7280 | ERROR_INJECT_ERROR("fail_add_partition_10" ) || |
| 7281 | (alter_partition_lock_handling(lpt), FALSE)) |
| 7282 | { |
| 7283 | handle_alter_part_error(lpt, action_completed, FALSE, frm_install, |
| 7284 | close_table_on_failure); |
| 7285 | goto err; |
| 7286 | } |
| 7287 | } |
| 7288 | else |
| 7289 | { |
| 7290 | /* |
| 7291 | ADD HASH PARTITION/ |
| 7292 | COALESCE PARTITION/ |
| 7293 | REBUILD PARTITION/ |
| 7294 | REORGANIZE PARTITION |
| 7295 | |
| 7296 | In this case all records are still around after the change although |
| 7297 | possibly organised into new partitions, thus by ensuring that all |
| 7298 | updates go to both the old and the new partitioning scheme we can |
| 7299 | actually perform this operation lock-free. The only exception to |
| 7300 | this is when REORGANIZE PARTITION adds/drops ranges. In this case |
| 7301 | there needs to be an exclusive lock during the time when the range |
| 7302 | changes occur. |
| 7303 | This is only possible if the handler can ensure double-write for a |
| 7304 | period. The double write will ensure that it doesn't matter where the |
| 7305 | data is read from since both places are updated for writes. If such |
| 7306 | double writing is not performed then it is necessary to perform the |
| 7307 | change with the usual exclusive lock. With double writes it is even |
| 7308 | possible to perform writes in parallel with the reorganisation of |
| 7309 | partitions. |
| 7310 | |
| 7311 | Without double write procedure we get the following procedure. |
| 7312 | The only difference with using double write is that we can downgrade |
| 7313 | the lock to TL_WRITE_ALLOW_WRITE. Double write in this case only |
| 7314 | double writes from old to new. If we had double writing in both |
| 7315 | directions we could perform the change completely without exclusive |
| 7316 | lock for HASH partitions. |
| 7317 | Handlers that perform double writing during the copy phase can actually |
| 7318 | use a lower lock level. This can be handled inside store_lock in the |
| 7319 | respective handler. |
| 7320 | |
| 7321 | 0) Write an entry that removes the shadow frm file if crash occurs. |
| 7322 | 1) Write the shadow frm file of new partitioning. |
| 7323 | 2) Log such that temporary partitions added in change phase are |
| 7324 | removed in a crash situation. |
| 7325 | 3) Add the new partitions. |
| 7326 | Copy from the reorganised partitions to the new partitions. |
| 7327 | 4) Get an exclusive metadata lock on the table (waits for all active |
| 7328 | transactions using this table). This ensures that we |
| 7329 | can release all other locks on the table and since no one can open |
| 7330 | the table, there can be no new threads accessing the table. They |
| 7331 | will be hanging on this exclusive lock. |
| 7332 | 5) Close the table. |
| 7333 | 6) Log that operation is completed and log all complete actions |
| 7334 | needed to complete operation from here. |
| 7335 | 7) Write bin log. |
| 7336 | 8) Prepare handlers for rename and delete of partitions. |
| 7337 | 9) Rename and drop the reorged partitions such that they are no |
| 7338 | longer used and rename those added to their real new names. |
| 7339 | 10) Install the shadow frm file. |
| 7340 | 11) Reopen the table if under lock tables. |
| 7341 | 12) Complete query. |
| 7342 | */ |
| 7343 | if (write_log_drop_shadow_frm(lpt) || |
| 7344 | ERROR_INJECT_CRASH("crash_change_partition_1" ) || |
| 7345 | ERROR_INJECT_ERROR("fail_change_partition_1" ) || |
| 7346 | mysql_write_frm(lpt, WFRM_WRITE_SHADOW) || |
| 7347 | ERROR_INJECT_CRASH("crash_change_partition_2" ) || |
| 7348 | ERROR_INJECT_ERROR("fail_change_partition_2" ) || |
| 7349 | (close_table_on_failure= TRUE, FALSE) || |
| 7350 | write_log_add_change_partition(lpt) || |
| 7351 | ERROR_INJECT_CRASH("crash_change_partition_3" ) || |
| 7352 | ERROR_INJECT_ERROR("fail_change_partition_3" ) || |
| 7353 | mysql_change_partitions(lpt) || |
| 7354 | ERROR_INJECT_CRASH("crash_change_partition_4" ) || |
| 7355 | ERROR_INJECT_ERROR("fail_change_partition_4" ) || |
| 7356 | wait_while_table_is_used(thd, table, HA_EXTRA_NOT_USED) || |
| 7357 | ERROR_INJECT_CRASH("crash_change_partition_5" ) || |
| 7358 | ERROR_INJECT_ERROR("fail_change_partition_5" ) || |
| 7359 | alter_close_table(lpt) || |
| 7360 | (close_table_on_failure= FALSE, FALSE) || |
| 7361 | ERROR_INJECT_CRASH("crash_change_partition_6" ) || |
| 7362 | ERROR_INJECT_ERROR("fail_change_partition_6" ) || |
| 7363 | write_log_final_change_partition(lpt) || |
| 7364 | (action_completed= TRUE, FALSE) || |
| 7365 | ERROR_INJECT_CRASH("crash_change_partition_7" ) || |
| 7366 | ERROR_INJECT_ERROR("fail_change_partition_7" ) || |
| 7367 | ((!thd->lex->no_write_to_binlog) && |
| 7368 | (write_bin_log(thd, FALSE, |
| 7369 | thd->query(), thd->query_length()), FALSE)) || |
| 7370 | ERROR_INJECT_CRASH("crash_change_partition_8" ) || |
| 7371 | ERROR_INJECT_ERROR("fail_change_partition_8" ) || |
| 7372 | ((frm_install= TRUE), FALSE) || |
| 7373 | mysql_write_frm(lpt, WFRM_INSTALL_SHADOW) || |
| 7374 | (frm_install= FALSE, FALSE) || |
| 7375 | ERROR_INJECT_CRASH("crash_change_partition_9" ) || |
| 7376 | ERROR_INJECT_ERROR("fail_change_partition_9" ) || |
| 7377 | mysql_drop_partitions(lpt) || |
| 7378 | ERROR_INJECT_CRASH("crash_change_partition_10" ) || |
| 7379 | ERROR_INJECT_ERROR("fail_change_partition_10" ) || |
| 7380 | mysql_rename_partitions(lpt) || |
| 7381 | ERROR_INJECT_CRASH("crash_change_partition_11" ) || |
| 7382 | ERROR_INJECT_ERROR("fail_change_partition_11" ) || |
| 7383 | (write_log_completed(lpt, FALSE), FALSE) || |
| 7384 | ERROR_INJECT_CRASH("crash_change_partition_12" ) || |
| 7385 | ERROR_INJECT_ERROR("fail_change_partition_12" ) || |
| 7386 | (alter_partition_lock_handling(lpt), FALSE)) |
| 7387 | { |
| 7388 | handle_alter_part_error(lpt, action_completed, FALSE, frm_install, |
| 7389 | close_table_on_failure); |
| 7390 | goto err; |
| 7391 | } |
| 7392 | } |
| 7393 | downgrade_mdl_if_lock_tables_mode(thd, mdl_ticket, MDL_SHARED_NO_READ_WRITE); |
| 7394 | /* |
| 7395 | A final step is to write the query to the binlog and send ok to the |
| 7396 | user |
| 7397 | */ |
| 7398 | DBUG_RETURN(fast_end_partition(thd, lpt->copied, lpt->deleted, table_list)); |
| 7399 | err: |
| 7400 | downgrade_mdl_if_lock_tables_mode(thd, mdl_ticket, MDL_SHARED_NO_READ_WRITE); |
| 7401 | DBUG_RETURN(TRUE); |
| 7402 | } |
| 7403 | #endif |
| 7404 | |
| 7405 | |
| 7406 | /* |
| 7407 | Prepare for calling val_int on partition function by setting fields to |
| 7408 | point to the record where the values of the PF-fields are stored. |
| 7409 | |
| 7410 | SYNOPSIS |
| 7411 | set_field_ptr() |
| 7412 | ptr Array of fields to change ptr |
| 7413 | new_buf New record pointer |
| 7414 | old_buf Old record pointer |
| 7415 | |
| 7416 | DESCRIPTION |
| 7417 | Set ptr in field objects of field array to refer to new_buf record |
| 7418 | instead of previously old_buf. Used before calling val_int and after |
| 7419 | it is used to restore pointers to table->record[0]. |
| 7420 | This routine is placed outside of partition code since it can be useful |
| 7421 | also for other programs. |
| 7422 | */ |
| 7423 | |
| 7424 | void set_field_ptr(Field **ptr, const uchar *new_buf, |
| 7425 | const uchar *old_buf) |
| 7426 | { |
| 7427 | my_ptrdiff_t diff= (new_buf - old_buf); |
| 7428 | DBUG_ENTER("set_field_ptr" ); |
| 7429 | |
| 7430 | do |
| 7431 | { |
| 7432 | (*ptr)->move_field_offset(diff); |
| 7433 | } while (*(++ptr)); |
| 7434 | DBUG_VOID_RETURN; |
| 7435 | } |
| 7436 | |
| 7437 | |
| 7438 | /* |
| 7439 | Prepare for calling val_int on partition function by setting fields to |
| 7440 | point to the record where the values of the PF-fields are stored. |
| 7441 | This variant works on a key_part reference. |
| 7442 | It is not required that all fields are NOT NULL fields. |
| 7443 | |
| 7444 | SYNOPSIS |
| 7445 | set_key_field_ptr() |
| 7446 | key_info key info with a set of fields to change ptr |
| 7447 | new_buf New record pointer |
| 7448 | old_buf Old record pointer |
| 7449 | |
| 7450 | DESCRIPTION |
| 7451 | Set ptr in field objects of field array to refer to new_buf record |
| 7452 | instead of previously old_buf. Used before calling val_int and after |
| 7453 | it is used to restore pointers to table->record[0]. |
| 7454 | This routine is placed outside of partition code since it can be useful |
| 7455 | also for other programs. |
| 7456 | */ |
| 7457 | |
| 7458 | void set_key_field_ptr(KEY *key_info, const uchar *new_buf, |
| 7459 | const uchar *old_buf) |
| 7460 | { |
| 7461 | KEY_PART_INFO *key_part= key_info->key_part; |
| 7462 | uint key_parts= key_info->user_defined_key_parts; |
| 7463 | uint i= 0; |
| 7464 | my_ptrdiff_t diff= (new_buf - old_buf); |
| 7465 | DBUG_ENTER("set_key_field_ptr" ); |
| 7466 | |
| 7467 | do |
| 7468 | { |
| 7469 | key_part->field->move_field_offset(diff); |
| 7470 | key_part++; |
| 7471 | } while (++i < key_parts); |
| 7472 | DBUG_VOID_RETURN; |
| 7473 | } |
| 7474 | |
| 7475 | |
| 7476 | /** |
| 7477 | Append all fields in read_set to string |
| 7478 | |
| 7479 | @param[in,out] str String to append to. |
| 7480 | @param[in] row Row to append. |
| 7481 | @param[in] table Table containing read_set and fields for the row. |
| 7482 | */ |
| 7483 | void append_row_to_str(String &str, const uchar *row, TABLE *table) |
| 7484 | { |
| 7485 | Field **fields, **field_ptr; |
| 7486 | const uchar *rec; |
| 7487 | uint num_fields= bitmap_bits_set(table->read_set); |
| 7488 | uint curr_field_index= 0; |
| 7489 | bool is_rec0= !row || row == table->record[0]; |
| 7490 | if (!row) |
| 7491 | rec= table->record[0]; |
| 7492 | else |
| 7493 | rec= row; |
| 7494 | |
| 7495 | /* Create a new array of all read fields. */ |
| 7496 | fields= (Field**) my_malloc(sizeof(void*) * (num_fields + 1), |
| 7497 | MYF(0)); |
| 7498 | if (!fields) |
| 7499 | return; |
| 7500 | fields[num_fields]= NULL; |
| 7501 | for (field_ptr= table->field; |
| 7502 | *field_ptr; |
| 7503 | field_ptr++) |
| 7504 | { |
| 7505 | if (!bitmap_is_set(table->read_set, (*field_ptr)->field_index)) |
| 7506 | continue; |
| 7507 | fields[curr_field_index++]= *field_ptr; |
| 7508 | } |
| 7509 | |
| 7510 | |
| 7511 | if (!is_rec0) |
| 7512 | set_field_ptr(fields, rec, table->record[0]); |
| 7513 | |
| 7514 | for (field_ptr= fields; |
| 7515 | *field_ptr; |
| 7516 | field_ptr++) |
| 7517 | { |
| 7518 | Field *field= *field_ptr; |
| 7519 | str.append(" " ); |
| 7520 | str.append(&field->field_name); |
| 7521 | str.append(":" ); |
| 7522 | field_unpack(&str, field, rec, 0, false); |
| 7523 | } |
| 7524 | |
| 7525 | if (!is_rec0) |
| 7526 | set_field_ptr(fields, table->record[0], rec); |
| 7527 | my_free(fields); |
| 7528 | } |
| 7529 | |
| 7530 | |
| 7531 | #ifdef WITH_PARTITION_STORAGE_ENGINE |
| 7532 | /** |
| 7533 | Return comma-separated list of used partitions in the provided given string. |
| 7534 | |
| 7535 | @param mem_root Where to allocate following list |
| 7536 | @param part_info Partitioning info |
| 7537 | @param[out] parts The resulting list of string to fill |
| 7538 | @param[out] used_partitions_list result list to fill |
| 7539 | |
| 7540 | Generate a list of used partitions (from bits in part_info->read_partitions |
| 7541 | bitmap), and store it into the provided String object. |
| 7542 | |
| 7543 | @note |
| 7544 | The produced string must not be longer then MAX_PARTITIONS * (1 + FN_LEN). |
| 7545 | In case of UPDATE, only the partitions read is given, not the partitions |
| 7546 | that was written or locked. |
| 7547 | */ |
| 7548 | |
| 7549 | void make_used_partitions_str(MEM_ROOT *alloc, |
| 7550 | partition_info *part_info, |
| 7551 | String *parts_str, |
| 7552 | String_list &used_partitions_list) |
| 7553 | { |
| 7554 | parts_str->length(0); |
| 7555 | partition_element *pe; |
| 7556 | uint partition_id= 0; |
| 7557 | List_iterator<partition_element> it(part_info->partitions); |
| 7558 | |
| 7559 | if (part_info->is_sub_partitioned()) |
| 7560 | { |
| 7561 | partition_element *head_pe; |
| 7562 | while ((head_pe= it++)) |
| 7563 | { |
| 7564 | List_iterator<partition_element> it2(head_pe->subpartitions); |
| 7565 | while ((pe= it2++)) |
| 7566 | { |
| 7567 | if (bitmap_is_set(&part_info->read_partitions, partition_id)) |
| 7568 | { |
| 7569 | if (parts_str->length()) |
| 7570 | parts_str->append(','); |
| 7571 | uint index= parts_str->length(); |
| 7572 | parts_str->append(head_pe->partition_name, |
| 7573 | strlen(head_pe->partition_name), |
| 7574 | system_charset_info); |
| 7575 | parts_str->append('_'); |
| 7576 | parts_str->append(pe->partition_name, |
| 7577 | strlen(pe->partition_name), |
| 7578 | system_charset_info); |
| 7579 | used_partitions_list.append_str(alloc, parts_str->ptr() + index); |
| 7580 | } |
| 7581 | partition_id++; |
| 7582 | } |
| 7583 | } |
| 7584 | } |
| 7585 | else |
| 7586 | { |
| 7587 | while ((pe= it++)) |
| 7588 | { |
| 7589 | if (bitmap_is_set(&part_info->read_partitions, partition_id)) |
| 7590 | { |
| 7591 | if (parts_str->length()) |
| 7592 | parts_str->append(','); |
| 7593 | used_partitions_list.append_str(alloc, pe->partition_name); |
| 7594 | parts_str->append(pe->partition_name, strlen(pe->partition_name), |
| 7595 | system_charset_info); |
| 7596 | } |
| 7597 | partition_id++; |
| 7598 | } |
| 7599 | } |
| 7600 | } |
| 7601 | #endif |
| 7602 | |
| 7603 | /**************************************************************************** |
| 7604 | * Partition interval analysis support |
| 7605 | ***************************************************************************/ |
| 7606 | |
| 7607 | /* |
| 7608 | Setup partition_info::* members related to partitioning range analysis |
| 7609 | |
| 7610 | SYNOPSIS |
| 7611 | set_up_partition_func_pointers() |
| 7612 | part_info Partitioning info structure |
| 7613 | |
| 7614 | DESCRIPTION |
| 7615 | Assuming that passed partition_info structure already has correct values |
| 7616 | for members that specify [sub]partitioning type, table fields, and |
| 7617 | functions, set up partition_info::* members that are related to |
| 7618 | Partitioning Interval Analysis (see get_partitions_in_range_iter for its |
| 7619 | definition) |
| 7620 | |
| 7621 | IMPLEMENTATION |
| 7622 | There are three available interval analyzer functions: |
| 7623 | (1) get_part_iter_for_interval_via_mapping |
| 7624 | (2) get_part_iter_for_interval_cols_via_map |
| 7625 | (3) get_part_iter_for_interval_via_walking |
| 7626 | |
| 7627 | They all have limited applicability: |
| 7628 | (1) is applicable for "PARTITION BY <RANGE|LIST>(func(t.field))", where |
| 7629 | func is a monotonic function. |
| 7630 | |
| 7631 | (2) is applicable for "PARTITION BY <RANGE|LIST> COLUMNS (field_list) |
| 7632 | |
| 7633 | (3) is applicable for |
| 7634 | "[SUB]PARTITION BY <any-partitioning-type>(any_func(t.integer_field))" |
| 7635 | |
| 7636 | If both (1) and (3) are applicable, (1) is preferred over (3). |
| 7637 | |
| 7638 | This function sets part_info::get_part_iter_for_interval according to |
| 7639 | this criteria, and also sets some auxilary fields that the function |
| 7640 | uses. |
| 7641 | */ |
| 7642 | #ifdef WITH_PARTITION_STORAGE_ENGINE |
| 7643 | static void set_up_range_analysis_info(partition_info *part_info) |
| 7644 | { |
| 7645 | /* Set the catch-all default */ |
| 7646 | part_info->get_part_iter_for_interval= NULL; |
| 7647 | part_info->get_subpart_iter_for_interval= NULL; |
| 7648 | |
| 7649 | /* |
| 7650 | Check if get_part_iter_for_interval_via_mapping() can be used for |
| 7651 | partitioning |
| 7652 | */ |
| 7653 | switch (part_info->part_type) { |
| 7654 | case RANGE_PARTITION: |
| 7655 | case LIST_PARTITION: |
| 7656 | if (!part_info->column_list) |
| 7657 | { |
| 7658 | if (part_info->part_expr->get_monotonicity_info() != NON_MONOTONIC) |
| 7659 | { |
| 7660 | part_info->get_part_iter_for_interval= |
| 7661 | get_part_iter_for_interval_via_mapping; |
| 7662 | goto setup_subparts; |
| 7663 | } |
| 7664 | } |
| 7665 | else |
| 7666 | { |
| 7667 | part_info->get_part_iter_for_interval= |
| 7668 | get_part_iter_for_interval_cols_via_map; |
| 7669 | goto setup_subparts; |
| 7670 | } |
| 7671 | default: |
| 7672 | ; |
| 7673 | } |
| 7674 | |
| 7675 | /* |
| 7676 | Check if get_part_iter_for_interval_via_walking() can be used for |
| 7677 | partitioning |
| 7678 | */ |
| 7679 | if (part_info->num_part_fields == 1) |
| 7680 | { |
| 7681 | Field *field= part_info->part_field_array[0]; |
| 7682 | switch (field->type()) { |
| 7683 | case MYSQL_TYPE_TINY: |
| 7684 | case MYSQL_TYPE_SHORT: |
| 7685 | case MYSQL_TYPE_INT24: |
| 7686 | case MYSQL_TYPE_LONG: |
| 7687 | case MYSQL_TYPE_LONGLONG: |
| 7688 | part_info->get_part_iter_for_interval= |
| 7689 | get_part_iter_for_interval_via_walking; |
| 7690 | break; |
| 7691 | default: |
| 7692 | ; |
| 7693 | } |
| 7694 | } |
| 7695 | |
| 7696 | setup_subparts: |
| 7697 | /* |
| 7698 | Check if get_part_iter_for_interval_via_walking() can be used for |
| 7699 | subpartitioning |
| 7700 | */ |
| 7701 | if (part_info->num_subpart_fields == 1) |
| 7702 | { |
| 7703 | Field *field= part_info->subpart_field_array[0]; |
| 7704 | switch (field->type()) { |
| 7705 | case MYSQL_TYPE_TINY: |
| 7706 | case MYSQL_TYPE_SHORT: |
| 7707 | case MYSQL_TYPE_LONG: |
| 7708 | case MYSQL_TYPE_LONGLONG: |
| 7709 | part_info->get_subpart_iter_for_interval= |
| 7710 | get_part_iter_for_interval_via_walking; |
| 7711 | break; |
| 7712 | default: |
| 7713 | ; |
| 7714 | } |
| 7715 | } |
| 7716 | } |
| 7717 | |
| 7718 | |
| 7719 | /* |
| 7720 | This function takes a memory of packed fields in opt-range format |
| 7721 | and stores it in record format. To avoid having to worry about how |
| 7722 | the length of fields are calculated in opt-range format we send |
| 7723 | an array of lengths used for each field in store_length_array. |
| 7724 | |
| 7725 | SYNOPSIS |
| 7726 | store_tuple_to_record() |
| 7727 | pfield Field array |
| 7728 | store_length_array Array of field lengths |
| 7729 | value Memory where fields are stored |
| 7730 | value_end End of memory |
| 7731 | |
| 7732 | RETURN VALUE |
| 7733 | nparts Number of fields assigned |
| 7734 | */ |
| 7735 | uint32 store_tuple_to_record(Field **pfield, |
| 7736 | uint32 *store_length_array, |
| 7737 | uchar *value, |
| 7738 | uchar *value_end) |
| 7739 | { |
| 7740 | /* This function is inspired by store_key_image_rec. */ |
| 7741 | uint32 nparts= 0; |
| 7742 | uchar *loc_value; |
| 7743 | while (value < value_end) |
| 7744 | { |
| 7745 | loc_value= value; |
| 7746 | if ((*pfield)->real_maybe_null()) |
| 7747 | { |
| 7748 | if (*loc_value) |
| 7749 | (*pfield)->set_null(); |
| 7750 | else |
| 7751 | (*pfield)->set_notnull(); |
| 7752 | loc_value++; |
| 7753 | } |
| 7754 | uint len= (*pfield)->pack_length(); |
| 7755 | (*pfield)->set_key_image(loc_value, len); |
| 7756 | value+= *store_length_array; |
| 7757 | store_length_array++; |
| 7758 | nparts++; |
| 7759 | pfield++; |
| 7760 | } |
| 7761 | return nparts; |
| 7762 | } |
| 7763 | |
| 7764 | /** |
| 7765 | RANGE(columns) partitioning: compare partition value bound and probe tuple. |
| 7766 | |
| 7767 | @param val Partition column values. |
| 7768 | @param nvals_in_rec Number of (prefix) fields to compare. |
| 7769 | |
| 7770 | @return Less than/Equal to/Greater than 0 if the record is L/E/G than val. |
| 7771 | |
| 7772 | @note The partition value bound is always a full tuple (but may include the |
| 7773 | MAXVALUE special value). The probe tuple may be a prefix of partitioning |
| 7774 | tuple. |
| 7775 | */ |
| 7776 | |
| 7777 | static int cmp_rec_and_tuple(part_column_list_val *val, uint32 nvals_in_rec) |
| 7778 | { |
| 7779 | partition_info *part_info= val->part_info; |
| 7780 | Field **field= part_info->part_field_array; |
| 7781 | Field **fields_end= field + nvals_in_rec; |
| 7782 | int res; |
| 7783 | |
| 7784 | for (; field != fields_end; field++, val++) |
| 7785 | { |
| 7786 | if (val->max_value) |
| 7787 | return -1; |
| 7788 | if ((*field)->is_null()) |
| 7789 | { |
| 7790 | if (val->null_value) |
| 7791 | continue; |
| 7792 | return -1; |
| 7793 | } |
| 7794 | if (val->null_value) |
| 7795 | return +1; |
| 7796 | res= (*field)->cmp((const uchar*)val->column_value); |
| 7797 | if (res) |
| 7798 | return res; |
| 7799 | } |
| 7800 | return 0; |
| 7801 | } |
| 7802 | |
| 7803 | |
| 7804 | /** |
| 7805 | Compare record and columns partition tuple including endpoint handling. |
| 7806 | |
| 7807 | @param val Columns partition tuple |
| 7808 | @param n_vals_in_rec Number of columns to compare |
| 7809 | @param is_left_endpoint True if left endpoint (part_tuple < rec or |
| 7810 | part_tuple <= rec) |
| 7811 | @param include_endpoint If endpoint is included (part_tuple <= rec or |
| 7812 | rec <= part_tuple) |
| 7813 | |
| 7814 | @return Less than/Equal to/Greater than 0 if the record is L/E/G than |
| 7815 | the partition tuple. |
| 7816 | |
| 7817 | @see get_list_array_idx_for_endpoint() and |
| 7818 | get_partition_id_range_for_endpoint(). |
| 7819 | */ |
| 7820 | |
| 7821 | static int cmp_rec_and_tuple_prune(part_column_list_val *val, |
| 7822 | uint32 n_vals_in_rec, |
| 7823 | bool is_left_endpoint, |
| 7824 | bool include_endpoint) |
| 7825 | { |
| 7826 | int cmp; |
| 7827 | Field **field; |
| 7828 | if ((cmp= cmp_rec_and_tuple(val, n_vals_in_rec))) |
| 7829 | return cmp; |
| 7830 | field= val->part_info->part_field_array + n_vals_in_rec; |
| 7831 | if (!(*field)) |
| 7832 | { |
| 7833 | /* Full match. Only equal if including endpoint. */ |
| 7834 | if (include_endpoint) |
| 7835 | return 0; |
| 7836 | |
| 7837 | if (is_left_endpoint) |
| 7838 | return +4; /* Start of range, part_tuple < rec, return higher. */ |
| 7839 | return -4; /* End of range, rec < part_tupe, return lesser. */ |
| 7840 | } |
| 7841 | /* |
| 7842 | The prefix is equal and there are more partition columns to compare. |
| 7843 | |
| 7844 | If including left endpoint or not including right endpoint |
| 7845 | then the record is considered lesser compared to the partition. |
| 7846 | |
| 7847 | i.e: |
| 7848 | part(10, x) <= rec(10, unknown) and rec(10, unknown) < part(10, x) |
| 7849 | part <= rec -> lesser (i.e. this or previous partitions) |
| 7850 | rec < part -> lesser (i.e. this or previous partitions) |
| 7851 | */ |
| 7852 | if (is_left_endpoint == include_endpoint) |
| 7853 | return -2; |
| 7854 | |
| 7855 | /* |
| 7856 | If right endpoint and the first additional partition value |
| 7857 | is MAXVALUE, then the record is lesser. |
| 7858 | */ |
| 7859 | if (!is_left_endpoint && (val + n_vals_in_rec)->max_value) |
| 7860 | return -3; |
| 7861 | |
| 7862 | /* |
| 7863 | Otherwise the record is considered greater. |
| 7864 | |
| 7865 | rec <= part -> greater (i.e. does not match this partition, seek higher). |
| 7866 | part < rec -> greater (i.e. does not match this partition, seek higher). |
| 7867 | */ |
| 7868 | return 2; |
| 7869 | } |
| 7870 | |
| 7871 | |
| 7872 | typedef uint32 (*get_endpoint_func)(partition_info*, bool left_endpoint, |
| 7873 | bool include_endpoint); |
| 7874 | |
| 7875 | typedef uint32 (*get_col_endpoint_func)(partition_info*, bool left_endpoint, |
| 7876 | bool include_endpoint, |
| 7877 | uint32 num_parts); |
| 7878 | |
| 7879 | /** |
| 7880 | Get partition for RANGE COLUMNS endpoint. |
| 7881 | |
| 7882 | @param part_info Partitioning metadata. |
| 7883 | @param is_left_endpoint True if left endpoint (const <=/< cols) |
| 7884 | @param include_endpoint True if range includes the endpoint (<=/>=) |
| 7885 | @param nparts Total number of partitions |
| 7886 | |
| 7887 | @return Partition id of matching partition. |
| 7888 | |
| 7889 | @see get_partition_id_cols_list_for_endpoint and |
| 7890 | get_partition_id_range_for_endpoint. |
| 7891 | */ |
| 7892 | |
| 7893 | uint32 get_partition_id_cols_range_for_endpoint(partition_info *part_info, |
| 7894 | bool is_left_endpoint, |
| 7895 | bool include_endpoint, |
| 7896 | uint32 nparts) |
| 7897 | { |
| 7898 | uint min_part_id= 0, max_part_id= part_info->num_parts, loc_part_id; |
| 7899 | part_column_list_val *range_col_array= part_info->range_col_array; |
| 7900 | uint num_columns= part_info->part_field_list.elements; |
| 7901 | DBUG_ENTER("get_partition_id_cols_range_for_endpoint" ); |
| 7902 | |
| 7903 | /* Find the matching partition (including taking endpoint into account). */ |
| 7904 | do |
| 7905 | { |
| 7906 | /* Midpoint, adjusted down, so it can never be > last partition. */ |
| 7907 | loc_part_id= (max_part_id + min_part_id) >> 1; |
| 7908 | if (0 <= cmp_rec_and_tuple_prune(range_col_array + |
| 7909 | loc_part_id * num_columns, |
| 7910 | nparts, |
| 7911 | is_left_endpoint, |
| 7912 | include_endpoint)) |
| 7913 | min_part_id= loc_part_id + 1; |
| 7914 | else |
| 7915 | max_part_id= loc_part_id; |
| 7916 | } while (max_part_id > min_part_id); |
| 7917 | loc_part_id= max_part_id; |
| 7918 | |
| 7919 | /* Given value must be LESS THAN the found partition. */ |
| 7920 | DBUG_ASSERT(loc_part_id == part_info->num_parts || |
| 7921 | (0 > cmp_rec_and_tuple_prune(range_col_array + |
| 7922 | loc_part_id * num_columns, |
| 7923 | nparts, is_left_endpoint, |
| 7924 | include_endpoint))); |
| 7925 | /* Given value must be GREATER THAN or EQUAL to the previous partition. */ |
| 7926 | DBUG_ASSERT(loc_part_id == 0 || |
| 7927 | (0 <= cmp_rec_and_tuple_prune(range_col_array + |
| 7928 | (loc_part_id - 1) * num_columns, |
| 7929 | nparts, is_left_endpoint, |
| 7930 | include_endpoint))); |
| 7931 | |
| 7932 | if (!is_left_endpoint) |
| 7933 | { |
| 7934 | /* Set the end after this partition if not already after the last. */ |
| 7935 | if (loc_part_id < part_info->num_parts) |
| 7936 | loc_part_id++; |
| 7937 | } |
| 7938 | DBUG_RETURN(loc_part_id); |
| 7939 | } |
| 7940 | |
| 7941 | |
| 7942 | static int get_part_iter_for_interval_cols_via_map(partition_info *part_info, |
| 7943 | bool is_subpart, uint32 *store_length_array, |
| 7944 | uchar *min_value, uchar *max_value, |
| 7945 | uint min_len, uint max_len, |
| 7946 | uint flags, PARTITION_ITERATOR *part_iter) |
| 7947 | { |
| 7948 | bool can_match_multiple_values; |
| 7949 | uint32 nparts; |
| 7950 | get_col_endpoint_func UNINIT_VAR(get_col_endpoint); |
| 7951 | uint full_length= 0; |
| 7952 | DBUG_ENTER("get_part_iter_for_interval_cols_via_map" ); |
| 7953 | |
| 7954 | if (part_info->part_type == RANGE_PARTITION || part_info->part_type == VERSIONING_PARTITION) |
| 7955 | { |
| 7956 | get_col_endpoint= get_partition_id_cols_range_for_endpoint; |
| 7957 | part_iter->get_next= get_next_partition_id_range; |
| 7958 | } |
| 7959 | else if (part_info->part_type == LIST_PARTITION) |
| 7960 | { |
| 7961 | if (part_info->has_default_partititon() && |
| 7962 | part_info->num_parts == 1) |
| 7963 | DBUG_RETURN(-1); //only DEFAULT partition |
| 7964 | get_col_endpoint= get_partition_id_cols_list_for_endpoint; |
| 7965 | part_iter->get_next= get_next_partition_id_list; |
| 7966 | part_iter->part_info= part_info; |
| 7967 | DBUG_ASSERT(part_info->num_list_values); |
| 7968 | } |
| 7969 | else |
| 7970 | assert(0); |
| 7971 | |
| 7972 | for (uint32 i= 0; i < part_info->num_columns; i++) |
| 7973 | full_length+= store_length_array[i]; |
| 7974 | |
| 7975 | can_match_multiple_values= ((flags & |
| 7976 | (NO_MIN_RANGE | NO_MAX_RANGE | NEAR_MIN | |
| 7977 | NEAR_MAX)) || |
| 7978 | (min_len != max_len) || |
| 7979 | (min_len != full_length) || |
| 7980 | memcmp(min_value, max_value, min_len)); |
| 7981 | DBUG_ASSERT(can_match_multiple_values || (flags & EQ_RANGE) || flags == 0); |
| 7982 | if (can_match_multiple_values && part_info->has_default_partititon()) |
| 7983 | part_iter->ret_default_part= part_iter->ret_default_part_orig= TRUE; |
| 7984 | |
| 7985 | if (flags & NO_MIN_RANGE) |
| 7986 | part_iter->part_nums.start= part_iter->part_nums.cur= 0; |
| 7987 | else |
| 7988 | { |
| 7989 | // Copy from min_value to record |
| 7990 | nparts= store_tuple_to_record(part_info->part_field_array, |
| 7991 | store_length_array, |
| 7992 | min_value, |
| 7993 | min_value + min_len); |
| 7994 | part_iter->part_nums.start= part_iter->part_nums.cur= |
| 7995 | get_col_endpoint(part_info, TRUE, !(flags & NEAR_MIN), |
| 7996 | nparts); |
| 7997 | } |
| 7998 | if (flags & NO_MAX_RANGE) |
| 7999 | { |
| 8000 | if (part_info->part_type == RANGE_PARTITION || part_info->part_type == VERSIONING_PARTITION) |
| 8001 | part_iter->part_nums.end= part_info->num_parts; |
| 8002 | else /* LIST_PARTITION */ |
| 8003 | { |
| 8004 | DBUG_ASSERT(part_info->part_type == LIST_PARTITION); |
| 8005 | part_iter->part_nums.end= part_info->num_list_values; |
| 8006 | } |
| 8007 | } |
| 8008 | else |
| 8009 | { |
| 8010 | // Copy from max_value to record |
| 8011 | nparts= store_tuple_to_record(part_info->part_field_array, |
| 8012 | store_length_array, |
| 8013 | max_value, |
| 8014 | max_value + max_len); |
| 8015 | part_iter->part_nums.end= get_col_endpoint(part_info, FALSE, |
| 8016 | !(flags & NEAR_MAX), |
| 8017 | nparts); |
| 8018 | } |
| 8019 | if (part_iter->part_nums.start == part_iter->part_nums.end) |
| 8020 | { |
| 8021 | // No matching partition found. |
| 8022 | if (part_info->has_default_partititon()) |
| 8023 | { |
| 8024 | part_iter->ret_default_part= part_iter->ret_default_part_orig= TRUE; |
| 8025 | DBUG_RETURN(1); |
| 8026 | } |
| 8027 | DBUG_RETURN(0); |
| 8028 | } |
| 8029 | DBUG_RETURN(1); |
| 8030 | } |
| 8031 | |
| 8032 | |
| 8033 | /** |
| 8034 | Partitioning Interval Analysis: Initialize the iterator for "mapping" case |
| 8035 | |
| 8036 | @param part_info Partition info |
| 8037 | @param is_subpart TRUE - act for subpartitioning |
| 8038 | FALSE - act for partitioning |
| 8039 | @param store_length_array Ignored. |
| 8040 | @param min_value minimum field value, in opt_range key format. |
| 8041 | @param max_value minimum field value, in opt_range key format. |
| 8042 | @param min_len Ignored. |
| 8043 | @param max_len Ignored. |
| 8044 | @param flags Some combination of NEAR_MIN, NEAR_MAX, NO_MIN_RANGE, |
| 8045 | NO_MAX_RANGE. |
| 8046 | @param part_iter Iterator structure to be initialized |
| 8047 | |
| 8048 | @details Initialize partition set iterator to walk over the interval in |
| 8049 | ordered-array-of-partitions (for RANGE partitioning) or |
| 8050 | ordered-array-of-list-constants (for LIST partitioning) space. |
| 8051 | |
| 8052 | This function is used when partitioning is done by |
| 8053 | <RANGE|LIST>(ascending_func(t.field)), and we can map an interval in |
| 8054 | t.field space into a sub-array of partition_info::range_int_array or |
| 8055 | partition_info::list_array (see get_partition_id_range_for_endpoint, |
| 8056 | get_list_array_idx_for_endpoint for details). |
| 8057 | |
| 8058 | The function performs this interval mapping, and sets the iterator to |
| 8059 | traverse the sub-array and return appropriate partitions. |
| 8060 | |
| 8061 | @return Status of iterator |
| 8062 | @retval 0 No matching partitions (iterator not initialized) |
| 8063 | @retval 1 Ok, iterator intialized for traversal of matching partitions. |
| 8064 | @retval -1 All partitions would match (iterator not initialized) |
| 8065 | */ |
| 8066 | |
| 8067 | static int get_part_iter_for_interval_via_mapping(partition_info *part_info, |
| 8068 | bool is_subpart, |
| 8069 | uint32 *store_length_array, /* ignored */ |
| 8070 | uchar *min_value, uchar *max_value, |
| 8071 | uint min_len, uint max_len, /* ignored */ |
| 8072 | uint flags, PARTITION_ITERATOR *part_iter) |
| 8073 | { |
| 8074 | Field *field= part_info->part_field_array[0]; |
| 8075 | uint32 UNINIT_VAR(max_endpoint_val); |
| 8076 | get_endpoint_func UNINIT_VAR(get_endpoint); |
| 8077 | bool can_match_multiple_values; /* is not '=' */ |
| 8078 | uint field_len= field->pack_length_in_rec(); |
| 8079 | MYSQL_TIME start_date; |
| 8080 | bool check_zero_dates= false; |
| 8081 | bool zero_in_start_date= true; |
| 8082 | DBUG_ENTER("get_part_iter_for_interval_via_mapping" ); |
| 8083 | DBUG_ASSERT(!is_subpart); |
| 8084 | (void) store_length_array; |
| 8085 | (void)min_len; |
| 8086 | (void)max_len; |
| 8087 | part_iter->ret_null_part= part_iter->ret_null_part_orig= FALSE; |
| 8088 | part_iter->ret_default_part= part_iter->ret_default_part_orig= FALSE; |
| 8089 | |
| 8090 | if (part_info->part_type == RANGE_PARTITION) |
| 8091 | { |
| 8092 | if (part_info->part_charset_field_array) |
| 8093 | get_endpoint= get_partition_id_range_for_endpoint_charset; |
| 8094 | else |
| 8095 | get_endpoint= get_partition_id_range_for_endpoint; |
| 8096 | max_endpoint_val= part_info->num_parts; |
| 8097 | part_iter->get_next= get_next_partition_id_range; |
| 8098 | } |
| 8099 | else if (part_info->part_type == LIST_PARTITION) |
| 8100 | { |
| 8101 | |
| 8102 | if (part_info->part_charset_field_array) |
| 8103 | get_endpoint= get_list_array_idx_for_endpoint_charset; |
| 8104 | else |
| 8105 | get_endpoint= get_list_array_idx_for_endpoint; |
| 8106 | max_endpoint_val= part_info->num_list_values; |
| 8107 | part_iter->get_next= get_next_partition_id_list; |
| 8108 | part_iter->part_info= part_info; |
| 8109 | if (max_endpoint_val == 0) |
| 8110 | { |
| 8111 | /* |
| 8112 | We handle this special case without optimisations since it is |
| 8113 | of little practical value but causes a great number of complex |
| 8114 | checks later in the code. |
| 8115 | */ |
| 8116 | part_iter->part_nums.start= part_iter->part_nums.end= 0; |
| 8117 | part_iter->part_nums.cur= 0; |
| 8118 | part_iter->ret_null_part= part_iter->ret_null_part_orig= TRUE; |
| 8119 | DBUG_RETURN(-1); |
| 8120 | } |
| 8121 | } |
| 8122 | else |
| 8123 | MY_ASSERT_UNREACHABLE(); |
| 8124 | |
| 8125 | can_match_multiple_values= ((flags & |
| 8126 | (NO_MIN_RANGE | NO_MAX_RANGE | NEAR_MIN | |
| 8127 | NEAR_MAX)) || |
| 8128 | memcmp(min_value, max_value, field_len)); |
| 8129 | DBUG_ASSERT(can_match_multiple_values || (flags & EQ_RANGE) || flags == 0); |
| 8130 | if (can_match_multiple_values && part_info->has_default_partititon()) |
| 8131 | part_iter->ret_default_part= part_iter->ret_default_part_orig= TRUE; |
| 8132 | if (can_match_multiple_values && |
| 8133 | (part_info->part_type == RANGE_PARTITION || |
| 8134 | part_info->has_null_value)) |
| 8135 | { |
| 8136 | /* Range scan on RANGE or LIST partitioned table */ |
| 8137 | enum_monotonicity_info monotonic; |
| 8138 | monotonic= part_info->part_expr->get_monotonicity_info(); |
| 8139 | if (monotonic == MONOTONIC_INCREASING_NOT_NULL || |
| 8140 | monotonic == MONOTONIC_STRICT_INCREASING_NOT_NULL) |
| 8141 | { |
| 8142 | /* col is NOT NULL, but F(col) can return NULL, add NULL partition */ |
| 8143 | part_iter->ret_null_part= part_iter->ret_null_part_orig= TRUE; |
| 8144 | check_zero_dates= true; |
| 8145 | } |
| 8146 | } |
| 8147 | |
| 8148 | /* |
| 8149 | Find minimum: Do special handling if the interval has left bound in form |
| 8150 | " NULL <= X ": |
| 8151 | */ |
| 8152 | if (field->real_maybe_null() && part_info->has_null_value && |
| 8153 | !(flags & (NO_MIN_RANGE | NEAR_MIN)) && *min_value) |
| 8154 | { |
| 8155 | part_iter->ret_null_part= part_iter->ret_null_part_orig= TRUE; |
| 8156 | part_iter->part_nums.start= part_iter->part_nums.cur= 0; |
| 8157 | if (!(flags & NO_MAX_RANGE) && *max_value) |
| 8158 | { |
| 8159 | /* The right bound is X <= NULL, i.e. it is a "X IS NULL" interval */ |
| 8160 | part_iter->part_nums.end= 0; |
| 8161 | /* |
| 8162 | It is something like select * from tbl where col IS NULL |
| 8163 | and we have partition with NULL to catch it, so we do not need |
| 8164 | DEFAULT partition |
| 8165 | */ |
| 8166 | part_iter->ret_default_part= part_iter->ret_default_part_orig= FALSE; |
| 8167 | DBUG_RETURN(1); |
| 8168 | } |
| 8169 | } |
| 8170 | else |
| 8171 | { |
| 8172 | if (flags & NO_MIN_RANGE) |
| 8173 | part_iter->part_nums.start= part_iter->part_nums.cur= 0; |
| 8174 | else |
| 8175 | { |
| 8176 | /* |
| 8177 | Store the interval edge in the record buffer, and call the |
| 8178 | function that maps the edge in table-field space to an edge |
| 8179 | in ordered-set-of-partitions (for RANGE partitioning) or |
| 8180 | index-in-ordered-array-of-list-constants (for LIST) space. |
| 8181 | */ |
| 8182 | store_key_image_to_rec(field, min_value, field_len); |
| 8183 | bool include_endp= !MY_TEST(flags & NEAR_MIN); |
| 8184 | part_iter->part_nums.start= get_endpoint(part_info, 1, include_endp); |
| 8185 | if (!can_match_multiple_values && part_info->part_expr->null_value) |
| 8186 | { |
| 8187 | /* col = x and F(x) = NULL -> only search NULL partition */ |
| 8188 | part_iter->part_nums.cur= part_iter->part_nums.start= 0; |
| 8189 | part_iter->part_nums.end= 0; |
| 8190 | /* |
| 8191 | if NULL partition exists: |
| 8192 | for RANGE it is the first partition (always exists); |
| 8193 | for LIST should be indicator that it is present |
| 8194 | */ |
| 8195 | if (part_info->part_type == RANGE_PARTITION || |
| 8196 | part_info->has_null_value) |
| 8197 | { |
| 8198 | part_iter->ret_null_part= part_iter->ret_null_part_orig= TRUE; |
| 8199 | DBUG_RETURN(1); |
| 8200 | } |
| 8201 | // If no NULL partition look up in DEFAULT or there is no such value |
| 8202 | goto not_found; |
| 8203 | } |
| 8204 | part_iter->part_nums.cur= part_iter->part_nums.start; |
| 8205 | if (check_zero_dates && !part_info->part_expr->null_value) |
| 8206 | { |
| 8207 | if (!(flags & NO_MAX_RANGE) && |
| 8208 | (field->type() == MYSQL_TYPE_DATE || |
| 8209 | field->type() == MYSQL_TYPE_DATETIME)) |
| 8210 | { |
| 8211 | /* Monotonic, but return NULL for dates with zeros in month/day. */ |
| 8212 | zero_in_start_date= field->get_date(&start_date, 0); |
| 8213 | DBUG_PRINT("info" , ("zero start %u %04d-%02d-%02d" , |
| 8214 | zero_in_start_date, start_date.year, |
| 8215 | start_date.month, start_date.day)); |
| 8216 | } |
| 8217 | } |
| 8218 | if (part_iter->part_nums.start == max_endpoint_val) |
| 8219 | goto not_found; |
| 8220 | } |
| 8221 | } |
| 8222 | |
| 8223 | /* Find maximum, do the same as above but for right interval bound */ |
| 8224 | if (flags & NO_MAX_RANGE) |
| 8225 | part_iter->part_nums.end= max_endpoint_val; |
| 8226 | else |
| 8227 | { |
| 8228 | store_key_image_to_rec(field, max_value, field_len); |
| 8229 | bool include_endp= !MY_TEST(flags & NEAR_MAX); |
| 8230 | part_iter->part_nums.end= get_endpoint(part_info, 0, include_endp); |
| 8231 | if (check_zero_dates && |
| 8232 | !zero_in_start_date && |
| 8233 | !part_info->part_expr->null_value) |
| 8234 | { |
| 8235 | MYSQL_TIME end_date; |
| 8236 | bool zero_in_end_date= field->get_date(&end_date, 0); |
| 8237 | /* |
| 8238 | This is an optimization for TO_DAYS()/TO_SECONDS() to avoid scanning |
| 8239 | the NULL partition for ranges that cannot include a date with 0 as |
| 8240 | month/day. |
| 8241 | */ |
| 8242 | DBUG_PRINT("info" , ("zero end %u %04d-%02d-%02d" , |
| 8243 | zero_in_end_date, |
| 8244 | end_date.year, end_date.month, end_date.day)); |
| 8245 | DBUG_ASSERT(!memcmp(((Item_func*) part_info->part_expr)->func_name(), |
| 8246 | "to_days" , 7) || |
| 8247 | !memcmp(((Item_func*) part_info->part_expr)->func_name(), |
| 8248 | "to_seconds" , 10)); |
| 8249 | if (!zero_in_end_date && |
| 8250 | start_date.month == end_date.month && |
| 8251 | start_date.year == end_date.year) |
| 8252 | part_iter->ret_null_part= part_iter->ret_null_part_orig= false; |
| 8253 | } |
| 8254 | if (part_iter->part_nums.start >= part_iter->part_nums.end && |
| 8255 | !part_iter->ret_null_part) |
| 8256 | goto not_found; |
| 8257 | } |
| 8258 | DBUG_RETURN(1); /* Ok, iterator initialized */ |
| 8259 | |
| 8260 | not_found: |
| 8261 | if (part_info->has_default_partititon()) |
| 8262 | { |
| 8263 | part_iter->ret_default_part= part_iter->ret_default_part_orig= TRUE; |
| 8264 | DBUG_RETURN(1); |
| 8265 | } |
| 8266 | DBUG_RETURN(0); /* No partitions */ |
| 8267 | } |
| 8268 | |
| 8269 | |
| 8270 | /* See get_part_iter_for_interval_via_walking for definition of what this is */ |
| 8271 | #define MAX_RANGE_TO_WALK 32 |
| 8272 | |
| 8273 | |
| 8274 | /* |
| 8275 | Partitioning Interval Analysis: Initialize iterator to walk field interval |
| 8276 | |
| 8277 | SYNOPSIS |
| 8278 | get_part_iter_for_interval_via_walking() |
| 8279 | part_info Partition info |
| 8280 | is_subpart TRUE - act for subpartitioning |
| 8281 | FALSE - act for partitioning |
| 8282 | min_value minimum field value, in opt_range key format. |
| 8283 | max_value minimum field value, in opt_range key format. |
| 8284 | flags Some combination of NEAR_MIN, NEAR_MAX, NO_MIN_RANGE, |
| 8285 | NO_MAX_RANGE. |
| 8286 | part_iter Iterator structure to be initialized |
| 8287 | |
| 8288 | DESCRIPTION |
| 8289 | Initialize partition set iterator to walk over interval in integer field |
| 8290 | space. That is, for "const1 <=? t.field <=? const2" interval, initialize |
| 8291 | the iterator to return a set of [sub]partitions obtained with the |
| 8292 | following procedure: |
| 8293 | get partition id for t.field = const1, return it |
| 8294 | get partition id for t.field = const1+1, return it |
| 8295 | ... t.field = const1+2, ... |
| 8296 | ... ... ... |
| 8297 | ... t.field = const2 ... |
| 8298 | |
| 8299 | IMPLEMENTATION |
| 8300 | See get_partitions_in_range_iter for general description of interval |
| 8301 | analysis. We support walking over the following intervals: |
| 8302 | "t.field IS NULL" |
| 8303 | "c1 <=? t.field <=? c2", where c1 and c2 are finite. |
| 8304 | Intervals with +inf/-inf, and [NULL, c1] interval can be processed but |
| 8305 | that is more tricky and I don't have time to do it right now. |
| 8306 | |
| 8307 | RETURN |
| 8308 | 0 - No matching partitions, iterator not initialized |
| 8309 | 1 - Some partitions would match, iterator intialized for traversing them |
| 8310 | -1 - All partitions would match, iterator not initialized |
| 8311 | */ |
| 8312 | |
| 8313 | static int get_part_iter_for_interval_via_walking(partition_info *part_info, |
| 8314 | bool is_subpart, |
| 8315 | uint32 *store_length_array, /* ignored */ |
| 8316 | uchar *min_value, uchar *max_value, |
| 8317 | uint min_len, uint max_len, /* ignored */ |
| 8318 | uint flags, PARTITION_ITERATOR *part_iter) |
| 8319 | { |
| 8320 | Field *field; |
| 8321 | uint total_parts; |
| 8322 | partition_iter_func get_next_func; |
| 8323 | DBUG_ENTER("get_part_iter_for_interval_via_walking" ); |
| 8324 | (void)store_length_array; |
| 8325 | (void)min_len; |
| 8326 | (void)max_len; |
| 8327 | |
| 8328 | part_iter->ret_null_part= part_iter->ret_null_part_orig= FALSE; |
| 8329 | part_iter->ret_default_part= part_iter->ret_default_part_orig= FALSE; |
| 8330 | |
| 8331 | if (is_subpart) |
| 8332 | { |
| 8333 | field= part_info->subpart_field_array[0]; |
| 8334 | total_parts= part_info->num_subparts; |
| 8335 | get_next_func= get_next_subpartition_via_walking; |
| 8336 | } |
| 8337 | else |
| 8338 | { |
| 8339 | field= part_info->part_field_array[0]; |
| 8340 | total_parts= part_info->num_parts; |
| 8341 | get_next_func= get_next_partition_via_walking; |
| 8342 | } |
| 8343 | |
| 8344 | /* Handle the "t.field IS NULL" interval, it is a special case */ |
| 8345 | if (field->real_maybe_null() && !(flags & (NO_MIN_RANGE | NO_MAX_RANGE)) && |
| 8346 | *min_value && *max_value) |
| 8347 | { |
| 8348 | /* |
| 8349 | We don't have a part_iter->get_next() function that would find which |
| 8350 | partition "t.field IS NULL" belongs to, so find partition that contains |
| 8351 | NULL right here, and return an iterator over singleton set. |
| 8352 | */ |
| 8353 | uint32 part_id; |
| 8354 | field->set_null(); |
| 8355 | if (is_subpart) |
| 8356 | { |
| 8357 | if (!part_info->get_subpartition_id(part_info, &part_id)) |
| 8358 | { |
| 8359 | init_single_partition_iterator(part_id, part_iter); |
| 8360 | DBUG_RETURN(1); /* Ok, iterator initialized */ |
| 8361 | } |
| 8362 | } |
| 8363 | else |
| 8364 | { |
| 8365 | longlong dummy; |
| 8366 | int res= part_info->is_sub_partitioned() ? |
| 8367 | part_info->get_part_partition_id(part_info, &part_id, |
| 8368 | &dummy): |
| 8369 | part_info->get_partition_id(part_info, &part_id, &dummy); |
| 8370 | if (!res) |
| 8371 | { |
| 8372 | init_single_partition_iterator(part_id, part_iter); |
| 8373 | DBUG_RETURN(1); /* Ok, iterator initialized */ |
| 8374 | } |
| 8375 | } |
| 8376 | DBUG_RETURN(0); /* No partitions match */ |
| 8377 | } |
| 8378 | |
| 8379 | if ((field->real_maybe_null() && |
| 8380 | ((!(flags & NO_MIN_RANGE) && *min_value) || // NULL <? X |
| 8381 | (!(flags & NO_MAX_RANGE) && *max_value))) || // X <? NULL |
| 8382 | (flags & (NO_MIN_RANGE | NO_MAX_RANGE))) // -inf at any bound |
| 8383 | { |
| 8384 | DBUG_RETURN(-1); /* Can't handle this interval, have to use all partitions */ |
| 8385 | } |
| 8386 | |
| 8387 | /* Get integers for left and right interval bound */ |
| 8388 | longlong a, b; |
| 8389 | uint len= field->pack_length_in_rec(); |
| 8390 | store_key_image_to_rec(field, min_value, len); |
| 8391 | a= field->val_int(); |
| 8392 | |
| 8393 | store_key_image_to_rec(field, max_value, len); |
| 8394 | b= field->val_int(); |
| 8395 | |
| 8396 | /* |
| 8397 | Handle a special case where the distance between interval bounds is |
| 8398 | exactly 4G-1. This interval is too big for range walking, and if it is an |
| 8399 | (x,y]-type interval then the following "b +=..." code will convert it to |
| 8400 | an empty interval by "wrapping around" a + 4G-1 + 1 = a. |
| 8401 | */ |
| 8402 | if ((ulonglong)b - (ulonglong)a == ~0ULL) |
| 8403 | DBUG_RETURN(-1); |
| 8404 | |
| 8405 | a+= MY_TEST(flags & NEAR_MIN); |
| 8406 | b+= MY_TEST(!(flags & NEAR_MAX)); |
| 8407 | ulonglong n_values= b - a; |
| 8408 | |
| 8409 | /* |
| 8410 | Will it pay off to enumerate all values in the [a..b] range and evaluate |
| 8411 | the partitioning function for every value? It depends on |
| 8412 | 1. whether we'll be able to infer that some partitions are not used |
| 8413 | 2. if time savings from not scanning these partitions will be greater |
| 8414 | than time spent in enumeration. |
| 8415 | We will assume that the cost of accessing one extra partition is greater |
| 8416 | than the cost of evaluating the partitioning function O(#partitions). |
| 8417 | This means we should jump at any chance to eliminate a partition, which |
| 8418 | gives us this logic: |
| 8419 | |
| 8420 | Do the enumeration if |
| 8421 | - the number of values to enumerate is comparable to the number of |
| 8422 | partitions, or |
| 8423 | - there are not many values to enumerate. |
| 8424 | */ |
| 8425 | if ((n_values > 2*total_parts) && n_values > MAX_RANGE_TO_WALK) |
| 8426 | DBUG_RETURN(-1); |
| 8427 | |
| 8428 | part_iter->field_vals.start= part_iter->field_vals.cur= a; |
| 8429 | part_iter->field_vals.end= b; |
| 8430 | part_iter->part_info= part_info; |
| 8431 | part_iter->get_next= get_next_func; |
| 8432 | DBUG_RETURN(1); |
| 8433 | } |
| 8434 | |
| 8435 | |
| 8436 | /* |
| 8437 | PARTITION_ITERATOR::get_next implementation: enumerate partitions in range |
| 8438 | |
| 8439 | SYNOPSIS |
| 8440 | get_next_partition_id_range() |
| 8441 | part_iter Partition set iterator structure |
| 8442 | |
| 8443 | DESCRIPTION |
| 8444 | This is implementation of PARTITION_ITERATOR::get_next() that returns |
| 8445 | [sub]partition ids in [min_partition_id, max_partition_id] range. |
| 8446 | The function conforms to partition_iter_func type. |
| 8447 | |
| 8448 | RETURN |
| 8449 | partition id |
| 8450 | NOT_A_PARTITION_ID if there are no more partitions |
| 8451 | */ |
| 8452 | |
| 8453 | uint32 get_next_partition_id_range(PARTITION_ITERATOR* part_iter) |
| 8454 | { |
| 8455 | if (part_iter->part_nums.cur >= part_iter->part_nums.end) |
| 8456 | { |
| 8457 | if (part_iter->ret_null_part) |
| 8458 | { |
| 8459 | part_iter->ret_null_part= FALSE; |
| 8460 | return 0; /* NULL always in first range partition */ |
| 8461 | } |
| 8462 | // we do not have default partition in RANGE partitioning |
| 8463 | DBUG_ASSERT(!part_iter->ret_default_part); |
| 8464 | |
| 8465 | part_iter->part_nums.cur= part_iter->part_nums.start; |
| 8466 | part_iter->ret_null_part= part_iter->ret_null_part_orig; |
| 8467 | return NOT_A_PARTITION_ID; |
| 8468 | } |
| 8469 | else |
| 8470 | return part_iter->part_nums.cur++; |
| 8471 | } |
| 8472 | |
| 8473 | |
| 8474 | /* |
| 8475 | PARTITION_ITERATOR::get_next implementation for LIST partitioning |
| 8476 | |
| 8477 | SYNOPSIS |
| 8478 | get_next_partition_id_list() |
| 8479 | part_iter Partition set iterator structure |
| 8480 | |
| 8481 | DESCRIPTION |
| 8482 | This implementation of PARTITION_ITERATOR::get_next() is special for |
| 8483 | LIST partitioning: it enumerates partition ids in |
| 8484 | part_info->list_array[i] (list_col_array[i*cols] for COLUMNS LIST |
| 8485 | partitioning) where i runs over [min_idx, max_idx] interval. |
| 8486 | The function conforms to partition_iter_func type. |
| 8487 | |
| 8488 | RETURN |
| 8489 | partition id |
| 8490 | NOT_A_PARTITION_ID if there are no more partitions |
| 8491 | */ |
| 8492 | |
| 8493 | uint32 get_next_partition_id_list(PARTITION_ITERATOR *part_iter) |
| 8494 | { |
| 8495 | if (part_iter->part_nums.cur >= part_iter->part_nums.end) |
| 8496 | { |
| 8497 | if (part_iter->ret_null_part) |
| 8498 | { |
| 8499 | part_iter->ret_null_part= FALSE; |
| 8500 | return part_iter->part_info->has_null_part_id; |
| 8501 | } |
| 8502 | if (part_iter->ret_default_part) |
| 8503 | { |
| 8504 | part_iter->ret_default_part= FALSE; |
| 8505 | return part_iter->part_info->default_partition_id; |
| 8506 | } |
| 8507 | /* Reset partition for next read */ |
| 8508 | part_iter->part_nums.cur= part_iter->part_nums.start; |
| 8509 | part_iter->ret_null_part= part_iter->ret_null_part_orig; |
| 8510 | part_iter->ret_default_part= part_iter->ret_default_part_orig; |
| 8511 | return NOT_A_PARTITION_ID; |
| 8512 | } |
| 8513 | else |
| 8514 | { |
| 8515 | partition_info *part_info= part_iter->part_info; |
| 8516 | uint32 num_part= part_iter->part_nums.cur++; |
| 8517 | if (part_info->column_list) |
| 8518 | { |
| 8519 | uint num_columns= part_info->part_field_list.elements; |
| 8520 | return part_info->list_col_array[num_part*num_columns].partition_id; |
| 8521 | } |
| 8522 | return part_info->list_array[num_part].partition_id; |
| 8523 | } |
| 8524 | } |
| 8525 | |
| 8526 | |
| 8527 | /* |
| 8528 | PARTITION_ITERATOR::get_next implementation: walk over field-space interval |
| 8529 | |
| 8530 | SYNOPSIS |
| 8531 | get_next_partition_via_walking() |
| 8532 | part_iter Partitioning iterator |
| 8533 | |
| 8534 | DESCRIPTION |
| 8535 | This implementation of PARTITION_ITERATOR::get_next() returns ids of |
| 8536 | partitions that contain records with partitioning field value within |
| 8537 | [start_val, end_val] interval. |
| 8538 | The function conforms to partition_iter_func type. |
| 8539 | |
| 8540 | RETURN |
| 8541 | partition id |
| 8542 | NOT_A_PARTITION_ID if there are no more partitioning. |
| 8543 | */ |
| 8544 | |
| 8545 | static uint32 get_next_partition_via_walking(PARTITION_ITERATOR *part_iter) |
| 8546 | { |
| 8547 | uint32 part_id; |
| 8548 | Field *field= part_iter->part_info->part_field_array[0]; |
| 8549 | while (part_iter->field_vals.cur != part_iter->field_vals.end) |
| 8550 | { |
| 8551 | longlong dummy; |
| 8552 | field->store(part_iter->field_vals.cur++, field->flags & UNSIGNED_FLAG); |
| 8553 | if ((part_iter->part_info->is_sub_partitioned() && |
| 8554 | !part_iter->part_info->get_part_partition_id(part_iter->part_info, |
| 8555 | &part_id, &dummy)) || |
| 8556 | !part_iter->part_info->get_partition_id(part_iter->part_info, |
| 8557 | &part_id, &dummy)) |
| 8558 | return part_id; |
| 8559 | } |
| 8560 | part_iter->field_vals.cur= part_iter->field_vals.start; |
| 8561 | return NOT_A_PARTITION_ID; |
| 8562 | } |
| 8563 | |
| 8564 | |
| 8565 | /* Same as get_next_partition_via_walking, but for subpartitions */ |
| 8566 | |
| 8567 | static uint32 get_next_subpartition_via_walking(PARTITION_ITERATOR *part_iter) |
| 8568 | { |
| 8569 | Field *field= part_iter->part_info->subpart_field_array[0]; |
| 8570 | uint32 res; |
| 8571 | if (part_iter->field_vals.cur == part_iter->field_vals.end) |
| 8572 | { |
| 8573 | part_iter->field_vals.cur= part_iter->field_vals.start; |
| 8574 | return NOT_A_PARTITION_ID; |
| 8575 | } |
| 8576 | field->store(part_iter->field_vals.cur++, field->flags & UNSIGNED_FLAG); |
| 8577 | if (part_iter->part_info->get_subpartition_id(part_iter->part_info, |
| 8578 | &res)) |
| 8579 | return NOT_A_PARTITION_ID; |
| 8580 | return res; |
| 8581 | } |
| 8582 | |
| 8583 | /* used in error messages below */ |
| 8584 | static const char *longest_str(const char *s1, const char *s2, |
| 8585 | const char *s3=0) |
| 8586 | { |
| 8587 | if (strlen(s2) > strlen(s1)) s1= s2; |
| 8588 | if (s3 && strlen(s3) > strlen(s1)) s1= s3; |
| 8589 | return s1; |
| 8590 | } |
| 8591 | |
| 8592 | |
| 8593 | /* |
| 8594 | Create partition names |
| 8595 | |
| 8596 | SYNOPSIS |
| 8597 | create_partition_name() |
| 8598 | out:out The buffer for the created partition name string |
| 8599 | must be *at least* of FN_REFLEN+1 bytes |
| 8600 | in1 First part |
| 8601 | in2 Second part |
| 8602 | name_variant Normal, temporary or renamed partition name |
| 8603 | |
| 8604 | RETURN VALUE |
| 8605 | 0 if ok, error if name too long |
| 8606 | |
| 8607 | DESCRIPTION |
| 8608 | This method is used to calculate the partition name, service routine to |
| 8609 | the del_ren_cre_table method. |
| 8610 | */ |
| 8611 | |
| 8612 | int create_partition_name(char *out, size_t outlen, const char *in1, |
| 8613 | const char *in2, uint name_variant, bool translate) |
| 8614 | { |
| 8615 | char transl_part_name[FN_REFLEN]; |
| 8616 | const char *transl_part, *end; |
| 8617 | DBUG_ASSERT(outlen >= FN_REFLEN + 1); // consistency! same limit everywhere |
| 8618 | |
| 8619 | if (translate) |
| 8620 | { |
| 8621 | tablename_to_filename(in2, transl_part_name, FN_REFLEN); |
| 8622 | transl_part= transl_part_name; |
| 8623 | } |
| 8624 | else |
| 8625 | transl_part= in2; |
| 8626 | |
| 8627 | if (name_variant == NORMAL_PART_NAME) |
| 8628 | end= strxnmov(out, outlen-1, in1, "#P#" , transl_part, NullS); |
| 8629 | else if (name_variant == TEMP_PART_NAME) |
| 8630 | end= strxnmov(out, outlen-1, in1, "#P#" , transl_part, "#TMP#" , NullS); |
| 8631 | else |
| 8632 | { |
| 8633 | DBUG_ASSERT(name_variant == RENAMED_PART_NAME); |
| 8634 | end= strxnmov(out, outlen-1, in1, "#P#" , transl_part, "#REN#" , NullS); |
| 8635 | } |
| 8636 | if (end - out == static_cast<ptrdiff_t>(outlen-1)) |
| 8637 | { |
| 8638 | my_error(ER_PATH_LENGTH, MYF(0), longest_str(in1, transl_part)); |
| 8639 | return HA_WRONG_CREATE_OPTION; |
| 8640 | } |
| 8641 | return 0; |
| 8642 | } |
| 8643 | |
| 8644 | /** |
| 8645 | Create subpartition name. This method is used to calculate the |
| 8646 | subpartition name, service routine to the del_ren_cre_table method. |
| 8647 | The output buffer size should be FN_REFLEN + 1(terminating '\0'). |
| 8648 | |
| 8649 | @param [out] out Created partition name string |
| 8650 | @param in1 First part |
| 8651 | @param in2 Second part |
| 8652 | @param in3 Third part |
| 8653 | @param name_variant Normal, temporary or renamed partition name |
| 8654 | |
| 8655 | @retval true Error. |
| 8656 | @retval false Success. |
| 8657 | */ |
| 8658 | |
| 8659 | int create_subpartition_name(char *out, size_t outlen, |
| 8660 | const char *in1, const char *in2, |
| 8661 | const char *in3, uint name_variant) |
| 8662 | { |
| 8663 | char transl_part_name[FN_REFLEN], transl_subpart_name[FN_REFLEN], *end; |
| 8664 | DBUG_ASSERT(outlen >= FN_REFLEN + 1); // consistency! same limit everywhere |
| 8665 | |
| 8666 | tablename_to_filename(in2, transl_part_name, FN_REFLEN); |
| 8667 | tablename_to_filename(in3, transl_subpart_name, FN_REFLEN); |
| 8668 | |
| 8669 | if (name_variant == NORMAL_PART_NAME) |
| 8670 | end= strxnmov(out, outlen-1, in1, "#P#" , transl_part_name, |
| 8671 | "#SP#" , transl_subpart_name, NullS); |
| 8672 | else if (name_variant == TEMP_PART_NAME) |
| 8673 | end= strxnmov(out, outlen-1, in1, "#P#" , transl_part_name, |
| 8674 | "#SP#" , transl_subpart_name, "#TMP#" , NullS); |
| 8675 | else |
| 8676 | { |
| 8677 | DBUG_ASSERT(name_variant == RENAMED_PART_NAME); |
| 8678 | end= strxnmov(out, outlen-1, in1, "#P#" , transl_part_name, |
| 8679 | "#SP#" , transl_subpart_name, "#REN#" , NullS); |
| 8680 | } |
| 8681 | if (end - out == static_cast<ptrdiff_t>(outlen-1)) |
| 8682 | { |
| 8683 | my_error(ER_PATH_LENGTH, MYF(0), |
| 8684 | longest_str(in1, transl_part_name, transl_subpart_name)); |
| 8685 | return HA_WRONG_CREATE_OPTION; |
| 8686 | } |
| 8687 | return 0; |
| 8688 | } |
| 8689 | |
| 8690 | uint get_partition_field_store_length(Field *field) |
| 8691 | { |
| 8692 | uint store_length; |
| 8693 | |
| 8694 | store_length= field->key_length(); |
| 8695 | if (field->real_maybe_null()) |
| 8696 | store_length+= HA_KEY_NULL_LENGTH; |
| 8697 | if (field->real_type() == MYSQL_TYPE_VARCHAR) |
| 8698 | store_length+= HA_KEY_BLOB_LENGTH; |
| 8699 | return store_length; |
| 8700 | } |
| 8701 | |
| 8702 | #endif |
| 8703 | |