| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* partbounds.c |
| 4 |
* Support routines for manipulating partition bounds |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* IDENTIFICATION |
| 10 |
* src/backend/partitioning/partbounds.c |
| 11 |
* |
| 12 |
*------------------------------------------------------------------------- |
| 13 |
*/ |
| 14 |
|
| 15 |
#include "postgres.h" |
| 16 |
|
| 17 |
#include "access/relation.h" |
| 18 |
#include "access/table.h" |
| 19 |
#include "access/tableam.h" |
| 20 |
#include "catalog/partition.h" |
| 21 |
#include "catalog/pg_inherits.h" |
| 22 |
#include "catalog/pg_type.h" |
| 23 |
#include "commands/tablecmds.h" |
| 24 |
#include "common/hashfn.h" |
| 25 |
#include "executor/executor.h" |
| 26 |
#include "miscadmin.h" |
| 27 |
#include "nodes/makefuncs.h" |
| 28 |
#include "nodes/nodeFuncs.h" |
| 29 |
#include "nodes/pathnodes.h" |
| 30 |
#include "parser/parse_coerce.h" |
| 31 |
#include "partitioning/partbounds.h" |
| 32 |
#include "partitioning/partdesc.h" |
| 33 |
#include "utils/array.h" |
| 34 |
#include "utils/builtins.h" |
| 35 |
#include "utils/datum.h" |
| 36 |
#include "utils/fmgroids.h" |
| 37 |
#include "utils/lsyscache.h" |
| 38 |
#include "utils/partcache.h" |
| 39 |
#include "utils/ruleutils.h" |
| 40 |
#include "utils/snapmgr.h" |
| 41 |
#include "utils/syscache.h" |
| 42 |
|
| 43 |
/* |
| 44 |
* When qsort'ing partition bounds after reading from the catalog, each bound |
| 45 |
* is represented with one of the following structs. |
| 46 |
*/ |
| 47 |
|
| 48 |
/* One bound of a hash partition */ |
| 49 |
typedef struct PartitionHashBound |
| 50 |
{ |
| 51 |
int modulus; |
| 52 |
int remainder; |
| 53 |
int index; |
| 54 |
} PartitionHashBound; |
| 55 |
|
| 56 |
/* One value coming from some (index'th) list partition */ |
| 57 |
typedef struct PartitionListValue |
| 58 |
{ |
| 59 |
int index; |
| 60 |
Datum value; |
| 61 |
} PartitionListValue; |
| 62 |
|
| 63 |
/* One bound of a range partition */ |
| 64 |
typedef struct PartitionRangeBound |
| 65 |
{ |
| 66 |
int index; |
| 67 |
Datum *datums; /* range bound datums */ |
| 68 |
PartitionRangeDatumKind *kind; /* the kind of each datum */ |
| 69 |
bool lower; /* this is the lower (vs upper) bound */ |
| 70 |
} PartitionRangeBound; |
| 71 |
|
| 72 |
/* |
| 73 |
* Mapping from partitions of a joining relation to partitions of a join |
| 74 |
* relation being computed (a.k.a merged partitions) |
| 75 |
*/ |
| 76 |
typedef struct PartitionMap |
| 77 |
{ |
| 78 |
int nparts; /* number of partitions */ |
| 79 |
int *merged_indexes; /* indexes of merged partitions */ |
| 80 |
bool *merged; /* flags to indicate whether partitions are |
| 81 |
* merged with non-dummy partitions */ |
| 82 |
bool did_remapping; /* did we re-map partitions? */ |
| 83 |
int *old_indexes; /* old indexes of merged partitions if |
| 84 |
* did_remapping */ |
| 85 |
} PartitionMap; |
| 86 |
|
| 87 |
/* Macro for comparing two range bounds */ |
| 88 |
#define compare_range_bounds(partnatts, partsupfunc, partcollations, \ |
| 89 |
bound1, bound2) \ |
| 90 |
(partition_rbound_cmp(partnatts, partsupfunc, partcollations, \ |
| 91 |
(bound1)->datums, (bound1)->kind, (bound1)->lower, \ |
| 92 |
bound2)) |
| 93 |
|
| 94 |
static int32 qsort_partition_hbound_cmp(const void *a, const void *b); |
| 95 |
static int32 qsort_partition_list_value_cmp(const void *a, const void *b, |
| 96 |
void *arg); |
| 97 |
static int32 qsort_partition_rbound_cmp(const void *a, const void *b, |
| 98 |
void *arg); |
| 99 |
static PartitionBoundInfo create_hash_bounds(PartitionBoundSpec **boundspecs, |
| 100 |
int nparts, PartitionKey key, int **mapping); |
| 101 |
static PartitionBoundInfo create_list_bounds(PartitionBoundSpec **boundspecs, |
| 102 |
int nparts, PartitionKey key, int **mapping); |
| 103 |
static PartitionBoundInfo create_range_bounds(PartitionBoundSpec **boundspecs, |
| 104 |
int nparts, PartitionKey key, int **mapping); |
| 105 |
static PartitionBoundInfo merge_list_bounds(FmgrInfo *partsupfunc, |
| 106 |
Oid *partcollation, |
| 107 |
RelOptInfo *outer_rel, |
| 108 |
RelOptInfo *inner_rel, |
| 109 |
JoinType jointype, |
| 110 |
List **outer_parts, |
| 111 |
List **inner_parts); |
| 112 |
static PartitionBoundInfo merge_range_bounds(int partnatts, |
| 113 |
FmgrInfo *partsupfuncs, |
| 114 |
Oid *partcollations, |
| 115 |
RelOptInfo *outer_rel, |
| 116 |
RelOptInfo *inner_rel, |
| 117 |
JoinType jointype, |
| 118 |
List **outer_parts, |
| 119 |
List **inner_parts); |
| 120 |
static void init_partition_map(RelOptInfo *rel, PartitionMap *map); |
| 121 |
static void free_partition_map(PartitionMap *map); |
| 122 |
static bool is_dummy_partition(RelOptInfo *rel, int part_index); |
| 123 |
static int merge_matching_partitions(PartitionMap *outer_map, |
| 124 |
PartitionMap *inner_map, |
| 125 |
int outer_index, |
| 126 |
int inner_index, |
| 127 |
int *next_index); |
| 128 |
static int process_outer_partition(PartitionMap *outer_map, |
| 129 |
PartitionMap *inner_map, |
| 130 |
bool outer_has_default, |
| 131 |
bool inner_has_default, |
| 132 |
int outer_index, |
| 133 |
int inner_default, |
| 134 |
JoinType jointype, |
| 135 |
int *next_index, |
| 136 |
int *default_index); |
| 137 |
static int process_inner_partition(PartitionMap *outer_map, |
| 138 |
PartitionMap *inner_map, |
| 139 |
bool outer_has_default, |
| 140 |
bool inner_has_default, |
| 141 |
int inner_index, |
| 142 |
int outer_default, |
| 143 |
JoinType jointype, |
| 144 |
int *next_index, |
| 145 |
int *default_index); |
| 146 |
static void merge_null_partitions(PartitionMap *outer_map, |
| 147 |
PartitionMap *inner_map, |
| 148 |
bool outer_has_null, |
| 149 |
bool inner_has_null, |
| 150 |
int outer_null, |
| 151 |
int inner_null, |
| 152 |
JoinType jointype, |
| 153 |
int *next_index, |
| 154 |
int *null_index); |
| 155 |
static void merge_default_partitions(PartitionMap *outer_map, |
| 156 |
PartitionMap *inner_map, |
| 157 |
bool outer_has_default, |
| 158 |
bool inner_has_default, |
| 159 |
int outer_default, |
| 160 |
int inner_default, |
| 161 |
JoinType jointype, |
| 162 |
int *next_index, |
| 163 |
int *default_index); |
| 164 |
static int merge_partition_with_dummy(PartitionMap *map, int index, |
| 165 |
int *next_index); |
| 166 |
static void fix_merged_indexes(PartitionMap *outer_map, |
| 167 |
PartitionMap *inner_map, |
| 168 |
int nmerged, List *merged_indexes); |
| 169 |
static void generate_matching_part_pairs(RelOptInfo *outer_rel, |
| 170 |
RelOptInfo *inner_rel, |
| 171 |
PartitionMap *outer_map, |
| 172 |
PartitionMap *inner_map, |
| 173 |
int nmerged, |
| 174 |
List **outer_parts, |
| 175 |
List **inner_parts); |
| 176 |
static PartitionBoundInfo build_merged_partition_bounds(char strategy, |
| 177 |
List *merged_datums, |
| 178 |
List *merged_kinds, |
| 179 |
List *merged_indexes, |
| 180 |
int null_index, |
| 181 |
int default_index); |
| 182 |
static int get_range_partition(RelOptInfo *rel, |
| 183 |
PartitionBoundInfo bi, |
| 184 |
int *lb_pos, |
| 185 |
PartitionRangeBound *lb, |
| 186 |
PartitionRangeBound *ub); |
| 187 |
static int get_range_partition_internal(PartitionBoundInfo bi, |
| 188 |
int *lb_pos, |
| 189 |
PartitionRangeBound *lb, |
| 190 |
PartitionRangeBound *ub); |
| 191 |
static bool compare_range_partitions(int partnatts, FmgrInfo *partsupfuncs, |
| 192 |
Oid *partcollations, |
| 193 |
PartitionRangeBound *outer_lb, |
| 194 |
PartitionRangeBound *outer_ub, |
| 195 |
PartitionRangeBound *inner_lb, |
| 196 |
PartitionRangeBound *inner_ub, |
| 197 |
int *lb_cmpval, int *ub_cmpval); |
| 198 |
static void get_merged_range_bounds(int partnatts, FmgrInfo *partsupfuncs, |
| 199 |
Oid *partcollations, JoinType jointype, |
| 200 |
PartitionRangeBound *outer_lb, |
| 201 |
PartitionRangeBound *outer_ub, |
| 202 |
PartitionRangeBound *inner_lb, |
| 203 |
PartitionRangeBound *inner_ub, |
| 204 |
int lb_cmpval, int ub_cmpval, |
| 205 |
PartitionRangeBound *merged_lb, |
| 206 |
PartitionRangeBound *merged_ub); |
| 207 |
static void add_merged_range_bounds(int partnatts, FmgrInfo *partsupfuncs, |
| 208 |
Oid *partcollations, |
| 209 |
PartitionRangeBound *merged_lb, |
| 210 |
PartitionRangeBound *merged_ub, |
| 211 |
int merged_index, |
| 212 |
List **merged_datums, |
| 213 |
List **merged_kinds, |
| 214 |
List **merged_indexes); |
| 215 |
static PartitionRangeBound *make_one_partition_rbound(PartitionKey key, int index, |
| 216 |
List *datums, bool lower); |
| 217 |
static int32 partition_hbound_cmp(int modulus1, int remainder1, int modulus2, |
| 218 |
int remainder2); |
| 219 |
static int32 partition_rbound_cmp(int partnatts, FmgrInfo *partsupfunc, |
| 220 |
Oid *partcollation, Datum *datums1, |
| 221 |
PartitionRangeDatumKind *kind1, bool lower1, |
| 222 |
PartitionRangeBound *b2); |
| 223 |
static int partition_range_bsearch(int partnatts, FmgrInfo *partsupfunc, |
| 224 |
Oid *partcollation, |
| 225 |
PartitionBoundInfo boundinfo, |
| 226 |
PartitionRangeBound *probe, int32 *cmpval); |
| 227 |
static Expr *make_partition_op_expr(PartitionKey key, int keynum, |
| 228 |
uint16 strategy, Expr *arg1, Expr *arg2); |
| 229 |
static Oid get_partition_operator(PartitionKey key, int col, |
| 230 |
StrategyNumber strategy, bool *need_relabel); |
| 231 |
static List *get_qual_for_hash(Relation parent, PartitionBoundSpec *spec); |
| 232 |
static List *get_qual_for_list(Relation parent, PartitionBoundSpec *spec); |
| 233 |
static List *get_qual_for_range(Relation parent, PartitionBoundSpec *spec, |
| 234 |
bool for_default); |
| 235 |
static void get_range_key_properties(PartitionKey key, int keynum, |
| 236 |
PartitionRangeDatum *ldatum, |
| 237 |
PartitionRangeDatum *udatum, |
| 238 |
ListCell **partexprs_item, |
| 239 |
Expr **keyCol, |
| 240 |
Const **lower_val, Const **upper_val); |
| 241 |
static List *get_range_nulltest(PartitionKey key); |
| 242 |
|
| 243 |
/* |
| 244 |
* get_qual_from_partbound |
| 245 |
* Given a parser node for partition bound, return the list of executable |
| 246 |
* expressions as partition constraint |
| 247 |
*/ |
| 248 |
List * |
| 249 |
get_qual_from_partbound(Relation parent, PartitionBoundSpec *spec) |
| 250 |
{ |
| 251 |
PartitionKey key = RelationGetPartitionKey(parent); |
| 252 |
List *my_qual = NIL; |
| 253 |
|
| 254 |
Assert(key != NULL); |
| 255 |
|
| 256 |
switch (key->strategy) |
| 257 |
{ |
| 258 |
case PARTITION_STRATEGY_HASH: |
| 259 |
Assert(spec->strategy == PARTITION_STRATEGY_HASH); |
| 260 |
my_qual = get_qual_for_hash(parent, spec); |
| 261 |
break; |
| 262 |
|
| 263 |
case PARTITION_STRATEGY_LIST: |
| 264 |
Assert(spec->strategy == PARTITION_STRATEGY_LIST); |
| 265 |
my_qual = get_qual_for_list(parent, spec); |
| 266 |
break; |
| 267 |
|
| 268 |
case PARTITION_STRATEGY_RANGE: |
| 269 |
Assert(spec->strategy == PARTITION_STRATEGY_RANGE); |
| 270 |
my_qual = get_qual_for_range(parent, spec, false); |
| 271 |
break; |
| 272 |
} |
| 273 |
|
| 274 |
return my_qual; |
| 275 |
} |
| 276 |
|
| 277 |
/* |
| 278 |
* partition_bounds_create |
| 279 |
* Build a PartitionBoundInfo struct from a list of PartitionBoundSpec |
| 280 |
* nodes |
| 281 |
* |
| 282 |
* This function creates a PartitionBoundInfo and fills the values of its |
| 283 |
* various members based on the input list. Importantly, 'datums' array will |
| 284 |
* contain Datum representation of individual bounds (possibly after |
| 285 |
* de-duplication as in case of range bounds), sorted in a canonical order |
| 286 |
* defined by qsort_partition_* functions of respective partitioning methods. |
| 287 |
* 'indexes' array will contain as many elements as there are bounds (specific |
| 288 |
* exceptions to this rule are listed in the function body), which represent |
| 289 |
* the 0-based canonical positions of partitions. |
| 290 |
* |
| 291 |
* Upon return from this function, *mapping is set to an array of |
| 292 |
* list_length(boundspecs) elements, each of which maps the original index of |
| 293 |
* a partition to its canonical index. |
| 294 |
* |
| 295 |
* Note: The objects returned by this function are wholly allocated in the |
| 296 |
* current memory context. |
| 297 |
*/ |
| 298 |
PartitionBoundInfo |
| 299 |
partition_bounds_create(PartitionBoundSpec **boundspecs, int nparts, |
| 300 |
PartitionKey key, int **mapping) |
| 301 |
{ |
| 302 |
int i; |
| 303 |
|
| 304 |
Assert(nparts > 0); |
| 305 |
|
| 306 |
/* |
| 307 |
* For each partitioning method, we first convert the partition bounds |
| 308 |
* from their parser node representation to the internal representation, |
| 309 |
* along with any additional preprocessing (such as de-duplicating range |
| 310 |
* bounds). Resulting bound datums are then added to the 'datums' array |
| 311 |
* in PartitionBoundInfo. For each datum added, an integer indicating the |
| 312 |
* canonical partition index is added to the 'indexes' array. |
| 313 |
* |
| 314 |
* For each bound, we remember its partition's position (0-based) in the |
| 315 |
* original list to later map it to the canonical index. |
| 316 |
*/ |
| 317 |
|
| 318 |
/* |
| 319 |
* Initialize mapping array with invalid values, this is filled within |
| 320 |
* each sub-routine below depending on the bound type. |
| 321 |
*/ |
| 322 |
*mapping = (int *) palloc(sizeof(int) * nparts); |
| 323 |
for (i = 0; i < nparts; i++) |
| 324 |
(*mapping)[i] = -1; |
| 325 |
|
| 326 |
switch (key->strategy) |
| 327 |
{ |
| 328 |
case PARTITION_STRATEGY_HASH: |
| 329 |
return create_hash_bounds(boundspecs, nparts, key, mapping); |
| 330 |
|
| 331 |
case PARTITION_STRATEGY_LIST: |
| 332 |
return create_list_bounds(boundspecs, nparts, key, mapping); |
| 333 |
|
| 334 |
case PARTITION_STRATEGY_RANGE: |
| 335 |
return create_range_bounds(boundspecs, nparts, key, mapping); |
| 336 |
} |
| 337 |
|
| 338 |
Assert(false); |
| 339 |
return NULL; /* keep compiler quiet */ |
| 340 |
} |
| 341 |
|
| 342 |
/* |
| 343 |
* create_hash_bounds |
| 344 |
* Create a PartitionBoundInfo for a hash partitioned table |
| 345 |
*/ |
| 346 |
static PartitionBoundInfo |
| 347 |
create_hash_bounds(PartitionBoundSpec **boundspecs, int nparts, |
| 348 |
PartitionKey key, int **mapping) |
| 349 |
{ |
| 350 |
PartitionBoundInfo boundinfo; |
| 351 |
PartitionHashBound *hbounds; |
| 352 |
int i; |
| 353 |
int greatest_modulus; |
| 354 |
Datum *boundDatums; |
| 355 |
|
| 356 |
boundinfo = (PartitionBoundInfoData *) |
| 357 |
palloc0(sizeof(PartitionBoundInfoData)); |
| 358 |
boundinfo->strategy = key->strategy; |
| 359 |
/* No special hash partitions. */ |
| 360 |
boundinfo->null_index = -1; |
| 361 |
boundinfo->default_index = -1; |
| 362 |
|
| 363 |
hbounds = (PartitionHashBound *) |
| 364 |
palloc(nparts * sizeof(PartitionHashBound)); |
| 365 |
|
| 366 |
/* Convert from node to the internal representation */ |
| 367 |
for (i = 0; i < nparts; i++) |
| 368 |
{ |
| 369 |
PartitionBoundSpec *spec = boundspecs[i]; |
| 370 |
|
| 371 |
if (spec->strategy != PARTITION_STRATEGY_HASH) |
| 372 |
elog(ERROR, "invalid strategy in partition bound spec"); |
| 373 |
|
| 374 |
hbounds[i].modulus = spec->modulus; |
| 375 |
hbounds[i].remainder = spec->remainder; |
| 376 |
hbounds[i].index = i; |
| 377 |
} |
| 378 |
|
| 379 |
/* Sort all the bounds in ascending order */ |
| 380 |
qsort(hbounds, nparts, sizeof(PartitionHashBound), |
| 381 |
qsort_partition_hbound_cmp); |
| 382 |
|
| 383 |
/* After sorting, moduli are now stored in ascending order. */ |
| 384 |
greatest_modulus = hbounds[nparts - 1].modulus; |
| 385 |
|
| 386 |
boundinfo->ndatums = nparts; |
| 387 |
boundinfo->datums = (Datum **) palloc0(nparts * sizeof(Datum *)); |
| 388 |
boundinfo->kind = NULL; |
| 389 |
boundinfo->interleaved_parts = NULL; |
| 390 |
boundinfo->nindexes = greatest_modulus; |
| 391 |
boundinfo->indexes = (int *) palloc(greatest_modulus * sizeof(int)); |
| 392 |
for (i = 0; i < greatest_modulus; i++) |
| 393 |
boundinfo->indexes[i] = -1; |
| 394 |
|
| 395 |
/* |
| 396 |
* In the loop below, to save from allocating a series of small datum |
| 397 |
* arrays, here we just allocate a single array and below we'll just |
| 398 |
* assign a portion of this array per partition. |
| 399 |
*/ |
| 400 |
boundDatums = (Datum *) palloc(nparts * 2 * sizeof(Datum)); |
| 401 |
|
| 402 |
/* |
| 403 |
* For hash partitioning, there are as many datums (modulus and remainder |
| 404 |
* pairs) as there are partitions. Indexes are simply values ranging from |
| 405 |
* 0 to (nparts - 1). |
| 406 |
*/ |
| 407 |
for (i = 0; i < nparts; i++) |
| 408 |
{ |
| 409 |
int modulus = hbounds[i].modulus; |
| 410 |
int remainder = hbounds[i].remainder; |
| 411 |
|
| 412 |
boundinfo->datums[i] = &boundDatums[i * 2]; |
| 413 |
boundinfo->datums[i][0] = Int32GetDatum(modulus); |
| 414 |
boundinfo->datums[i][1] = Int32GetDatum(remainder); |
| 415 |
|
| 416 |
while (remainder < greatest_modulus) |
| 417 |
{ |
| 418 |
/* overlap? */ |
| 419 |
Assert(boundinfo->indexes[remainder] == -1); |
| 420 |
boundinfo->indexes[remainder] = i; |
| 421 |
remainder += modulus; |
| 422 |
} |
| 423 |
|
| 424 |
(*mapping)[hbounds[i].index] = i; |
| 425 |
} |
| 426 |
pfree(hbounds); |
| 427 |
|
| 428 |
return boundinfo; |
| 429 |
} |
| 430 |
|
| 431 |
/* |
| 432 |
* get_non_null_list_datum_count |
| 433 |
* Counts the number of non-null Datums in each partition. |
| 434 |
*/ |
| 435 |
static int |
| 436 |
get_non_null_list_datum_count(PartitionBoundSpec **boundspecs, int nparts) |
| 437 |
{ |
| 438 |
int i; |
| 439 |
int count = 0; |
| 440 |
|
| 441 |
for (i = 0; i < nparts; i++) |
| 442 |
{ |
| 443 |
ListCell *lc; |
| 444 |
|
| 445 |
foreach(lc, boundspecs[i]->listdatums) |
| 446 |
{ |
| 447 |
Const *val = lfirst_node(Const, lc); |
| 448 |
|
| 449 |
if (!val->constisnull) |
| 450 |
count++; |
| 451 |
} |
| 452 |
} |
| 453 |
|
| 454 |
return count; |
| 455 |
} |
| 456 |
|
| 457 |
/* |
| 458 |
* create_list_bounds |
| 459 |
* Create a PartitionBoundInfo for a list partitioned table |
| 460 |
*/ |
| 461 |
static PartitionBoundInfo |
| 462 |
create_list_bounds(PartitionBoundSpec **boundspecs, int nparts, |
| 463 |
PartitionKey key, int **mapping) |
| 464 |
{ |
| 465 |
PartitionBoundInfo boundinfo; |
| 466 |
PartitionListValue *all_values; |
| 467 |
int i; |
| 468 |
int j; |
| 469 |
int ndatums; |
| 470 |
int next_index = 0; |
| 471 |
int default_index = -1; |
| 472 |
int null_index = -1; |
| 473 |
Datum *boundDatums; |
| 474 |
|
| 475 |
boundinfo = (PartitionBoundInfoData *) |
| 476 |
palloc0(sizeof(PartitionBoundInfoData)); |
| 477 |
boundinfo->strategy = key->strategy; |
| 478 |
/* Will be set correctly below. */ |
| 479 |
boundinfo->null_index = -1; |
| 480 |
boundinfo->default_index = -1; |
| 481 |
|
| 482 |
ndatums = get_non_null_list_datum_count(boundspecs, nparts); |
| 483 |
all_values = (PartitionListValue *) |
| 484 |
palloc(ndatums * sizeof(PartitionListValue)); |
| 485 |
|
| 486 |
/* Create a unified list of non-null values across all partitions. */ |
| 487 |
for (j = 0, i = 0; i < nparts; i++) |
| 488 |
{ |
| 489 |
PartitionBoundSpec *spec = boundspecs[i]; |
| 490 |
ListCell *c; |
| 491 |
|
| 492 |
if (spec->strategy != PARTITION_STRATEGY_LIST) |
| 493 |
elog(ERROR, "invalid strategy in partition bound spec"); |
| 494 |
|
| 495 |
/* |
| 496 |
* Note the index of the partition bound spec for the default |
| 497 |
* partition. There's no datum to add to the list on non-null datums |
| 498 |
* for this partition. |
| 499 |
*/ |
| 500 |
if (spec->is_default) |
| 501 |
{ |
| 502 |
default_index = i; |
| 503 |
continue; |
| 504 |
} |
| 505 |
|
| 506 |
foreach(c, spec->listdatums) |
| 507 |
{ |
| 508 |
Const *val = lfirst_node(Const, c); |
| 509 |
|
| 510 |
if (!val->constisnull) |
| 511 |
{ |
| 512 |
all_values[j].index = i; |
| 513 |
all_values[j].value = val->constvalue; |
| 514 |
j++; |
| 515 |
} |
| 516 |
else |
| 517 |
{ |
| 518 |
/* |
| 519 |
* Never put a null into the values array; save the index of |
| 520 |
* the partition that stores nulls, instead. |
| 521 |
*/ |
| 522 |
if (null_index != -1) |
| 523 |
elog(ERROR, "found null more than once"); |
| 524 |
null_index = i; |
| 525 |
} |
| 526 |
} |
| 527 |
} |
| 528 |
|
| 529 |
/* ensure we found a Datum for every slot in the all_values array */ |
| 530 |
Assert(j == ndatums); |
| 531 |
|
| 532 |
qsort_arg(all_values, ndatums, sizeof(PartitionListValue), |
| 533 |
qsort_partition_list_value_cmp, key); |
| 534 |
|
| 535 |
boundinfo->ndatums = ndatums; |
| 536 |
boundinfo->datums = (Datum **) palloc0(ndatums * sizeof(Datum *)); |
| 537 |
boundinfo->kind = NULL; |
| 538 |
boundinfo->interleaved_parts = NULL; |
| 539 |
boundinfo->nindexes = ndatums; |
| 540 |
boundinfo->indexes = (int *) palloc(ndatums * sizeof(int)); |
| 541 |
|
| 542 |
/* |
| 543 |
* In the loop below, to save from allocating a series of small datum |
| 544 |
* arrays, here we just allocate a single array and below we'll just |
| 545 |
* assign a portion of this array per datum. |
| 546 |
*/ |
| 547 |
boundDatums = (Datum *) palloc(ndatums * sizeof(Datum)); |
| 548 |
|
| 549 |
/* |
| 550 |
* Copy values. Canonical indexes are values ranging from 0 to (nparts - |
| 551 |
* 1) assigned to each partition such that all datums of a given partition |
| 552 |
* receive the same value. The value for a given partition is the index of |
| 553 |
* that partition's smallest datum in the all_values[] array. |
| 554 |
*/ |
| 555 |
for (i = 0; i < ndatums; i++) |
| 556 |
{ |
| 557 |
int orig_index = all_values[i].index; |
| 558 |
|
| 559 |
boundinfo->datums[i] = &boundDatums[i]; |
| 560 |
boundinfo->datums[i][0] = datumCopy(all_values[i].value, |
| 561 |
key->parttypbyval[0], |
| 562 |
key->parttyplen[0]); |
| 563 |
|
| 564 |
/* If the old index has no mapping, assign one */ |
| 565 |
if ((*mapping)[orig_index] == -1) |
| 566 |
(*mapping)[orig_index] = next_index++; |
| 567 |
|
| 568 |
boundinfo->indexes[i] = (*mapping)[orig_index]; |
| 569 |
} |
| 570 |
|
| 571 |
pfree(all_values); |
| 572 |
|
| 573 |
/* |
| 574 |
* Set the canonical value for null_index, if any. |
| 575 |
* |
| 576 |
* It is possible that the null-accepting partition has not been assigned |
| 577 |
* an index yet, which could happen if such partition accepts only null |
| 578 |
* and hence not handled in the above loop which only looked at non-null |
| 579 |
* values. |
| 580 |
*/ |
| 581 |
if (null_index != -1) |
| 582 |
{ |
| 583 |
Assert(null_index >= 0); |
| 584 |
if ((*mapping)[null_index] == -1) |
| 585 |
(*mapping)[null_index] = next_index++; |
| 586 |
boundinfo->null_index = (*mapping)[null_index]; |
| 587 |
} |
| 588 |
|
| 589 |
/* Set the canonical value for default_index, if any. */ |
| 590 |
if (default_index != -1) |
| 591 |
{ |
| 592 |
/* |
| 593 |
* The default partition accepts any value not specified in the lists |
| 594 |
* of other partitions, hence it should not get mapped index while |
| 595 |
* assigning those for non-null datums. |
| 596 |
*/ |
| 597 |
Assert(default_index >= 0); |
| 598 |
Assert((*mapping)[default_index] == -1); |
| 599 |
(*mapping)[default_index] = next_index++; |
| 600 |
boundinfo->default_index = (*mapping)[default_index]; |
| 601 |
} |
| 602 |
|
| 603 |
/* |
| 604 |
* Calculate interleaved partitions. Here we look for partitions which |
| 605 |
* might be interleaved with other partitions and set a bit in |
| 606 |
* interleaved_parts for any partitions which may be interleaved with |
| 607 |
* another partition. |
| 608 |
*/ |
| 609 |
|
| 610 |
/* |
| 611 |
* There must be multiple partitions to have any interleaved partitions, |
| 612 |
* otherwise there's nothing to interleave with. |
| 613 |
*/ |
| 614 |
if (nparts > 1) |
| 615 |
{ |
| 616 |
/* |
| 617 |
* Short-circuit check to see if only 1 Datum is allowed per |
| 618 |
* partition. When this is true there's no need to do the more |
| 619 |
* expensive checks to look for interleaved values. |
| 620 |
*/ |
| 621 |
if (boundinfo->ndatums + |
| 622 |
partition_bound_accepts_nulls(boundinfo) + |
| 623 |
partition_bound_has_default(boundinfo) != nparts) |
| 624 |
{ |
| 625 |
int last_index = -1; |
| 626 |
|
| 627 |
/* |
| 628 |
* Since the indexes array is sorted in Datum order, if any |
| 629 |
* partitions are interleaved then it will show up by the |
| 630 |
* partition indexes not being in ascending order. Here we check |
| 631 |
* for that and record all partitions that are out of order. |
| 632 |
*/ |
| 633 |
for (i = 0; i < boundinfo->nindexes; i++) |
| 634 |
{ |
| 635 |
int index = boundinfo->indexes[i]; |
| 636 |
|
| 637 |
if (index < last_index) |
| 638 |
boundinfo->interleaved_parts = bms_add_member(boundinfo->interleaved_parts, |
| 639 |
index); |
| 640 |
|
| 641 |
/* |
| 642 |
* Otherwise, if the null_index exists in the indexes array, |
| 643 |
* then the NULL partition must also allow some other Datum, |
| 644 |
* therefore it's "interleaved". |
| 645 |
*/ |
| 646 |
else if (partition_bound_accepts_nulls(boundinfo) && |
| 647 |
index == boundinfo->null_index) |
| 648 |
boundinfo->interleaved_parts = bms_add_member(boundinfo->interleaved_parts, |
| 649 |
index); |
| 650 |
|
| 651 |
last_index = index; |
| 652 |
} |
| 653 |
} |
| 654 |
|
| 655 |
/* |
| 656 |
* The DEFAULT partition is the "catch-all" partition that can contain |
| 657 |
* anything that does not belong to any other partition. If there are |
| 658 |
* any other partitions then the DEFAULT partition must be marked as |
| 659 |
* interleaved. |
| 660 |
*/ |
| 661 |
if (partition_bound_has_default(boundinfo)) |
| 662 |
boundinfo->interleaved_parts = bms_add_member(boundinfo->interleaved_parts, |
| 663 |
boundinfo->default_index); |
| 664 |
} |
| 665 |
|
| 666 |
|
| 667 |
/* All partitions must now have been assigned canonical indexes. */ |
| 668 |
Assert(next_index == nparts); |
| 669 |
return boundinfo; |
| 670 |
} |
| 671 |
|
| 672 |
/* |
| 673 |
* create_range_bounds |
| 674 |
* Create a PartitionBoundInfo for a range partitioned table |
| 675 |
*/ |
| 676 |
static PartitionBoundInfo |
| 677 |
create_range_bounds(PartitionBoundSpec **boundspecs, int nparts, |
| 678 |
PartitionKey key, int **mapping) |
| 679 |
{ |
| 680 |
PartitionBoundInfo boundinfo; |
| 681 |
PartitionRangeBound **rbounds = NULL; |
| 682 |
PartitionRangeBound **all_bounds, |
| 683 |
*prev; |
| 684 |
int i, |
| 685 |
k, |
| 686 |
partnatts; |
| 687 |
int ndatums = 0; |
| 688 |
int default_index = -1; |
| 689 |
int next_index = 0; |
| 690 |
Datum *boundDatums; |
| 691 |
PartitionRangeDatumKind *boundKinds; |
| 692 |
|
| 693 |
boundinfo = (PartitionBoundInfoData *) |
| 694 |
palloc0(sizeof(PartitionBoundInfoData)); |
| 695 |
boundinfo->strategy = key->strategy; |
| 696 |
/* There is no special null-accepting range partition. */ |
| 697 |
boundinfo->null_index = -1; |
| 698 |
/* Will be set correctly below. */ |
| 699 |
boundinfo->default_index = -1; |
| 700 |
|
| 701 |
all_bounds = (PartitionRangeBound **) |
| 702 |
palloc0(2 * nparts * sizeof(PartitionRangeBound *)); |
| 703 |
|
| 704 |
/* Create a unified list of range bounds across all the partitions. */ |
| 705 |
ndatums = 0; |
| 706 |
for (i = 0; i < nparts; i++) |
| 707 |
{ |
| 708 |
PartitionBoundSpec *spec = boundspecs[i]; |
| 709 |
PartitionRangeBound *lower, |
| 710 |
*upper; |
| 711 |
|
| 712 |
if (spec->strategy != PARTITION_STRATEGY_RANGE) |
| 713 |
elog(ERROR, "invalid strategy in partition bound spec"); |
| 714 |
|
| 715 |
/* |
| 716 |
* Note the index of the partition bound spec for the default |
| 717 |
* partition. There's no datum to add to the all_bounds array for |
| 718 |
* this partition. |
| 719 |
*/ |
| 720 |
if (spec->is_default) |
| 721 |
{ |
| 722 |
default_index = i; |
| 723 |
continue; |
| 724 |
} |
| 725 |
|
| 726 |
lower = make_one_partition_rbound(key, i, spec->lowerdatums, true); |
| 727 |
upper = make_one_partition_rbound(key, i, spec->upperdatums, false); |
| 728 |
all_bounds[ndatums++] = lower; |
| 729 |
all_bounds[ndatums++] = upper; |
| 730 |
} |
| 731 |
|
| 732 |
Assert(ndatums == nparts * 2 || |
| 733 |
(default_index != -1 && ndatums == (nparts - 1) * 2)); |
| 734 |
|
| 735 |
/* Sort all the bounds in ascending order */ |
| 736 |
qsort_arg(all_bounds, ndatums, |
| 737 |
sizeof(PartitionRangeBound *), |
| 738 |
qsort_partition_rbound_cmp, |
| 739 |
key); |
| 740 |
|
| 741 |
/* Save distinct bounds from all_bounds into rbounds. */ |
| 742 |
rbounds = (PartitionRangeBound **) |
| 743 |
palloc(ndatums * sizeof(PartitionRangeBound *)); |
| 744 |
k = 0; |
| 745 |
prev = NULL; |
| 746 |
for (i = 0; i < ndatums; i++) |
| 747 |
{ |
| 748 |
PartitionRangeBound *cur = all_bounds[i]; |
| 749 |
bool is_distinct = false; |
| 750 |
int j; |
| 751 |
|
| 752 |
/* Is the current bound distinct from the previous one? */ |
| 753 |
for (j = 0; j < key->partnatts; j++) |
| 754 |
{ |
| 755 |
Datum cmpval; |
| 756 |
|
| 757 |
if (prev == NULL || cur->kind[j] != prev->kind[j]) |
| 758 |
{ |
| 759 |
is_distinct = true; |
| 760 |
break; |
| 761 |
} |
| 762 |
|
| 763 |
/* |
| 764 |
* If the bounds are both MINVALUE or MAXVALUE, stop now and treat |
| 765 |
* them as equal, since any values after this point must be |
| 766 |
* ignored. |
| 767 |
*/ |
| 768 |
if (cur->kind[j] != PARTITION_RANGE_DATUM_VALUE) |
| 769 |
break; |
| 770 |
|
| 771 |
cmpval = FunctionCall2Coll(&key->partsupfunc[j], |
| 772 |
key->partcollation[j], |
| 773 |
cur->datums[j], |
| 774 |
prev->datums[j]); |
| 775 |
if (DatumGetInt32(cmpval) != 0) |
| 776 |
{ |
| 777 |
is_distinct = true; |
| 778 |
break; |
| 779 |
} |
| 780 |
} |
| 781 |
|
| 782 |
/* |
| 783 |
* Only if the bound is distinct save it into a temporary array, i.e, |
| 784 |
* rbounds which is later copied into boundinfo datums array. |
| 785 |
*/ |
| 786 |
if (is_distinct) |
| 787 |
rbounds[k++] = all_bounds[i]; |
| 788 |
|
| 789 |
prev = cur; |
| 790 |
} |
| 791 |
|
| 792 |
pfree(all_bounds); |
| 793 |
|
| 794 |
/* Update ndatums to hold the count of distinct datums. */ |
| 795 |
ndatums = k; |
| 796 |
|
| 797 |
/* |
| 798 |
* Add datums to boundinfo. Canonical indexes are values ranging from 0 |
| 799 |
* to nparts - 1, assigned in that order to each partition's upper bound. |
| 800 |
* For 'datums' elements that are lower bounds, there is -1 in the |
| 801 |
* 'indexes' array to signify that no partition exists for the values less |
| 802 |
* than such a bound and greater than or equal to the previous upper |
| 803 |
* bound. |
| 804 |
*/ |
| 805 |
boundinfo->ndatums = ndatums; |
| 806 |
boundinfo->datums = (Datum **) palloc0(ndatums * sizeof(Datum *)); |
| 807 |
boundinfo->kind = (PartitionRangeDatumKind **) |
| 808 |
palloc(ndatums * |
| 809 |
sizeof(PartitionRangeDatumKind *)); |
| 810 |
boundinfo->interleaved_parts = NULL; |
| 811 |
|
| 812 |
/* |
| 813 |
* For range partitioning, an additional value of -1 is stored as the last |
| 814 |
* element of the indexes[] array. |
| 815 |
*/ |
| 816 |
boundinfo->nindexes = ndatums + 1; |
| 817 |
boundinfo->indexes = (int *) palloc((ndatums + 1) * sizeof(int)); |
| 818 |
|
| 819 |
/* |
| 820 |
* In the loop below, to save from allocating a series of small arrays, |
| 821 |
* here we just allocate a single array for Datums and another for |
| 822 |
* PartitionRangeDatumKinds, below we'll just assign a portion of these |
| 823 |
* arrays in each loop. |
| 824 |
*/ |
| 825 |
partnatts = key->partnatts; |
| 826 |
boundDatums = (Datum *) palloc(ndatums * partnatts * sizeof(Datum)); |
| 827 |
boundKinds = (PartitionRangeDatumKind *) palloc(ndatums * partnatts * |
| 828 |
sizeof(PartitionRangeDatumKind)); |
| 829 |
|
| 830 |
for (i = 0; i < ndatums; i++) |
| 831 |
{ |
| 832 |
int j; |
| 833 |
|
| 834 |
boundinfo->datums[i] = &boundDatums[i * partnatts]; |
| 835 |
boundinfo->kind[i] = &boundKinds[i * partnatts]; |
| 836 |
for (j = 0; j < partnatts; j++) |
| 837 |
{ |
| 838 |
if (rbounds[i]->kind[j] == PARTITION_RANGE_DATUM_VALUE) |
| 839 |
boundinfo->datums[i][j] = |
| 840 |
datumCopy(rbounds[i]->datums[j], |
| 841 |
key->parttypbyval[j], |
| 842 |
key->parttyplen[j]); |
| 843 |
boundinfo->kind[i][j] = rbounds[i]->kind[j]; |
| 844 |
} |
| 845 |
|
| 846 |
/* |
| 847 |
* There is no mapping for invalid indexes. |
| 848 |
* |
| 849 |
* Any lower bounds in the rbounds array have invalid indexes |
| 850 |
* assigned, because the values between the previous bound (if there |
| 851 |
* is one) and this (lower) bound are not part of the range of any |
| 852 |
* existing partition. |
| 853 |
*/ |
| 854 |
if (rbounds[i]->lower) |
| 855 |
boundinfo->indexes[i] = -1; |
| 856 |
else |
| 857 |
{ |
| 858 |
int orig_index = rbounds[i]->index; |
| 859 |
|
| 860 |
/* If the old index has no mapping, assign one */ |
| 861 |
if ((*mapping)[orig_index] == -1) |
| 862 |
(*mapping)[orig_index] = next_index++; |
| 863 |
|
| 864 |
boundinfo->indexes[i] = (*mapping)[orig_index]; |
| 865 |
} |
| 866 |
} |
| 867 |
|
| 868 |
pfree(rbounds); |
| 869 |
|
| 870 |
/* Set the canonical value for default_index, if any. */ |
| 871 |
if (default_index != -1) |
| 872 |
{ |
| 873 |
Assert(default_index >= 0 && (*mapping)[default_index] == -1); |
| 874 |
(*mapping)[default_index] = next_index++; |
| 875 |
boundinfo->default_index = (*mapping)[default_index]; |
| 876 |
} |
| 877 |
|
| 878 |
/* The extra -1 element. */ |
| 879 |
Assert(i == ndatums); |
| 880 |
boundinfo->indexes[i] = -1; |
| 881 |
|
| 882 |
/* All partitions must now have been assigned canonical indexes. */ |
| 883 |
Assert(next_index == nparts); |
| 884 |
return boundinfo; |
| 885 |
} |
| 886 |
|
| 887 |
/* |
| 888 |
* Are two partition bound collections logically equal? |
| 889 |
* |
| 890 |
* Used in the keep logic of relcache.c (ie, in RelationClearRelation()). |
| 891 |
* This is also useful when b1 and b2 are bound collections of two separate |
| 892 |
* relations, respectively, because PartitionBoundInfo is a canonical |
| 893 |
* representation of partition bounds. |
| 894 |
*/ |
| 895 |
bool |
| 896 |
partition_bounds_equal(int partnatts, int16 *parttyplen, bool *parttypbyval, |
| 897 |
PartitionBoundInfo b1, PartitionBoundInfo b2) |
| 898 |
{ |
| 899 |
int i; |
| 900 |
|
| 901 |
if (b1->strategy != b2->strategy) |
| 902 |
return false; |
| 903 |
|
| 904 |
if (b1->ndatums != b2->ndatums) |
| 905 |
return false; |
| 906 |
|
| 907 |
if (b1->nindexes != b2->nindexes) |
| 908 |
return false; |
| 909 |
|
| 910 |
if (b1->null_index != b2->null_index) |
| 911 |
return false; |
| 912 |
|
| 913 |
if (b1->default_index != b2->default_index) |
| 914 |
return false; |
| 915 |
|
| 916 |
/* For all partition strategies, the indexes[] arrays have to match */ |
| 917 |
for (i = 0; i < b1->nindexes; i++) |
| 918 |
{ |
| 919 |
if (b1->indexes[i] != b2->indexes[i]) |
| 920 |
return false; |
| 921 |
} |
| 922 |
|
| 923 |
/* Finally, compare the datums[] arrays */ |
| 924 |
if (b1->strategy == PARTITION_STRATEGY_HASH) |
| 925 |
{ |
| 926 |
/* |
| 927 |
* We arrange the partitions in the ascending order of their moduli |
| 928 |
* and remainders. Also every modulus is factor of next larger |
| 929 |
* modulus. Therefore we can safely store index of a given partition |
| 930 |
* in indexes array at remainder of that partition. Also entries at |
| 931 |
* (remainder + N * modulus) positions in indexes array are all same |
| 932 |
* for (modulus, remainder) specification for any partition. Thus the |
| 933 |
* datums arrays from the given bounds are the same, if and only if |
| 934 |
* their indexes arrays are the same. So, it suffices to compare the |
| 935 |
* indexes arrays. |
| 936 |
* |
| 937 |
* Nonetheless make sure that the bounds are indeed the same when the |
| 938 |
* indexes match. Hash partition bound stores modulus and remainder |
| 939 |
* at b1->datums[i][0] and b1->datums[i][1] position respectively. |
| 940 |
*/ |
| 941 |
#ifdef USE_ASSERT_CHECKING |
| 942 |
for (i = 0; i < b1->ndatums; i++) |
| 943 |
Assert((b1->datums[i][0] == b2->datums[i][0] && |
| 944 |
b1->datums[i][1] == b2->datums[i][1])); |
| 945 |
#endif |
| 946 |
} |
| 947 |
else |
| 948 |
{ |
| 949 |
for (i = 0; i < b1->ndatums; i++) |
| 950 |
{ |
| 951 |
int j; |
| 952 |
|
| 953 |
for (j = 0; j < partnatts; j++) |
| 954 |
{ |
| 955 |
/* For range partitions, the bounds might not be finite. */ |
| 956 |
if (b1->kind != NULL) |
| 957 |
{ |
| 958 |
/* The different kinds of bound all differ from each other */ |
| 959 |
if (b1->kind[i][j] != b2->kind[i][j]) |
| 960 |
return false; |
| 961 |
|
| 962 |
/* |
| 963 |
* Non-finite bounds are equal without further |
| 964 |
* examination. |
| 965 |
*/ |
| 966 |
if (b1->kind[i][j] != PARTITION_RANGE_DATUM_VALUE) |
| 967 |
continue; |
| 968 |
} |
| 969 |
|
| 970 |
/* |
| 971 |
* Compare the actual values. Note that it would be both |
| 972 |
* incorrect and unsafe to invoke the comparison operator |
| 973 |
* derived from the partitioning specification here. It would |
| 974 |
* be incorrect because we want the relcache entry to be |
| 975 |
* updated for ANY change to the partition bounds, not just |
| 976 |
* those that the partitioning operator thinks are |
| 977 |
* significant. It would be unsafe because we might reach |
| 978 |
* this code in the context of an aborted transaction, and an |
| 979 |
* arbitrary partitioning operator might not be safe in that |
| 980 |
* context. datumIsEqual() should be simple enough to be |
| 981 |
* safe. |
| 982 |
*/ |
| 983 |
if (!datumIsEqual(b1->datums[i][j], b2->datums[i][j], |
| 984 |
parttypbyval[j], parttyplen[j])) |
| 985 |
return false; |
| 986 |
} |
| 987 |
} |
| 988 |
} |
| 989 |
return true; |
| 990 |
} |
| 991 |
|
| 992 |
/* |
| 993 |
* Return a copy of given PartitionBoundInfo structure. The data types of bounds |
| 994 |
* are described by given partition key specification. |
| 995 |
* |
| 996 |
* Note: it's important that this function and its callees not do any catalog |
| 997 |
* access, nor anything else that would result in allocating memory other than |
| 998 |
* the returned data structure. Since this is called in a long-lived context, |
| 999 |
* that would result in unwanted memory leaks. |
| 1000 |
*/ |
| 1001 |
PartitionBoundInfo |
| 1002 |
partition_bounds_copy(PartitionBoundInfo src, |
| 1003 |
PartitionKey key) |
| 1004 |
{ |
| 1005 |
PartitionBoundInfo dest; |
| 1006 |
int i; |
| 1007 |
int ndatums; |
| 1008 |
int nindexes; |
| 1009 |
int partnatts; |
| 1010 |
bool hash_part; |
| 1011 |
int natts; |
| 1012 |
Datum *boundDatums; |
| 1013 |
|
| 1014 |
dest = (PartitionBoundInfo) palloc(sizeof(PartitionBoundInfoData)); |
| 1015 |
|
| 1016 |
dest->strategy = src->strategy; |
| 1017 |
ndatums = dest->ndatums = src->ndatums; |
| 1018 |
nindexes = dest->nindexes = src->nindexes; |
| 1019 |
partnatts = key->partnatts; |
| 1020 |
|
| 1021 |
/* List partitioned tables have only a single partition key. */ |
| 1022 |
Assert(key->strategy != PARTITION_STRATEGY_LIST || partnatts == 1); |
| 1023 |
|
| 1024 |
dest->datums = (Datum **) palloc(sizeof(Datum *) * ndatums); |
| 1025 |
|
| 1026 |
if (src->kind != NULL) |
| 1027 |
{ |
| 1028 |
PartitionRangeDatumKind *boundKinds; |
| 1029 |
|
| 1030 |
/* only RANGE partition should have a non-NULL kind */ |
| 1031 |
Assert(key->strategy == PARTITION_STRATEGY_RANGE); |
| 1032 |
|
| 1033 |
dest->kind = (PartitionRangeDatumKind **) palloc(ndatums * |
| 1034 |
sizeof(PartitionRangeDatumKind *)); |
| 1035 |
|
| 1036 |
/* |
| 1037 |
* In the loop below, to save from allocating a series of small arrays |
| 1038 |
* for storing the PartitionRangeDatumKind, we allocate a single chunk |
| 1039 |
* here and use a smaller portion of it for each datum. |
| 1040 |
*/ |
| 1041 |
boundKinds = (PartitionRangeDatumKind *) palloc(ndatums * partnatts * |
| 1042 |
sizeof(PartitionRangeDatumKind)); |
| 1043 |
|
| 1044 |
for (i = 0; i < ndatums; i++) |
| 1045 |
{ |
| 1046 |
dest->kind[i] = &boundKinds[i * partnatts]; |
| 1047 |
memcpy(dest->kind[i], src->kind[i], |
| 1048 |
sizeof(PartitionRangeDatumKind) * partnatts); |
| 1049 |
} |
| 1050 |
} |
| 1051 |
else |
| 1052 |
dest->kind = NULL; |
| 1053 |
|
| 1054 |
/* copy interleaved partitions for LIST partitioned tables */ |
| 1055 |
dest->interleaved_parts = bms_copy(src->interleaved_parts); |
| 1056 |
|
| 1057 |
/* |
| 1058 |
* For hash partitioning, datums array will have two elements - modulus |
| 1059 |
* and remainder. |
| 1060 |
*/ |
| 1061 |
hash_part = (key->strategy == PARTITION_STRATEGY_HASH); |
| 1062 |
natts = hash_part ? 2 : partnatts; |
| 1063 |
boundDatums = palloc(ndatums * natts * sizeof(Datum)); |
| 1064 |
|
| 1065 |
for (i = 0; i < ndatums; i++) |
| 1066 |
{ |
| 1067 |
int j; |
| 1068 |
|
| 1069 |
dest->datums[i] = &boundDatums[i * natts]; |
| 1070 |
|
| 1071 |
for (j = 0; j < natts; j++) |
| 1072 |
{ |
| 1073 |
bool byval; |
| 1074 |
int typlen; |
| 1075 |
|
| 1076 |
if (hash_part) |
| 1077 |
{ |
| 1078 |
typlen = sizeof(int32); /* Always int4 */ |
| 1079 |
byval = true; /* int4 is pass-by-value */ |
| 1080 |
} |
| 1081 |
else |
| 1082 |
{ |
| 1083 |
byval = key->parttypbyval[j]; |
| 1084 |
typlen = key->parttyplen[j]; |
| 1085 |
} |
| 1086 |
|
| 1087 |
if (dest->kind == NULL || |
| 1088 |
dest->kind[i][j] == PARTITION_RANGE_DATUM_VALUE) |
| 1089 |
dest->datums[i][j] = datumCopy(src->datums[i][j], |
| 1090 |
byval, typlen); |
| 1091 |
} |
| 1092 |
} |
| 1093 |
|
| 1094 |
dest->indexes = (int *) palloc(sizeof(int) * nindexes); |
| 1095 |
memcpy(dest->indexes, src->indexes, sizeof(int) * nindexes); |
| 1096 |
|
| 1097 |
dest->null_index = src->null_index; |
| 1098 |
dest->default_index = src->default_index; |
| 1099 |
|
| 1100 |
return dest; |
| 1101 |
} |
| 1102 |
|
| 1103 |
/* |
| 1104 |
* partition_bounds_merge |
| 1105 |
* Check to see whether every partition of 'outer_rel' matches/overlaps |
| 1106 |
* one partition of 'inner_rel' at most, and vice versa; and if so, build |
| 1107 |
* and return the partition bounds for a join relation between the rels, |
| 1108 |
* generating two lists of the matching/overlapping partitions, which are |
| 1109 |
* returned to *outer_parts and *inner_parts respectively. |
| 1110 |
* |
| 1111 |
* The lists contain the same number of partitions, and the partitions at the |
| 1112 |
* same positions in the lists indicate join pairs used for partitioned join. |
| 1113 |
* If a partition on one side matches/overlaps multiple partitions on the other |
| 1114 |
* side, this function returns NULL, setting *outer_parts and *inner_parts to |
| 1115 |
* NIL. |
| 1116 |
*/ |
| 1117 |
PartitionBoundInfo |
| 1118 |
partition_bounds_merge(int partnatts, |
| 1119 |
FmgrInfo *partsupfunc, Oid *partcollation, |
| 1120 |
RelOptInfo *outer_rel, RelOptInfo *inner_rel, |
| 1121 |
JoinType jointype, |
| 1122 |
List **outer_parts, List **inner_parts) |
| 1123 |
{ |
| 1124 |
/* |
| 1125 |
* Currently, this function is called only from try_partitionwise_join(), |
| 1126 |
* so the join type should be INNER, LEFT, FULL, SEMI, or ANTI. |
| 1127 |
*/ |
| 1128 |
Assert(jointype == JOIN_INNER || jointype == JOIN_LEFT || |
| 1129 |
jointype == JOIN_FULL || jointype == JOIN_SEMI || |
| 1130 |
jointype == JOIN_ANTI); |
| 1131 |
|
| 1132 |
/* The partitioning strategies should be the same. */ |
| 1133 |
Assert(outer_rel->boundinfo->strategy == inner_rel->boundinfo->strategy); |
| 1134 |
|
| 1135 |
*outer_parts = *inner_parts = NIL; |
| 1136 |
switch (outer_rel->boundinfo->strategy) |
| 1137 |
{ |
| 1138 |
case PARTITION_STRATEGY_HASH: |
| 1139 |
|
| 1140 |
/* |
| 1141 |
* For hash partitioned tables, we currently support partitioned |
| 1142 |
* join only when they have exactly the same partition bounds. |
| 1143 |
* |
| 1144 |
* XXX: it might be possible to relax the restriction to support |
| 1145 |
* cases where hash partitioned tables have missing partitions |
| 1146 |
* and/or different moduli, but it's not clear if it would be |
| 1147 |
* useful to support the former case since it's unusual to have |
| 1148 |
* missing partitions. On the other hand, it would be useful to |
| 1149 |
* support the latter case, but in that case, there is a high |
| 1150 |
* probability that a partition on one side will match multiple |
| 1151 |
* partitions on the other side, which is the scenario the current |
| 1152 |
* implementation of partitioned join can't handle. |
| 1153 |
*/ |
| 1154 |
return NULL; |
| 1155 |
|
| 1156 |
case PARTITION_STRATEGY_LIST: |
| 1157 |
return merge_list_bounds(partsupfunc, |
| 1158 |
partcollation, |
| 1159 |
outer_rel, |
| 1160 |
inner_rel, |
| 1161 |
jointype, |
| 1162 |
outer_parts, |
| 1163 |
inner_parts); |
| 1164 |
|
| 1165 |
case PARTITION_STRATEGY_RANGE: |
| 1166 |
return merge_range_bounds(partnatts, |
| 1167 |
partsupfunc, |
| 1168 |
partcollation, |
| 1169 |
outer_rel, |
| 1170 |
inner_rel, |
| 1171 |
jointype, |
| 1172 |
outer_parts, |
| 1173 |
inner_parts); |
| 1174 |
} |
| 1175 |
|
| 1176 |
return NULL; |
| 1177 |
} |
| 1178 |
|
| 1179 |
/* |
| 1180 |
* merge_list_bounds |
| 1181 |
* Create the partition bounds for a join relation between list |
| 1182 |
* partitioned tables, if possible |
| 1183 |
* |
| 1184 |
* In this function we try to find sets of matching partitions from both sides |
| 1185 |
* by comparing list values stored in their partition bounds. Since the list |
| 1186 |
* values appear in the ascending order, an algorithm similar to merge join is |
| 1187 |
* used for that. If a partition on one side doesn't have a matching |
| 1188 |
* partition on the other side, the algorithm tries to match it with the |
| 1189 |
* default partition on the other side if any; if not, the algorithm tries to |
| 1190 |
* match it with a dummy partition on the other side if it's on the |
| 1191 |
* non-nullable side of an outer join. Also, if both sides have the default |
| 1192 |
* partitions, the algorithm tries to match them with each other. We give up |
| 1193 |
* if the algorithm finds a partition matching multiple partitions on the |
| 1194 |
* other side, which is the scenario the current implementation of partitioned |
| 1195 |
* join can't handle. |
| 1196 |
*/ |
| 1197 |
static PartitionBoundInfo |
| 1198 |
merge_list_bounds(FmgrInfo *partsupfunc, Oid *partcollation, |
| 1199 |
RelOptInfo *outer_rel, RelOptInfo *inner_rel, |
| 1200 |
JoinType jointype, |
| 1201 |
List **outer_parts, List **inner_parts) |
| 1202 |
{ |
| 1203 |
PartitionBoundInfo merged_bounds = NULL; |
| 1204 |
PartitionBoundInfo outer_bi = outer_rel->boundinfo; |
| 1205 |
PartitionBoundInfo inner_bi = inner_rel->boundinfo; |
| 1206 |
bool outer_has_default = partition_bound_has_default(outer_bi); |
| 1207 |
bool inner_has_default = partition_bound_has_default(inner_bi); |
| 1208 |
int outer_default = outer_bi->default_index; |
| 1209 |
int inner_default = inner_bi->default_index; |
| 1210 |
bool outer_has_null = partition_bound_accepts_nulls(outer_bi); |
| 1211 |
bool inner_has_null = partition_bound_accepts_nulls(inner_bi); |
| 1212 |
PartitionMap outer_map; |
| 1213 |
PartitionMap inner_map; |
| 1214 |
int outer_pos; |
| 1215 |
int inner_pos; |
| 1216 |
int next_index = 0; |
| 1217 |
int null_index = -1; |
| 1218 |
int default_index = -1; |
| 1219 |
List *merged_datums = NIL; |
| 1220 |
List *merged_indexes = NIL; |
| 1221 |
|
| 1222 |
Assert(*outer_parts == NIL); |
| 1223 |
Assert(*inner_parts == NIL); |
| 1224 |
Assert(outer_bi->strategy == inner_bi->strategy && |
| 1225 |
outer_bi->strategy == PARTITION_STRATEGY_LIST); |
| 1226 |
/* List partitioning doesn't require kinds. */ |
| 1227 |
Assert(!outer_bi->kind && !inner_bi->kind); |
| 1228 |
|
| 1229 |
init_partition_map(outer_rel, &outer_map); |
| 1230 |
init_partition_map(inner_rel, &inner_map); |
| 1231 |
|
| 1232 |
/* |
| 1233 |
* If the default partitions (if any) have been proven empty, deem them |
| 1234 |
* non-existent. |
| 1235 |
*/ |
| 1236 |
if (outer_has_default && is_dummy_partition(outer_rel, outer_default)) |
| 1237 |
outer_has_default = false; |
| 1238 |
if (inner_has_default && is_dummy_partition(inner_rel, inner_default)) |
| 1239 |
inner_has_default = false; |
| 1240 |
|
| 1241 |
/* |
| 1242 |
* Merge partitions from both sides. In each iteration we compare a pair |
| 1243 |
* of list values, one from each side, and decide whether the |
| 1244 |
* corresponding partitions match or not. If the two values match |
| 1245 |
* exactly, move to the next pair of list values, otherwise move to the |
| 1246 |
* next list value on the side with a smaller list value. |
| 1247 |
*/ |
| 1248 |
outer_pos = inner_pos = 0; |
| 1249 |
while (outer_pos < outer_bi->ndatums || inner_pos < inner_bi->ndatums) |
| 1250 |
{ |
| 1251 |
int outer_index = -1; |
| 1252 |
int inner_index = -1; |
| 1253 |
Datum *outer_datums; |
| 1254 |
Datum *inner_datums; |
| 1255 |
int cmpval; |
| 1256 |
Datum *merged_datum = NULL; |
| 1257 |
int merged_index = -1; |
| 1258 |
|
| 1259 |
if (outer_pos < outer_bi->ndatums) |
| 1260 |
{ |
| 1261 |
/* |
| 1262 |
* If the partition on the outer side has been proven empty, |
| 1263 |
* ignore it and move to the next datum on the outer side. |
| 1264 |
*/ |
| 1265 |
outer_index = outer_bi->indexes[outer_pos]; |
| 1266 |
if (is_dummy_partition(outer_rel, outer_index)) |
| 1267 |
{ |
| 1268 |
outer_pos++; |
| 1269 |
continue; |
| 1270 |
} |
| 1271 |
} |
| 1272 |
if (inner_pos < inner_bi->ndatums) |
| 1273 |
{ |
| 1274 |
/* |
| 1275 |
* If the partition on the inner side has been proven empty, |
| 1276 |
* ignore it and move to the next datum on the inner side. |
| 1277 |
*/ |
| 1278 |
inner_index = inner_bi->indexes[inner_pos]; |
| 1279 |
if (is_dummy_partition(inner_rel, inner_index)) |
| 1280 |
{ |
| 1281 |
inner_pos++; |
| 1282 |
continue; |
| 1283 |
} |
| 1284 |
} |
| 1285 |
|
| 1286 |
/* Get the list values. */ |
| 1287 |
outer_datums = outer_pos < outer_bi->ndatums ? |
| 1288 |
outer_bi->datums[outer_pos] : NULL; |
| 1289 |
inner_datums = inner_pos < inner_bi->ndatums ? |
| 1290 |
inner_bi->datums[inner_pos] : NULL; |
| 1291 |
|
| 1292 |
/* |
| 1293 |
* We run this loop till both sides finish. This allows us to avoid |
| 1294 |
* duplicating code to handle the remaining values on the side which |
| 1295 |
* finishes later. For that we set the comparison parameter cmpval in |
| 1296 |
* such a way that it appears as if the side which finishes earlier |
| 1297 |
* has an extra value higher than any other value on the unfinished |
| 1298 |
* side. That way we advance the values on the unfinished side till |
| 1299 |
* all of its values are exhausted. |
| 1300 |
*/ |
| 1301 |
if (outer_pos >= outer_bi->ndatums) |
| 1302 |
cmpval = 1; |
| 1303 |
else if (inner_pos >= inner_bi->ndatums) |
| 1304 |
cmpval = -1; |
| 1305 |
else |
| 1306 |
{ |
| 1307 |
Assert(outer_datums != NULL && inner_datums != NULL); |
| 1308 |
cmpval = DatumGetInt32(FunctionCall2Coll(&partsupfunc[0], |
| 1309 |
partcollation[0], |
| 1310 |
outer_datums[0], |
| 1311 |
inner_datums[0])); |
| 1312 |
} |
| 1313 |
|
| 1314 |
if (cmpval == 0) |
| 1315 |
{ |
| 1316 |
/* Two list values match exactly. */ |
| 1317 |
Assert(outer_pos < outer_bi->ndatums); |
| 1318 |
Assert(inner_pos < inner_bi->ndatums); |
| 1319 |
Assert(outer_index >= 0); |
| 1320 |
Assert(inner_index >= 0); |
| 1321 |
|
| 1322 |
/* |
| 1323 |
* Try merging both partitions. If successful, add the list value |
| 1324 |
* and index of the merged partition below. |
| 1325 |
*/ |
| 1326 |
merged_index = merge_matching_partitions(&outer_map, &inner_map, |
| 1327 |
outer_index, inner_index, |
| 1328 |
&next_index); |
| 1329 |
if (merged_index == -1) |
| 1330 |
goto cleanup; |
| 1331 |
|
| 1332 |
merged_datum = outer_datums; |
| 1333 |
|
| 1334 |
/* Move to the next pair of list values. */ |
| 1335 |
outer_pos++; |
| 1336 |
inner_pos++; |
| 1337 |
} |
| 1338 |
else if (cmpval < 0) |
| 1339 |
{ |
| 1340 |
/* A list value missing from the inner side. */ |
| 1341 |
Assert(outer_pos < outer_bi->ndatums); |
| 1342 |
|
| 1343 |
/* |
| 1344 |
* If the inner side has the default partition, or this is an |
| 1345 |
* outer join, try to assign a merged partition to the outer |
| 1346 |
* partition (see process_outer_partition()). Otherwise, the |
| 1347 |
* outer partition will not contribute to the result. |
| 1348 |
*/ |
| 1349 |
if (inner_has_default || IS_OUTER_JOIN(jointype)) |
| 1350 |
{ |
| 1351 |
/* Get the outer partition. */ |
| 1352 |
outer_index = outer_bi->indexes[outer_pos]; |
| 1353 |
Assert(outer_index >= 0); |
| 1354 |
merged_index = process_outer_partition(&outer_map, |
| 1355 |
&inner_map, |
| 1356 |
outer_has_default, |
| 1357 |
inner_has_default, |
| 1358 |
outer_index, |
| 1359 |
inner_default, |
| 1360 |
jointype, |
| 1361 |
&next_index, |
| 1362 |
&default_index); |
| 1363 |
if (merged_index == -1) |
| 1364 |
goto cleanup; |
| 1365 |
merged_datum = outer_datums; |
| 1366 |
} |
| 1367 |
|
| 1368 |
/* Move to the next list value on the outer side. */ |
| 1369 |
outer_pos++; |
| 1370 |
} |
| 1371 |
else |
| 1372 |
{ |
| 1373 |
/* A list value missing from the outer side. */ |
| 1374 |
Assert(cmpval > 0); |
| 1375 |
Assert(inner_pos < inner_bi->ndatums); |
| 1376 |
|
| 1377 |
/* |
| 1378 |
* If the outer side has the default partition, or this is a FULL |
| 1379 |
* join, try to assign a merged partition to the inner partition |
| 1380 |
* (see process_inner_partition()). Otherwise, the inner |
| 1381 |
* partition will not contribute to the result. |
| 1382 |
*/ |
| 1383 |
if (outer_has_default || jointype == JOIN_FULL) |
| 1384 |
{ |
| 1385 |
/* Get the inner partition. */ |
| 1386 |
inner_index = inner_bi->indexes[inner_pos]; |
| 1387 |
Assert(inner_index >= 0); |
| 1388 |
merged_index = process_inner_partition(&outer_map, |
| 1389 |
&inner_map, |
| 1390 |
outer_has_default, |
| 1391 |
inner_has_default, |
| 1392 |
inner_index, |
| 1393 |
outer_default, |
| 1394 |
jointype, |
| 1395 |
&next_index, |
| 1396 |
&default_index); |
| 1397 |
if (merged_index == -1) |
| 1398 |
goto cleanup; |
| 1399 |
merged_datum = inner_datums; |
| 1400 |
} |
| 1401 |
|
| 1402 |
/* Move to the next list value on the inner side. */ |
| 1403 |
inner_pos++; |
| 1404 |
} |
| 1405 |
|
| 1406 |
/* |
| 1407 |
* If we assigned a merged partition, add the list value and index of |
| 1408 |
* the merged partition if appropriate. |
| 1409 |
*/ |
| 1410 |
if (merged_index >= 0 && merged_index != default_index) |
| 1411 |
{ |
| 1412 |
merged_datums = lappend(merged_datums, merged_datum); |
| 1413 |
merged_indexes = lappend_int(merged_indexes, merged_index); |
| 1414 |
} |
| 1415 |
} |
| 1416 |
|
| 1417 |
/* |
| 1418 |
* If the NULL partitions (if any) have been proven empty, deem them |
| 1419 |
* non-existent. |
| 1420 |
*/ |
| 1421 |
if (outer_has_null && |
| 1422 |
is_dummy_partition(outer_rel, outer_bi->null_index)) |
| 1423 |
outer_has_null = false; |
| 1424 |
if (inner_has_null && |
| 1425 |
is_dummy_partition(inner_rel, inner_bi->null_index)) |
| 1426 |
inner_has_null = false; |
| 1427 |
|
| 1428 |
/* Merge the NULL partitions if any. */ |
| 1429 |
if (outer_has_null || inner_has_null) |
| 1430 |
merge_null_partitions(&outer_map, &inner_map, |
| 1431 |
outer_has_null, inner_has_null, |
| 1432 |
outer_bi->null_index, inner_bi->null_index, |
| 1433 |
jointype, &next_index, &null_index); |
| 1434 |
else |
| 1435 |
Assert(null_index == -1); |
| 1436 |
|
| 1437 |
/* Merge the default partitions if any. */ |
| 1438 |
if (outer_has_default || inner_has_default) |
| 1439 |
merge_default_partitions(&outer_map, &inner_map, |
| 1440 |
outer_has_default, inner_has_default, |
| 1441 |
outer_default, inner_default, |
| 1442 |
jointype, &next_index, &default_index); |
| 1443 |
else |
| 1444 |
Assert(default_index == -1); |
| 1445 |
|
| 1446 |
/* If we have merged partitions, create the partition bounds. */ |
| 1447 |
if (next_index > 0) |
| 1448 |
{ |
| 1449 |
/* Fix the merged_indexes list if necessary. */ |
| 1450 |
if (outer_map.did_remapping || inner_map.did_remapping) |
| 1451 |
{ |
| 1452 |
Assert(jointype == JOIN_FULL); |
| 1453 |
fix_merged_indexes(&outer_map, &inner_map, |
| 1454 |
next_index, merged_indexes); |
| 1455 |
} |
| 1456 |
|
| 1457 |
/* Use maps to match partitions from inputs. */ |
| 1458 |
generate_matching_part_pairs(outer_rel, inner_rel, |
| 1459 |
&outer_map, &inner_map, |
| 1460 |
next_index, |
| 1461 |
outer_parts, inner_parts); |
| 1462 |
Assert(*outer_parts != NIL); |
| 1463 |
Assert(*inner_parts != NIL); |
| 1464 |
Assert(list_length(*outer_parts) == list_length(*inner_parts)); |
| 1465 |
Assert(list_length(*outer_parts) <= next_index); |
| 1466 |
|
| 1467 |
/* Make a PartitionBoundInfo struct to return. */ |
| 1468 |
merged_bounds = build_merged_partition_bounds(outer_bi->strategy, |
| 1469 |
merged_datums, |
| 1470 |
NIL, |
| 1471 |
merged_indexes, |
| 1472 |
null_index, |
| 1473 |
default_index); |
| 1474 |
Assert(merged_bounds); |
| 1475 |
} |
| 1476 |
|
| 1477 |
cleanup: |
| 1478 |
/* Free local memory before returning. */ |
| 1479 |
list_free(merged_datums); |
| 1480 |
list_free(merged_indexes); |
| 1481 |
free_partition_map(&outer_map); |
| 1482 |
free_partition_map(&inner_map); |
| 1483 |
|
| 1484 |
return merged_bounds; |
| 1485 |
} |
| 1486 |
|
| 1487 |
/* |
| 1488 |
* merge_range_bounds |
| 1489 |
* Create the partition bounds for a join relation between range |
| 1490 |
* partitioned tables, if possible |
| 1491 |
* |
| 1492 |
* In this function we try to find sets of overlapping partitions from both |
| 1493 |
* sides by comparing ranges stored in their partition bounds. Since the |
| 1494 |
* ranges appear in the ascending order, an algorithm similar to merge join is |
| 1495 |
* used for that. If a partition on one side doesn't have an overlapping |
| 1496 |
* partition on the other side, the algorithm tries to match it with the |
| 1497 |
* default partition on the other side if any; if not, the algorithm tries to |
| 1498 |
* match it with a dummy partition on the other side if it's on the |
| 1499 |
* non-nullable side of an outer join. Also, if both sides have the default |
| 1500 |
* partitions, the algorithm tries to match them with each other. We give up |
| 1501 |
* if the algorithm finds a partition overlapping multiple partitions on the |
| 1502 |
* other side, which is the scenario the current implementation of partitioned |
| 1503 |
* join can't handle. |
| 1504 |
*/ |
| 1505 |
static PartitionBoundInfo |
| 1506 |
merge_range_bounds(int partnatts, FmgrInfo *partsupfuncs, |
| 1507 |
Oid *partcollations, |
| 1508 |
RelOptInfo *outer_rel, RelOptInfo *inner_rel, |
| 1509 |
JoinType jointype, |
| 1510 |
List **outer_parts, List **inner_parts) |
| 1511 |
{ |
| 1512 |
PartitionBoundInfo merged_bounds = NULL; |
| 1513 |
PartitionBoundInfo outer_bi = outer_rel->boundinfo; |
| 1514 |
PartitionBoundInfo inner_bi = inner_rel->boundinfo; |
| 1515 |
bool outer_has_default = partition_bound_has_default(outer_bi); |
| 1516 |
bool inner_has_default = partition_bound_has_default(inner_bi); |
| 1517 |
int outer_default = outer_bi->default_index; |
| 1518 |
int inner_default = inner_bi->default_index; |
| 1519 |
PartitionMap outer_map; |
| 1520 |
PartitionMap inner_map; |
| 1521 |
int outer_index; |
| 1522 |
int inner_index; |
| 1523 |
int outer_lb_pos; |
| 1524 |
int inner_lb_pos; |
| 1525 |
PartitionRangeBound outer_lb; |
| 1526 |
PartitionRangeBound outer_ub; |
| 1527 |
PartitionRangeBound inner_lb; |
| 1528 |
PartitionRangeBound inner_ub; |
| 1529 |
int next_index = 0; |
| 1530 |
int default_index = -1; |
| 1531 |
List *merged_datums = NIL; |
| 1532 |
List *merged_kinds = NIL; |
| 1533 |
List *merged_indexes = NIL; |
| 1534 |
|
| 1535 |
Assert(*outer_parts == NIL); |
| 1536 |
Assert(*inner_parts == NIL); |
| 1537 |
Assert(outer_bi->strategy == inner_bi->strategy && |
| 1538 |
outer_bi->strategy == PARTITION_STRATEGY_RANGE); |
| 1539 |
|
| 1540 |
init_partition_map(outer_rel, &outer_map); |
| 1541 |
init_partition_map(inner_rel, &inner_map); |
| 1542 |
|
| 1543 |
/* |
| 1544 |
* If the default partitions (if any) have been proven empty, deem them |
| 1545 |
* non-existent. |
| 1546 |
*/ |
| 1547 |
if (outer_has_default && is_dummy_partition(outer_rel, outer_default)) |
| 1548 |
outer_has_default = false; |
| 1549 |
if (inner_has_default && is_dummy_partition(inner_rel, inner_default)) |
| 1550 |
inner_has_default = false; |
| 1551 |
|
| 1552 |
/* |
| 1553 |
* Merge partitions from both sides. In each iteration we compare a pair |
| 1554 |
* of ranges, one from each side, and decide whether the corresponding |
| 1555 |
* partitions match or not. If the two ranges overlap, move to the next |
| 1556 |
* pair of ranges, otherwise move to the next range on the side with a |
| 1557 |
* lower range. outer_lb_pos/inner_lb_pos keep track of the positions of |
| 1558 |
* lower bounds in the datums arrays in the outer/inner |
| 1559 |
* PartitionBoundInfos respectively. |
| 1560 |
*/ |
| 1561 |
outer_lb_pos = inner_lb_pos = 0; |
| 1562 |
outer_index = get_range_partition(outer_rel, outer_bi, &outer_lb_pos, |
| 1563 |
&outer_lb, &outer_ub); |
| 1564 |
inner_index = get_range_partition(inner_rel, inner_bi, &inner_lb_pos, |
| 1565 |
&inner_lb, &inner_ub); |
| 1566 |
while (outer_index >= 0 || inner_index >= 0) |
| 1567 |
{ |
| 1568 |
bool overlap; |
| 1569 |
int ub_cmpval; |
| 1570 |
int lb_cmpval; |
| 1571 |
PartitionRangeBound merged_lb = {-1, NULL, NULL, true}; |
| 1572 |
PartitionRangeBound merged_ub = {-1, NULL, NULL, false}; |
| 1573 |
int merged_index = -1; |
| 1574 |
|
| 1575 |
/* |
| 1576 |
* We run this loop till both sides finish. This allows us to avoid |
| 1577 |
* duplicating code to handle the remaining ranges on the side which |
| 1578 |
* finishes later. For that we set the comparison parameter cmpval in |
| 1579 |
* such a way that it appears as if the side which finishes earlier |
| 1580 |
* has an extra range higher than any other range on the unfinished |
| 1581 |
* side. That way we advance the ranges on the unfinished side till |
| 1582 |
* all of its ranges are exhausted. |
| 1583 |
*/ |
| 1584 |
if (outer_index == -1) |
| 1585 |
{ |
| 1586 |
overlap = false; |
| 1587 |
lb_cmpval = 1; |
| 1588 |
ub_cmpval = 1; |
| 1589 |
} |
| 1590 |
else if (inner_index == -1) |
| 1591 |
{ |
| 1592 |
overlap = false; |
| 1593 |
lb_cmpval = -1; |
| 1594 |
ub_cmpval = -1; |
| 1595 |
} |
| 1596 |
else |
| 1597 |
overlap = compare_range_partitions(partnatts, partsupfuncs, |
| 1598 |
partcollations, |
| 1599 |
&outer_lb, &outer_ub, |
| 1600 |
&inner_lb, &inner_ub, |
| 1601 |
&lb_cmpval, &ub_cmpval); |
| 1602 |
|
| 1603 |
if (overlap) |
| 1604 |
{ |
| 1605 |
/* Two ranges overlap; form a join pair. */ |
| 1606 |
|
| 1607 |
PartitionRangeBound save_outer_ub; |
| 1608 |
PartitionRangeBound save_inner_ub; |
| 1609 |
|
| 1610 |
/* Both partitions should not have been merged yet. */ |
| 1611 |
Assert(outer_index >= 0); |
| 1612 |
Assert(outer_map.merged_indexes[outer_index] == -1 && |
| 1613 |
outer_map.merged[outer_index] == false); |
| 1614 |
Assert(inner_index >= 0); |
| 1615 |
Assert(inner_map.merged_indexes[inner_index] == -1 && |
| 1616 |
inner_map.merged[inner_index] == false); |
| 1617 |
|
| 1618 |
/* |
| 1619 |
* Get the index of the merged partition. Both partitions aren't |
| 1620 |
* merged yet, so the partitions should be merged successfully. |
| 1621 |
*/ |
| 1622 |
merged_index = merge_matching_partitions(&outer_map, &inner_map, |
| 1623 |
outer_index, inner_index, |
| 1624 |
&next_index); |
| 1625 |
Assert(merged_index >= 0); |
| 1626 |
|
| 1627 |
/* Get the range bounds of the merged partition. */ |
| 1628 |
get_merged_range_bounds(partnatts, partsupfuncs, |
| 1629 |
partcollations, jointype, |
| 1630 |
&outer_lb, &outer_ub, |
| 1631 |
&inner_lb, &inner_ub, |
| 1632 |
lb_cmpval, ub_cmpval, |
| 1633 |
&merged_lb, &merged_ub); |
| 1634 |
|
| 1635 |
/* Save the upper bounds of both partitions for use below. */ |
| 1636 |
save_outer_ub = outer_ub; |
| 1637 |
save_inner_ub = inner_ub; |
| 1638 |
|
| 1639 |
/* Move to the next pair of ranges. */ |
| 1640 |
outer_index = get_range_partition(outer_rel, outer_bi, &outer_lb_pos, |
| 1641 |
&outer_lb, &outer_ub); |
| 1642 |
inner_index = get_range_partition(inner_rel, inner_bi, &inner_lb_pos, |
| 1643 |
&inner_lb, &inner_ub); |
| 1644 |
|
| 1645 |
/* |
| 1646 |
* If the range of a partition on one side overlaps the range of |
| 1647 |
* the next partition on the other side, that will cause the |
| 1648 |
* partition on one side to match at least two partitions on the |
| 1649 |
* other side, which is the case that we currently don't support |
| 1650 |
* partitioned join for; give up. |
| 1651 |
*/ |
| 1652 |
if (ub_cmpval > 0 && inner_index >= 0 && |
| 1653 |
compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 1654 |
&save_outer_ub, &inner_lb) > 0) |
| 1655 |
goto cleanup; |
| 1656 |
if (ub_cmpval < 0 && outer_index >= 0 && |
| 1657 |
compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 1658 |
&outer_lb, &save_inner_ub) < 0) |
| 1659 |
goto cleanup; |
| 1660 |
|
| 1661 |
/* |
| 1662 |
* A row from a non-overlapping portion (if any) of a partition on |
| 1663 |
* one side might find its join partner in the default partition |
| 1664 |
* (if any) on the other side, causing the same situation as |
| 1665 |
* above; give up in that case. |
| 1666 |
*/ |
| 1667 |
if ((outer_has_default && (lb_cmpval > 0 || ub_cmpval < 0)) || |
| 1668 |
(inner_has_default && (lb_cmpval < 0 || ub_cmpval > 0))) |
| 1669 |
goto cleanup; |
| 1670 |
} |
| 1671 |
else if (ub_cmpval < 0) |
| 1672 |
{ |
| 1673 |
/* A non-overlapping outer range. */ |
| 1674 |
|
| 1675 |
/* The outer partition should not have been merged yet. */ |
| 1676 |
Assert(outer_index >= 0); |
| 1677 |
Assert(outer_map.merged_indexes[outer_index] == -1 && |
| 1678 |
outer_map.merged[outer_index] == false); |
| 1679 |
|
| 1680 |
/* |
| 1681 |
* If the inner side has the default partition, or this is an |
| 1682 |
* outer join, try to assign a merged partition to the outer |
| 1683 |
* partition (see process_outer_partition()). Otherwise, the |
| 1684 |
* outer partition will not contribute to the result. |
| 1685 |
*/ |
| 1686 |
if (inner_has_default || IS_OUTER_JOIN(jointype)) |
| 1687 |
{ |
| 1688 |
merged_index = process_outer_partition(&outer_map, |
| 1689 |
&inner_map, |
| 1690 |
outer_has_default, |
| 1691 |
inner_has_default, |
| 1692 |
outer_index, |
| 1693 |
inner_default, |
| 1694 |
jointype, |
| 1695 |
&next_index, |
| 1696 |
&default_index); |
| 1697 |
if (merged_index == -1) |
| 1698 |
goto cleanup; |
| 1699 |
merged_lb = outer_lb; |
| 1700 |
merged_ub = outer_ub; |
| 1701 |
} |
| 1702 |
|
| 1703 |
/* Move to the next range on the outer side. */ |
| 1704 |
outer_index = get_range_partition(outer_rel, outer_bi, &outer_lb_pos, |
| 1705 |
&outer_lb, &outer_ub); |
| 1706 |
} |
| 1707 |
else |
| 1708 |
{ |
| 1709 |
/* A non-overlapping inner range. */ |
| 1710 |
Assert(ub_cmpval > 0); |
| 1711 |
|
| 1712 |
/* The inner partition should not have been merged yet. */ |
| 1713 |
Assert(inner_index >= 0); |
| 1714 |
Assert(inner_map.merged_indexes[inner_index] == -1 && |
| 1715 |
inner_map.merged[inner_index] == false); |
| 1716 |
|
| 1717 |
/* |
| 1718 |
* If the outer side has the default partition, or this is a FULL |
| 1719 |
* join, try to assign a merged partition to the inner partition |
| 1720 |
* (see process_inner_partition()). Otherwise, the inner |
| 1721 |
* partition will not contribute to the result. |
| 1722 |
*/ |
| 1723 |
if (outer_has_default || jointype == JOIN_FULL) |
| 1724 |
{ |
| 1725 |
merged_index = process_inner_partition(&outer_map, |
| 1726 |
&inner_map, |
| 1727 |
outer_has_default, |
| 1728 |
inner_has_default, |
| 1729 |
inner_index, |
| 1730 |
outer_default, |
| 1731 |
jointype, |
| 1732 |
&next_index, |
| 1733 |
&default_index); |
| 1734 |
if (merged_index == -1) |
| 1735 |
goto cleanup; |
| 1736 |
merged_lb = inner_lb; |
| 1737 |
merged_ub = inner_ub; |
| 1738 |
} |
| 1739 |
|
| 1740 |
/* Move to the next range on the inner side. */ |
| 1741 |
inner_index = get_range_partition(inner_rel, inner_bi, &inner_lb_pos, |
| 1742 |
&inner_lb, &inner_ub); |
| 1743 |
} |
| 1744 |
|
| 1745 |
/* |
| 1746 |
* If we assigned a merged partition, add the range bounds and index |
| 1747 |
* of the merged partition if appropriate. |
| 1748 |
*/ |
| 1749 |
if (merged_index >= 0 && merged_index != default_index) |
| 1750 |
add_merged_range_bounds(partnatts, partsupfuncs, partcollations, |
| 1751 |
&merged_lb, &merged_ub, merged_index, |
| 1752 |
&merged_datums, &merged_kinds, |
| 1753 |
&merged_indexes); |
| 1754 |
} |
| 1755 |
|
| 1756 |
/* Merge the default partitions if any. */ |
| 1757 |
if (outer_has_default || inner_has_default) |
| 1758 |
merge_default_partitions(&outer_map, &inner_map, |
| 1759 |
outer_has_default, inner_has_default, |
| 1760 |
outer_default, inner_default, |
| 1761 |
jointype, &next_index, &default_index); |
| 1762 |
else |
| 1763 |
Assert(default_index == -1); |
| 1764 |
|
| 1765 |
/* If we have merged partitions, create the partition bounds. */ |
| 1766 |
if (next_index > 0) |
| 1767 |
{ |
| 1768 |
/* |
| 1769 |
* Unlike the case of list partitioning, we wouldn't have re-merged |
| 1770 |
* partitions, so did_remapping should be left alone. |
| 1771 |
*/ |
| 1772 |
Assert(!outer_map.did_remapping); |
| 1773 |
Assert(!inner_map.did_remapping); |
| 1774 |
|
| 1775 |
/* Use maps to match partitions from inputs. */ |
| 1776 |
generate_matching_part_pairs(outer_rel, inner_rel, |
| 1777 |
&outer_map, &inner_map, |
| 1778 |
next_index, |
| 1779 |
outer_parts, inner_parts); |
| 1780 |
Assert(*outer_parts != NIL); |
| 1781 |
Assert(*inner_parts != NIL); |
| 1782 |
Assert(list_length(*outer_parts) == list_length(*inner_parts)); |
| 1783 |
Assert(list_length(*outer_parts) == next_index); |
| 1784 |
|
| 1785 |
/* Make a PartitionBoundInfo struct to return. */ |
| 1786 |
merged_bounds = build_merged_partition_bounds(outer_bi->strategy, |
| 1787 |
merged_datums, |
| 1788 |
merged_kinds, |
| 1789 |
merged_indexes, |
| 1790 |
-1, |
| 1791 |
default_index); |
| 1792 |
Assert(merged_bounds); |
| 1793 |
} |
| 1794 |
|
| 1795 |
cleanup: |
| 1796 |
/* Free local memory before returning. */ |
| 1797 |
list_free(merged_datums); |
| 1798 |
list_free(merged_kinds); |
| 1799 |
list_free(merged_indexes); |
| 1800 |
free_partition_map(&outer_map); |
| 1801 |
free_partition_map(&inner_map); |
| 1802 |
|
| 1803 |
return merged_bounds; |
| 1804 |
} |
| 1805 |
|
| 1806 |
/* |
| 1807 |
* init_partition_map |
| 1808 |
* Initialize a PartitionMap struct for given relation |
| 1809 |
*/ |
| 1810 |
static void |
| 1811 |
init_partition_map(RelOptInfo *rel, PartitionMap *map) |
| 1812 |
{ |
| 1813 |
int nparts = rel->nparts; |
| 1814 |
int i; |
| 1815 |
|
| 1816 |
map->nparts = nparts; |
| 1817 |
map->merged_indexes = (int *) palloc(sizeof(int) * nparts); |
| 1818 |
map->merged = (bool *) palloc(sizeof(bool) * nparts); |
| 1819 |
map->did_remapping = false; |
| 1820 |
map->old_indexes = (int *) palloc(sizeof(int) * nparts); |
| 1821 |
for (i = 0; i < nparts; i++) |
| 1822 |
{ |
| 1823 |
map->merged_indexes[i] = map->old_indexes[i] = -1; |
| 1824 |
map->merged[i] = false; |
| 1825 |
} |
| 1826 |
} |
| 1827 |
|
| 1828 |
/* |
| 1829 |
* free_partition_map |
| 1830 |
*/ |
| 1831 |
static void |
| 1832 |
free_partition_map(PartitionMap *map) |
| 1833 |
{ |
| 1834 |
pfree(map->merged_indexes); |
| 1835 |
pfree(map->merged); |
| 1836 |
pfree(map->old_indexes); |
| 1837 |
} |
| 1838 |
|
| 1839 |
/* |
| 1840 |
* is_dummy_partition --- has partition been proven empty? |
| 1841 |
*/ |
| 1842 |
static bool |
| 1843 |
is_dummy_partition(RelOptInfo *rel, int part_index) |
| 1844 |
{ |
| 1845 |
RelOptInfo *part_rel; |
| 1846 |
|
| 1847 |
Assert(part_index >= 0); |
| 1848 |
part_rel = rel->part_rels[part_index]; |
| 1849 |
if (part_rel == NULL || IS_DUMMY_REL(part_rel)) |
| 1850 |
return true; |
| 1851 |
return false; |
| 1852 |
} |
| 1853 |
|
| 1854 |
/* |
| 1855 |
* merge_matching_partitions |
| 1856 |
* Try to merge given outer/inner partitions, and return the index of a |
| 1857 |
* merged partition produced from them if successful, -1 otherwise |
| 1858 |
* |
| 1859 |
* If the merged partition is newly created, *next_index is incremented. |
| 1860 |
*/ |
| 1861 |
static int |
| 1862 |
merge_matching_partitions(PartitionMap *outer_map, PartitionMap *inner_map, |
| 1863 |
int outer_index, int inner_index, int *next_index) |
| 1864 |
{ |
| 1865 |
int outer_merged_index; |
| 1866 |
int inner_merged_index; |
| 1867 |
bool outer_merged; |
| 1868 |
bool inner_merged; |
| 1869 |
|
| 1870 |
Assert(outer_index >= 0 && outer_index < outer_map->nparts); |
| 1871 |
outer_merged_index = outer_map->merged_indexes[outer_index]; |
| 1872 |
outer_merged = outer_map->merged[outer_index]; |
| 1873 |
Assert(inner_index >= 0 && inner_index < inner_map->nparts); |
| 1874 |
inner_merged_index = inner_map->merged_indexes[inner_index]; |
| 1875 |
inner_merged = inner_map->merged[inner_index]; |
| 1876 |
|
| 1877 |
/* |
| 1878 |
* Handle cases where we have already assigned a merged partition to each |
| 1879 |
* of the given partitions. |
| 1880 |
*/ |
| 1881 |
if (outer_merged_index >= 0 && inner_merged_index >= 0) |
| 1882 |
{ |
| 1883 |
/* |
| 1884 |
* If the merged partitions are the same, no need to do anything; |
| 1885 |
* return the index of the merged partitions. Otherwise, if each of |
| 1886 |
* the given partitions has been merged with a dummy partition on the |
| 1887 |
* other side, re-map them to either of the two merged partitions. |
| 1888 |
* Otherwise, they can't be merged, so return -1. |
| 1889 |
*/ |
| 1890 |
if (outer_merged_index == inner_merged_index) |
| 1891 |
{ |
| 1892 |
Assert(outer_merged); |
| 1893 |
Assert(inner_merged); |
| 1894 |
return outer_merged_index; |
| 1895 |
} |
| 1896 |
if (!outer_merged && !inner_merged) |
| 1897 |
{ |
| 1898 |
/* |
| 1899 |
* This can only happen for a list-partitioning case. We re-map |
| 1900 |
* them to the merged partition with the smaller of the two merged |
| 1901 |
* indexes to preserve the property that the canonical order of |
| 1902 |
* list partitions is determined by the indexes assigned to the |
| 1903 |
* smallest list value of each partition. |
| 1904 |
*/ |
| 1905 |
if (outer_merged_index < inner_merged_index) |
| 1906 |
{ |
| 1907 |
outer_map->merged[outer_index] = true; |
| 1908 |
inner_map->merged_indexes[inner_index] = outer_merged_index; |
| 1909 |
inner_map->merged[inner_index] = true; |
| 1910 |
inner_map->did_remapping = true; |
| 1911 |
inner_map->old_indexes[inner_index] = inner_merged_index; |
| 1912 |
return outer_merged_index; |
| 1913 |
} |
| 1914 |
else |
| 1915 |
{ |
| 1916 |
inner_map->merged[inner_index] = true; |
| 1917 |
outer_map->merged_indexes[outer_index] = inner_merged_index; |
| 1918 |
outer_map->merged[outer_index] = true; |
| 1919 |
outer_map->did_remapping = true; |
| 1920 |
outer_map->old_indexes[outer_index] = outer_merged_index; |
| 1921 |
return inner_merged_index; |
| 1922 |
} |
| 1923 |
} |
| 1924 |
return -1; |
| 1925 |
} |
| 1926 |
|
| 1927 |
/* At least one of the given partitions should not have yet been merged. */ |
| 1928 |
Assert(outer_merged_index == -1 || inner_merged_index == -1); |
| 1929 |
|
| 1930 |
/* |
| 1931 |
* If neither of them has been merged, merge them. Otherwise, if one has |
| 1932 |
* been merged with a dummy partition on the other side (and the other |
| 1933 |
* hasn't yet been merged with anything), re-merge them. Otherwise, they |
| 1934 |
* can't be merged, so return -1. |
| 1935 |
*/ |
| 1936 |
if (outer_merged_index == -1 && inner_merged_index == -1) |
| 1937 |
{ |
| 1938 |
int merged_index = *next_index; |
| 1939 |
|
| 1940 |
Assert(!outer_merged); |
| 1941 |
Assert(!inner_merged); |
| 1942 |
outer_map->merged_indexes[outer_index] = merged_index; |
| 1943 |
outer_map->merged[outer_index] = true; |
| 1944 |
inner_map->merged_indexes[inner_index] = merged_index; |
| 1945 |
inner_map->merged[inner_index] = true; |
| 1946 |
*next_index = *next_index + 1; |
| 1947 |
return merged_index; |
| 1948 |
} |
| 1949 |
if (outer_merged_index >= 0 && !outer_map->merged[outer_index]) |
| 1950 |
{ |
| 1951 |
Assert(inner_merged_index == -1); |
| 1952 |
Assert(!inner_merged); |
| 1953 |
inner_map->merged_indexes[inner_index] = outer_merged_index; |
| 1954 |
inner_map->merged[inner_index] = true; |
| 1955 |
outer_map->merged[outer_index] = true; |
| 1956 |
return outer_merged_index; |
| 1957 |
} |
| 1958 |
if (inner_merged_index >= 0 && !inner_map->merged[inner_index]) |
| 1959 |
{ |
| 1960 |
Assert(outer_merged_index == -1); |
| 1961 |
Assert(!outer_merged); |
| 1962 |
outer_map->merged_indexes[outer_index] = inner_merged_index; |
| 1963 |
outer_map->merged[outer_index] = true; |
| 1964 |
inner_map->merged[inner_index] = true; |
| 1965 |
return inner_merged_index; |
| 1966 |
} |
| 1967 |
return -1; |
| 1968 |
} |
| 1969 |
|
| 1970 |
/* |
| 1971 |
* process_outer_partition |
| 1972 |
* Try to assign given outer partition a merged partition, and return the |
| 1973 |
* index of the merged partition if successful, -1 otherwise |
| 1974 |
* |
| 1975 |
* If the partition is newly created, *next_index is incremented. Also, if it |
| 1976 |
* is the default partition of the join relation, *default_index is set to the |
| 1977 |
* index if not already done. |
| 1978 |
*/ |
| 1979 |
static int |
| 1980 |
process_outer_partition(PartitionMap *outer_map, |
| 1981 |
PartitionMap *inner_map, |
| 1982 |
bool outer_has_default, |
| 1983 |
bool inner_has_default, |
| 1984 |
int outer_index, |
| 1985 |
int inner_default, |
| 1986 |
JoinType jointype, |
| 1987 |
int *next_index, |
| 1988 |
int *default_index) |
| 1989 |
{ |
| 1990 |
int merged_index = -1; |
| 1991 |
|
| 1992 |
Assert(outer_index >= 0); |
| 1993 |
|
| 1994 |
/* |
| 1995 |
* If the inner side has the default partition, a row from the outer |
| 1996 |
* partition might find its join partner in the default partition; try |
| 1997 |
* merging the outer partition with the default partition. Otherwise, |
| 1998 |
* this should be an outer join, in which case the outer partition has to |
| 1999 |
* be scanned all the way anyway; merge the outer partition with a dummy |
| 2000 |
* partition on the other side. |
| 2001 |
*/ |
| 2002 |
if (inner_has_default) |
| 2003 |
{ |
| 2004 |
Assert(inner_default >= 0); |
| 2005 |
|
| 2006 |
/* |
| 2007 |
* If the outer side has the default partition as well, the default |
| 2008 |
* partition on the inner side will have two matching partitions on |
| 2009 |
* the other side: the outer partition and the default partition on |
| 2010 |
* the outer side. Partitionwise join doesn't handle this scenario |
| 2011 |
* yet. |
| 2012 |
*/ |
| 2013 |
if (outer_has_default) |
| 2014 |
return -1; |
| 2015 |
|
| 2016 |
merged_index = merge_matching_partitions(outer_map, inner_map, |
| 2017 |
outer_index, inner_default, |
| 2018 |
next_index); |
| 2019 |
if (merged_index == -1) |
| 2020 |
return -1; |
| 2021 |
|
| 2022 |
/* |
| 2023 |
* If this is a FULL join, the default partition on the inner side has |
| 2024 |
* to be scanned all the way anyway, so the resulting partition will |
| 2025 |
* contain all key values from the default partition, which any other |
| 2026 |
* partition of the join relation will not contain. Thus the |
| 2027 |
* resulting partition will act as the default partition of the join |
| 2028 |
* relation; record the index in *default_index if not already done. |
| 2029 |
*/ |
| 2030 |
if (jointype == JOIN_FULL) |
| 2031 |
{ |
| 2032 |
if (*default_index == -1) |
| 2033 |
*default_index = merged_index; |
| 2034 |
else |
| 2035 |
Assert(*default_index == merged_index); |
| 2036 |
} |
| 2037 |
} |
| 2038 |
else |
| 2039 |
{ |
| 2040 |
Assert(IS_OUTER_JOIN(jointype)); |
| 2041 |
Assert(jointype != JOIN_RIGHT); |
| 2042 |
|
| 2043 |
/* If we have already assigned a partition, no need to do anything. */ |
| 2044 |
merged_index = outer_map->merged_indexes[outer_index]; |
| 2045 |
if (merged_index == -1) |
| 2046 |
merged_index = merge_partition_with_dummy(outer_map, outer_index, |
| 2047 |
next_index); |
| 2048 |
} |
| 2049 |
return merged_index; |
| 2050 |
} |
| 2051 |
|
| 2052 |
/* |
| 2053 |
* process_inner_partition |
| 2054 |
* Try to assign given inner partition a merged partition, and return the |
| 2055 |
* index of the merged partition if successful, -1 otherwise |
| 2056 |
* |
| 2057 |
* If the partition is newly created, *next_index is incremented. Also, if it |
| 2058 |
* is the default partition of the join relation, *default_index is set to the |
| 2059 |
* index if not already done. |
| 2060 |
*/ |
| 2061 |
static int |
| 2062 |
process_inner_partition(PartitionMap *outer_map, |
| 2063 |
PartitionMap *inner_map, |
| 2064 |
bool outer_has_default, |
| 2065 |
bool inner_has_default, |
| 2066 |
int inner_index, |
| 2067 |
int outer_default, |
| 2068 |
JoinType jointype, |
| 2069 |
int *next_index, |
| 2070 |
int *default_index) |
| 2071 |
{ |
| 2072 |
int merged_index = -1; |
| 2073 |
|
| 2074 |
Assert(inner_index >= 0); |
| 2075 |
|
| 2076 |
/* |
| 2077 |
* If the outer side has the default partition, a row from the inner |
| 2078 |
* partition might find its join partner in the default partition; try |
| 2079 |
* merging the inner partition with the default partition. Otherwise, |
| 2080 |
* this should be a FULL join, in which case the inner partition has to be |
| 2081 |
* scanned all the way anyway; merge the inner partition with a dummy |
| 2082 |
* partition on the other side. |
| 2083 |
*/ |
| 2084 |
if (outer_has_default) |
| 2085 |
{ |
| 2086 |
Assert(outer_default >= 0); |
| 2087 |
|
| 2088 |
/* |
| 2089 |
* If the inner side has the default partition as well, the default |
| 2090 |
* partition on the outer side will have two matching partitions on |
| 2091 |
* the other side: the inner partition and the default partition on |
| 2092 |
* the inner side. Partitionwise join doesn't handle this scenario |
| 2093 |
* yet. |
| 2094 |
*/ |
| 2095 |
if (inner_has_default) |
| 2096 |
return -1; |
| 2097 |
|
| 2098 |
merged_index = merge_matching_partitions(outer_map, inner_map, |
| 2099 |
outer_default, inner_index, |
| 2100 |
next_index); |
| 2101 |
if (merged_index == -1) |
| 2102 |
return -1; |
| 2103 |
|
| 2104 |
/* |
| 2105 |
* If this is an outer join, the default partition on the outer side |
| 2106 |
* has to be scanned all the way anyway, so the resulting partition |
| 2107 |
* will contain all key values from the default partition, which any |
| 2108 |
* other partition of the join relation will not contain. Thus the |
| 2109 |
* resulting partition will act as the default partition of the join |
| 2110 |
* relation; record the index in *default_index if not already done. |
| 2111 |
*/ |
| 2112 |
if (IS_OUTER_JOIN(jointype)) |
| 2113 |
{ |
| 2114 |
Assert(jointype != JOIN_RIGHT); |
| 2115 |
if (*default_index == -1) |
| 2116 |
*default_index = merged_index; |
| 2117 |
else |
| 2118 |
Assert(*default_index == merged_index); |
| 2119 |
} |
| 2120 |
} |
| 2121 |
else |
| 2122 |
{ |
| 2123 |
Assert(jointype == JOIN_FULL); |
| 2124 |
|
| 2125 |
/* If we have already assigned a partition, no need to do anything. */ |
| 2126 |
merged_index = inner_map->merged_indexes[inner_index]; |
| 2127 |
if (merged_index == -1) |
| 2128 |
merged_index = merge_partition_with_dummy(inner_map, inner_index, |
| 2129 |
next_index); |
| 2130 |
} |
| 2131 |
return merged_index; |
| 2132 |
} |
| 2133 |
|
| 2134 |
/* |
| 2135 |
* merge_null_partitions |
| 2136 |
* Merge the NULL partitions from a join's outer and inner sides. |
| 2137 |
* |
| 2138 |
* If the merged partition produced from them is the NULL partition of the join |
| 2139 |
* relation, *null_index is set to the index of the merged partition. |
| 2140 |
* |
| 2141 |
* Note: We assume here that the join clause for a partitioned join is strict |
| 2142 |
* because have_partkey_equi_join() requires that the corresponding operator |
| 2143 |
* be mergejoinable, and we currently assume that mergejoinable operators are |
| 2144 |
* strict (see MJEvalOuterValues()/MJEvalInnerValues()). |
| 2145 |
*/ |
| 2146 |
static void |
| 2147 |
merge_null_partitions(PartitionMap *outer_map, |
| 2148 |
PartitionMap *inner_map, |
| 2149 |
bool outer_has_null, |
| 2150 |
bool inner_has_null, |
| 2151 |
int outer_null, |
| 2152 |
int inner_null, |
| 2153 |
JoinType jointype, |
| 2154 |
int *next_index, |
| 2155 |
int *null_index) |
| 2156 |
{ |
| 2157 |
bool consider_outer_null = false; |
| 2158 |
bool consider_inner_null = false; |
| 2159 |
|
| 2160 |
Assert(outer_has_null || inner_has_null); |
| 2161 |
Assert(*null_index == -1); |
| 2162 |
|
| 2163 |
/* |
| 2164 |
* Check whether the NULL partitions have already been merged and if so, |
| 2165 |
* set the consider_outer_null/consider_inner_null flags. |
| 2166 |
*/ |
| 2167 |
if (outer_has_null) |
| 2168 |
{ |
| 2169 |
Assert(outer_null >= 0 && outer_null < outer_map->nparts); |
| 2170 |
if (outer_map->merged_indexes[outer_null] == -1) |
| 2171 |
consider_outer_null = true; |
| 2172 |
} |
| 2173 |
if (inner_has_null) |
| 2174 |
{ |
| 2175 |
Assert(inner_null >= 0 && inner_null < inner_map->nparts); |
| 2176 |
if (inner_map->merged_indexes[inner_null] == -1) |
| 2177 |
consider_inner_null = true; |
| 2178 |
} |
| 2179 |
|
| 2180 |
/* If both flags are set false, we don't need to do anything. */ |
| 2181 |
if (!consider_outer_null && !consider_inner_null) |
| 2182 |
return; |
| 2183 |
|
| 2184 |
if (consider_outer_null && !consider_inner_null) |
| 2185 |
{ |
| 2186 |
Assert(outer_has_null); |
| 2187 |
|
| 2188 |
/* |
| 2189 |
* If this is an outer join, the NULL partition on the outer side has |
| 2190 |
* to be scanned all the way anyway; merge the NULL partition with a |
| 2191 |
* dummy partition on the other side. In that case |
| 2192 |
* consider_outer_null means that the NULL partition only contains |
| 2193 |
* NULL values as the key values, so the merged partition will do so; |
| 2194 |
* treat it as the NULL partition of the join relation. |
| 2195 |
*/ |
| 2196 |
if (IS_OUTER_JOIN(jointype)) |
| 2197 |
{ |
| 2198 |
Assert(jointype != JOIN_RIGHT); |
| 2199 |
*null_index = merge_partition_with_dummy(outer_map, outer_null, |
| 2200 |
next_index); |
| 2201 |
} |
| 2202 |
} |
| 2203 |
else if (!consider_outer_null && consider_inner_null) |
| 2204 |
{ |
| 2205 |
Assert(inner_has_null); |
| 2206 |
|
| 2207 |
/* |
| 2208 |
* If this is a FULL join, the NULL partition on the inner side has to |
| 2209 |
* be scanned all the way anyway; merge the NULL partition with a |
| 2210 |
* dummy partition on the other side. In that case |
| 2211 |
* consider_inner_null means that the NULL partition only contains |
| 2212 |
* NULL values as the key values, so the merged partition will do so; |
| 2213 |
* treat it as the NULL partition of the join relation. |
| 2214 |
*/ |
| 2215 |
if (jointype == JOIN_FULL) |
| 2216 |
*null_index = merge_partition_with_dummy(inner_map, inner_null, |
| 2217 |
next_index); |
| 2218 |
} |
| 2219 |
else |
| 2220 |
{ |
| 2221 |
Assert(consider_outer_null && consider_inner_null); |
| 2222 |
Assert(outer_has_null); |
| 2223 |
Assert(inner_has_null); |
| 2224 |
|
| 2225 |
/* |
| 2226 |
* If this is an outer join, the NULL partition on the outer side (and |
| 2227 |
* that on the inner side if this is a FULL join) have to be scanned |
| 2228 |
* all the way anyway, so merge them. Note that each of the NULL |
| 2229 |
* partitions isn't merged yet, so they should be merged successfully. |
| 2230 |
* Like the above, each of the NULL partitions only contains NULL |
| 2231 |
* values as the key values, so the merged partition will do so; treat |
| 2232 |
* it as the NULL partition of the join relation. |
| 2233 |
* |
| 2234 |
* Note: if this an INNER/SEMI join, the join clause will never be |
| 2235 |
* satisfied by two NULL values (see comments above), so both the NULL |
| 2236 |
* partitions can be eliminated. |
| 2237 |
*/ |
| 2238 |
if (IS_OUTER_JOIN(jointype)) |
| 2239 |
{ |
| 2240 |
Assert(jointype != JOIN_RIGHT); |
| 2241 |
*null_index = merge_matching_partitions(outer_map, inner_map, |
| 2242 |
outer_null, inner_null, |
| 2243 |
next_index); |
| 2244 |
Assert(*null_index >= 0); |
| 2245 |
} |
| 2246 |
} |
| 2247 |
} |
| 2248 |
|
| 2249 |
/* |
| 2250 |
* merge_default_partitions |
| 2251 |
* Merge the default partitions from a join's outer and inner sides. |
| 2252 |
* |
| 2253 |
* If the merged partition produced from them is the default partition of the |
| 2254 |
* join relation, *default_index is set to the index of the merged partition. |
| 2255 |
*/ |
| 2256 |
static void |
| 2257 |
merge_default_partitions(PartitionMap *outer_map, |
| 2258 |
PartitionMap *inner_map, |
| 2259 |
bool outer_has_default, |
| 2260 |
bool inner_has_default, |
| 2261 |
int outer_default, |
| 2262 |
int inner_default, |
| 2263 |
JoinType jointype, |
| 2264 |
int *next_index, |
| 2265 |
int *default_index) |
| 2266 |
{ |
| 2267 |
int outer_merged_index = -1; |
| 2268 |
int inner_merged_index = -1; |
| 2269 |
|
| 2270 |
Assert(outer_has_default || inner_has_default); |
| 2271 |
|
| 2272 |
/* Get the merged partition indexes for the default partitions. */ |
| 2273 |
if (outer_has_default) |
| 2274 |
{ |
| 2275 |
Assert(outer_default >= 0 && outer_default < outer_map->nparts); |
| 2276 |
outer_merged_index = outer_map->merged_indexes[outer_default]; |
| 2277 |
} |
| 2278 |
if (inner_has_default) |
| 2279 |
{ |
| 2280 |
Assert(inner_default >= 0 && inner_default < inner_map->nparts); |
| 2281 |
inner_merged_index = inner_map->merged_indexes[inner_default]; |
| 2282 |
} |
| 2283 |
|
| 2284 |
if (outer_has_default && !inner_has_default) |
| 2285 |
{ |
| 2286 |
/* |
| 2287 |
* If this is an outer join, the default partition on the outer side |
| 2288 |
* has to be scanned all the way anyway; if we have not yet assigned a |
| 2289 |
* partition, merge the default partition with a dummy partition on |
| 2290 |
* the other side. The merged partition will act as the default |
| 2291 |
* partition of the join relation (see comments in |
| 2292 |
* process_inner_partition()). |
| 2293 |
*/ |
| 2294 |
if (IS_OUTER_JOIN(jointype)) |
| 2295 |
{ |
| 2296 |
Assert(jointype != JOIN_RIGHT); |
| 2297 |
if (outer_merged_index == -1) |
| 2298 |
{ |
| 2299 |
Assert(*default_index == -1); |
| 2300 |
*default_index = merge_partition_with_dummy(outer_map, |
| 2301 |
outer_default, |
| 2302 |
next_index); |
| 2303 |
} |
| 2304 |
else |
| 2305 |
Assert(*default_index == outer_merged_index); |
| 2306 |
} |
| 2307 |
else |
| 2308 |
Assert(*default_index == -1); |
| 2309 |
} |
| 2310 |
else if (!outer_has_default && inner_has_default) |
| 2311 |
{ |
| 2312 |
/* |
| 2313 |
* If this is a FULL join, the default partition on the inner side has |
| 2314 |
* to be scanned all the way anyway; if we have not yet assigned a |
| 2315 |
* partition, merge the default partition with a dummy partition on |
| 2316 |
* the other side. The merged partition will act as the default |
| 2317 |
* partition of the join relation (see comments in |
| 2318 |
* process_outer_partition()). |
| 2319 |
*/ |
| 2320 |
if (jointype == JOIN_FULL) |
| 2321 |
{ |
| 2322 |
if (inner_merged_index == -1) |
| 2323 |
{ |
| 2324 |
Assert(*default_index == -1); |
| 2325 |
*default_index = merge_partition_with_dummy(inner_map, |
| 2326 |
inner_default, |
| 2327 |
next_index); |
| 2328 |
} |
| 2329 |
else |
| 2330 |
Assert(*default_index == inner_merged_index); |
| 2331 |
} |
| 2332 |
else |
| 2333 |
Assert(*default_index == -1); |
| 2334 |
} |
| 2335 |
else |
| 2336 |
{ |
| 2337 |
Assert(outer_has_default && inner_has_default); |
| 2338 |
|
| 2339 |
/* |
| 2340 |
* The default partitions have to be joined with each other, so merge |
| 2341 |
* them. Note that each of the default partitions isn't merged yet |
| 2342 |
* (see, process_outer_partition()/process_inner_partition()), so they |
| 2343 |
* should be merged successfully. The merged partition will act as |
| 2344 |
* the default partition of the join relation. |
| 2345 |
*/ |
| 2346 |
Assert(outer_merged_index == -1); |
| 2347 |
Assert(inner_merged_index == -1); |
| 2348 |
Assert(*default_index == -1); |
| 2349 |
*default_index = merge_matching_partitions(outer_map, |
| 2350 |
inner_map, |
| 2351 |
outer_default, |
| 2352 |
inner_default, |
| 2353 |
next_index); |
| 2354 |
Assert(*default_index >= 0); |
| 2355 |
} |
| 2356 |
} |
| 2357 |
|
| 2358 |
/* |
| 2359 |
* merge_partition_with_dummy |
| 2360 |
* Assign given partition a new partition of a join relation |
| 2361 |
* |
| 2362 |
* Note: The caller assumes that the given partition doesn't have a non-dummy |
| 2363 |
* matching partition on the other side, but if the given partition finds the |
| 2364 |
* matching partition later, we will adjust the assignment. |
| 2365 |
*/ |
| 2366 |
static int |
| 2367 |
merge_partition_with_dummy(PartitionMap *map, int index, int *next_index) |
| 2368 |
{ |
| 2369 |
int merged_index = *next_index; |
| 2370 |
|
| 2371 |
Assert(index >= 0 && index < map->nparts); |
| 2372 |
Assert(map->merged_indexes[index] == -1); |
| 2373 |
Assert(!map->merged[index]); |
| 2374 |
map->merged_indexes[index] = merged_index; |
| 2375 |
/* Leave the merged flag alone! */ |
| 2376 |
*next_index = *next_index + 1; |
| 2377 |
return merged_index; |
| 2378 |
} |
| 2379 |
|
| 2380 |
/* |
| 2381 |
* fix_merged_indexes |
| 2382 |
* Adjust merged indexes of re-merged partitions |
| 2383 |
*/ |
| 2384 |
static void |
| 2385 |
fix_merged_indexes(PartitionMap *outer_map, PartitionMap *inner_map, |
| 2386 |
int nmerged, List *merged_indexes) |
| 2387 |
{ |
| 2388 |
int *new_indexes; |
| 2389 |
int merged_index; |
| 2390 |
int i; |
| 2391 |
ListCell *lc; |
| 2392 |
|
| 2393 |
Assert(nmerged > 0); |
| 2394 |
|
| 2395 |
new_indexes = (int *) palloc(sizeof(int) * nmerged); |
| 2396 |
for (i = 0; i < nmerged; i++) |
| 2397 |
new_indexes[i] = -1; |
| 2398 |
|
| 2399 |
/* Build the mapping of old merged indexes to new merged indexes. */ |
| 2400 |
if (outer_map->did_remapping) |
| 2401 |
{ |
| 2402 |
for (i = 0; i < outer_map->nparts; i++) |
| 2403 |
{ |
| 2404 |
merged_index = outer_map->old_indexes[i]; |
| 2405 |
if (merged_index >= 0) |
| 2406 |
new_indexes[merged_index] = outer_map->merged_indexes[i]; |
| 2407 |
} |
| 2408 |
} |
| 2409 |
if (inner_map->did_remapping) |
| 2410 |
{ |
| 2411 |
for (i = 0; i < inner_map->nparts; i++) |
| 2412 |
{ |
| 2413 |
merged_index = inner_map->old_indexes[i]; |
| 2414 |
if (merged_index >= 0) |
| 2415 |
new_indexes[merged_index] = inner_map->merged_indexes[i]; |
| 2416 |
} |
| 2417 |
} |
| 2418 |
|
| 2419 |
/* Fix the merged_indexes list using the mapping. */ |
| 2420 |
foreach(lc, merged_indexes) |
| 2421 |
{ |
| 2422 |
merged_index = lfirst_int(lc); |
| 2423 |
Assert(merged_index >= 0); |
| 2424 |
if (new_indexes[merged_index] >= 0) |
| 2425 |
lfirst_int(lc) = new_indexes[merged_index]; |
| 2426 |
} |
| 2427 |
|
| 2428 |
pfree(new_indexes); |
| 2429 |
} |
| 2430 |
|
| 2431 |
/* |
| 2432 |
* generate_matching_part_pairs |
| 2433 |
* Generate a pair of lists of partitions that produce merged partitions |
| 2434 |
* |
| 2435 |
* The lists of partitions are built in the order of merged partition indexes, |
| 2436 |
* and returned in *outer_parts and *inner_parts. |
| 2437 |
*/ |
| 2438 |
static void |
| 2439 |
generate_matching_part_pairs(RelOptInfo *outer_rel, RelOptInfo *inner_rel, |
| 2440 |
PartitionMap *outer_map, PartitionMap *inner_map, |
| 2441 |
int nmerged, |
| 2442 |
List **outer_parts, List **inner_parts) |
| 2443 |
{ |
| 2444 |
int outer_nparts = outer_map->nparts; |
| 2445 |
int inner_nparts = inner_map->nparts; |
| 2446 |
int *outer_indexes; |
| 2447 |
int *inner_indexes; |
| 2448 |
int max_nparts; |
| 2449 |
int i; |
| 2450 |
|
| 2451 |
Assert(nmerged > 0); |
| 2452 |
Assert(*outer_parts == NIL); |
| 2453 |
Assert(*inner_parts == NIL); |
| 2454 |
|
| 2455 |
outer_indexes = (int *) palloc(sizeof(int) * nmerged); |
| 2456 |
inner_indexes = (int *) palloc(sizeof(int) * nmerged); |
| 2457 |
for (i = 0; i < nmerged; i++) |
| 2458 |
outer_indexes[i] = inner_indexes[i] = -1; |
| 2459 |
|
| 2460 |
/* Set pairs of matching partitions. */ |
| 2461 |
Assert(outer_nparts == outer_rel->nparts); |
| 2462 |
Assert(inner_nparts == inner_rel->nparts); |
| 2463 |
max_nparts = Max(outer_nparts, inner_nparts); |
| 2464 |
for (i = 0; i < max_nparts; i++) |
| 2465 |
{ |
| 2466 |
if (i < outer_nparts) |
| 2467 |
{ |
| 2468 |
int merged_index = outer_map->merged_indexes[i]; |
| 2469 |
|
| 2470 |
if (merged_index >= 0) |
| 2471 |
{ |
| 2472 |
Assert(merged_index < nmerged); |
| 2473 |
outer_indexes[merged_index] = i; |
| 2474 |
} |
| 2475 |
} |
| 2476 |
if (i < inner_nparts) |
| 2477 |
{ |
| 2478 |
int merged_index = inner_map->merged_indexes[i]; |
| 2479 |
|
| 2480 |
if (merged_index >= 0) |
| 2481 |
{ |
| 2482 |
Assert(merged_index < nmerged); |
| 2483 |
inner_indexes[merged_index] = i; |
| 2484 |
} |
| 2485 |
} |
| 2486 |
} |
| 2487 |
|
| 2488 |
/* Build the list pairs. */ |
| 2489 |
for (i = 0; i < nmerged; i++) |
| 2490 |
{ |
| 2491 |
int outer_index = outer_indexes[i]; |
| 2492 |
int inner_index = inner_indexes[i]; |
| 2493 |
|
| 2494 |
/* |
| 2495 |
* If both partitions are dummy, it means the merged partition that |
| 2496 |
* had been assigned to the outer/inner partition was removed when |
| 2497 |
* re-merging the outer/inner partition in |
| 2498 |
* merge_matching_partitions(); ignore the merged partition. |
| 2499 |
*/ |
| 2500 |
if (outer_index == -1 && inner_index == -1) |
| 2501 |
continue; |
| 2502 |
|
| 2503 |
*outer_parts = lappend(*outer_parts, outer_index >= 0 ? |
| 2504 |
outer_rel->part_rels[outer_index] : NULL); |
| 2505 |
*inner_parts = lappend(*inner_parts, inner_index >= 0 ? |
| 2506 |
inner_rel->part_rels[inner_index] : NULL); |
| 2507 |
} |
| 2508 |
|
| 2509 |
pfree(outer_indexes); |
| 2510 |
pfree(inner_indexes); |
| 2511 |
} |
| 2512 |
|
| 2513 |
/* |
| 2514 |
* build_merged_partition_bounds |
| 2515 |
* Create a PartitionBoundInfo struct from merged partition bounds |
| 2516 |
*/ |
| 2517 |
static PartitionBoundInfo |
| 2518 |
build_merged_partition_bounds(char strategy, List *merged_datums, |
| 2519 |
List *merged_kinds, List *merged_indexes, |
| 2520 |
int null_index, int default_index) |
| 2521 |
{ |
| 2522 |
PartitionBoundInfo merged_bounds; |
| 2523 |
int ndatums = list_length(merged_datums); |
| 2524 |
int pos; |
| 2525 |
ListCell *lc; |
| 2526 |
|
| 2527 |
merged_bounds = (PartitionBoundInfo) palloc(sizeof(PartitionBoundInfoData)); |
| 2528 |
merged_bounds->strategy = strategy; |
| 2529 |
merged_bounds->ndatums = ndatums; |
| 2530 |
|
| 2531 |
merged_bounds->datums = (Datum **) palloc(sizeof(Datum *) * ndatums); |
| 2532 |
pos = 0; |
| 2533 |
foreach(lc, merged_datums) |
| 2534 |
merged_bounds->datums[pos++] = (Datum *) lfirst(lc); |
| 2535 |
|
| 2536 |
if (strategy == PARTITION_STRATEGY_RANGE) |
| 2537 |
{ |
| 2538 |
Assert(list_length(merged_kinds) == ndatums); |
| 2539 |
merged_bounds->kind = (PartitionRangeDatumKind **) |
| 2540 |
palloc(sizeof(PartitionRangeDatumKind *) * ndatums); |
| 2541 |
pos = 0; |
| 2542 |
foreach(lc, merged_kinds) |
| 2543 |
merged_bounds->kind[pos++] = (PartitionRangeDatumKind *) lfirst(lc); |
| 2544 |
|
| 2545 |
/* There are ndatums+1 indexes in the case of range partitioning. */ |
| 2546 |
merged_indexes = lappend_int(merged_indexes, -1); |
| 2547 |
ndatums++; |
| 2548 |
} |
| 2549 |
else |
| 2550 |
{ |
| 2551 |
Assert(strategy == PARTITION_STRATEGY_LIST); |
| 2552 |
Assert(merged_kinds == NIL); |
| 2553 |
merged_bounds->kind = NULL; |
| 2554 |
} |
| 2555 |
|
| 2556 |
/* interleaved_parts is always NULL for join relations. */ |
| 2557 |
merged_bounds->interleaved_parts = NULL; |
| 2558 |
|
| 2559 |
Assert(list_length(merged_indexes) == ndatums); |
| 2560 |
merged_bounds->nindexes = ndatums; |
| 2561 |
merged_bounds->indexes = (int *) palloc(sizeof(int) * ndatums); |
| 2562 |
pos = 0; |
| 2563 |
foreach(lc, merged_indexes) |
| 2564 |
merged_bounds->indexes[pos++] = lfirst_int(lc); |
| 2565 |
|
| 2566 |
merged_bounds->null_index = null_index; |
| 2567 |
merged_bounds->default_index = default_index; |
| 2568 |
|
| 2569 |
return merged_bounds; |
| 2570 |
} |
| 2571 |
|
| 2572 |
/* |
| 2573 |
* get_range_partition |
| 2574 |
* Get the next non-dummy partition of a range-partitioned relation, |
| 2575 |
* returning the index of that partition |
| 2576 |
* |
| 2577 |
* *lb and *ub are set to the lower and upper bounds of that partition |
| 2578 |
* respectively, and *lb_pos is advanced to the next lower bound, if any. |
| 2579 |
*/ |
| 2580 |
static int |
| 2581 |
get_range_partition(RelOptInfo *rel, |
| 2582 |
PartitionBoundInfo bi, |
| 2583 |
int *lb_pos, |
| 2584 |
PartitionRangeBound *lb, |
| 2585 |
PartitionRangeBound *ub) |
| 2586 |
{ |
| 2587 |
int part_index; |
| 2588 |
|
| 2589 |
Assert(bi->strategy == PARTITION_STRATEGY_RANGE); |
| 2590 |
|
| 2591 |
do |
| 2592 |
{ |
| 2593 |
part_index = get_range_partition_internal(bi, lb_pos, lb, ub); |
| 2594 |
if (part_index == -1) |
| 2595 |
return -1; |
| 2596 |
} while (is_dummy_partition(rel, part_index)); |
| 2597 |
|
| 2598 |
return part_index; |
| 2599 |
} |
| 2600 |
|
| 2601 |
static int |
| 2602 |
get_range_partition_internal(PartitionBoundInfo bi, |
| 2603 |
int *lb_pos, |
| 2604 |
PartitionRangeBound *lb, |
| 2605 |
PartitionRangeBound *ub) |
| 2606 |
{ |
| 2607 |
/* Return the index as -1 if we've exhausted all lower bounds. */ |
| 2608 |
if (*lb_pos >= bi->ndatums) |
| 2609 |
return -1; |
| 2610 |
|
| 2611 |
/* A lower bound should have at least one more bound after it. */ |
| 2612 |
Assert(*lb_pos + 1 < bi->ndatums); |
| 2613 |
|
| 2614 |
/* Set the lower bound. */ |
| 2615 |
lb->index = bi->indexes[*lb_pos]; |
| 2616 |
lb->datums = bi->datums[*lb_pos]; |
| 2617 |
lb->kind = bi->kind[*lb_pos]; |
| 2618 |
lb->lower = true; |
| 2619 |
/* Set the upper bound. */ |
| 2620 |
ub->index = bi->indexes[*lb_pos + 1]; |
| 2621 |
ub->datums = bi->datums[*lb_pos + 1]; |
| 2622 |
ub->kind = bi->kind[*lb_pos + 1]; |
| 2623 |
ub->lower = false; |
| 2624 |
|
| 2625 |
/* The index assigned to an upper bound should be valid. */ |
| 2626 |
Assert(ub->index >= 0); |
| 2627 |
|
| 2628 |
/* |
| 2629 |
* Advance the position to the next lower bound. If there are no bounds |
| 2630 |
* left beyond the upper bound, we have reached the last lower bound. |
| 2631 |
*/ |
| 2632 |
if (*lb_pos + 2 >= bi->ndatums) |
| 2633 |
*lb_pos = bi->ndatums; |
| 2634 |
else |
| 2635 |
{ |
| 2636 |
/* |
| 2637 |
* If the index assigned to the bound next to the upper bound isn't |
| 2638 |
* valid, that is the next lower bound; else, the upper bound is also |
| 2639 |
* the lower bound of the next range partition. |
| 2640 |
*/ |
| 2641 |
if (bi->indexes[*lb_pos + 2] < 0) |
| 2642 |
*lb_pos = *lb_pos + 2; |
| 2643 |
else |
| 2644 |
*lb_pos = *lb_pos + 1; |
| 2645 |
} |
| 2646 |
|
| 2647 |
return ub->index; |
| 2648 |
} |
| 2649 |
|
| 2650 |
/* |
| 2651 |
* compare_range_partitions |
| 2652 |
* Compare the bounds of two range partitions, and return true if the |
| 2653 |
* two partitions overlap, false otherwise |
| 2654 |
* |
| 2655 |
* *lb_cmpval is set to -1, 0, or 1 if the outer partition's lower bound is |
| 2656 |
* lower than, equal to, or higher than the inner partition's lower bound |
| 2657 |
* respectively. Likewise, *ub_cmpval is set to -1, 0, or 1 if the outer |
| 2658 |
* partition's upper bound is lower than, equal to, or higher than the inner |
| 2659 |
* partition's upper bound respectively. |
| 2660 |
*/ |
| 2661 |
static bool |
| 2662 |
compare_range_partitions(int partnatts, FmgrInfo *partsupfuncs, |
| 2663 |
Oid *partcollations, |
| 2664 |
PartitionRangeBound *outer_lb, |
| 2665 |
PartitionRangeBound *outer_ub, |
| 2666 |
PartitionRangeBound *inner_lb, |
| 2667 |
PartitionRangeBound *inner_ub, |
| 2668 |
int *lb_cmpval, int *ub_cmpval) |
| 2669 |
{ |
| 2670 |
/* |
| 2671 |
* Check if the outer partition's upper bound is lower than the inner |
| 2672 |
* partition's lower bound; if so the partitions aren't overlapping. |
| 2673 |
*/ |
| 2674 |
if (compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 2675 |
outer_ub, inner_lb) < 0) |
| 2676 |
{ |
| 2677 |
*lb_cmpval = -1; |
| 2678 |
*ub_cmpval = -1; |
| 2679 |
return false; |
| 2680 |
} |
| 2681 |
|
| 2682 |
/* |
| 2683 |
* Check if the outer partition's lower bound is higher than the inner |
| 2684 |
* partition's upper bound; if so the partitions aren't overlapping. |
| 2685 |
*/ |
| 2686 |
if (compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 2687 |
outer_lb, inner_ub) > 0) |
| 2688 |
{ |
| 2689 |
*lb_cmpval = 1; |
| 2690 |
*ub_cmpval = 1; |
| 2691 |
return false; |
| 2692 |
} |
| 2693 |
|
| 2694 |
/* All other cases indicate overlapping partitions. */ |
| 2695 |
*lb_cmpval = compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 2696 |
outer_lb, inner_lb); |
| 2697 |
*ub_cmpval = compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 2698 |
outer_ub, inner_ub); |
| 2699 |
return true; |
| 2700 |
} |
| 2701 |
|
| 2702 |
/* |
| 2703 |
* get_merged_range_bounds |
| 2704 |
* Given the bounds of range partitions to be joined, determine the bounds |
| 2705 |
* of a merged partition produced from the range partitions |
| 2706 |
* |
| 2707 |
* *merged_lb and *merged_ub are set to the lower and upper bounds of the |
| 2708 |
* merged partition. |
| 2709 |
*/ |
| 2710 |
static void |
| 2711 |
get_merged_range_bounds(int partnatts, FmgrInfo *partsupfuncs, |
| 2712 |
Oid *partcollations, JoinType jointype, |
| 2713 |
PartitionRangeBound *outer_lb, |
| 2714 |
PartitionRangeBound *outer_ub, |
| 2715 |
PartitionRangeBound *inner_lb, |
| 2716 |
PartitionRangeBound *inner_ub, |
| 2717 |
int lb_cmpval, int ub_cmpval, |
| 2718 |
PartitionRangeBound *merged_lb, |
| 2719 |
PartitionRangeBound *merged_ub) |
| 2720 |
{ |
| 2721 |
Assert(compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 2722 |
outer_lb, inner_lb) == lb_cmpval); |
| 2723 |
Assert(compare_range_bounds(partnatts, partsupfuncs, partcollations, |
| 2724 |
outer_ub, inner_ub) == ub_cmpval); |
| 2725 |
|
| 2726 |
switch (jointype) |
| 2727 |
{ |
| 2728 |
case JOIN_INNER: |
| 2729 |
case JOIN_SEMI: |
| 2730 |
|
| 2731 |
/* |
| 2732 |
* An INNER/SEMI join will have the rows that fit both sides, so |
| 2733 |
* the lower bound of the merged partition will be the higher of |
| 2734 |
* the two lower bounds, and the upper bound of the merged |
| 2735 |
* partition will be the lower of the two upper bounds. |
| 2736 |
*/ |
| 2737 |
*merged_lb = (lb_cmpval > 0) ? *outer_lb : *inner_lb; |
| 2738 |
*merged_ub = (ub_cmpval < 0) ? *outer_ub : *inner_ub; |
| 2739 |
break; |
| 2740 |
|
| 2741 |
case JOIN_LEFT: |
| 2742 |
case JOIN_ANTI: |
| 2743 |
|
| 2744 |
/* |
| 2745 |
* A LEFT/ANTI join will have all the rows from the outer side, so |
| 2746 |
* the bounds of the merged partition will be the same as the |
| 2747 |
* outer bounds. |
| 2748 |
*/ |
| 2749 |
*merged_lb = *outer_lb; |
| 2750 |
*merged_ub = *outer_ub; |
| 2751 |
break; |
| 2752 |
|
| 2753 |
case JOIN_FULL: |
| 2754 |
|
| 2755 |
/* |
| 2756 |
* A FULL join will have all the rows from both sides, so the |
| 2757 |
* lower bound of the merged partition will be the lower of the |
| 2758 |
* two lower bounds, and the upper bound of the merged partition |
| 2759 |
* will be the higher of the two upper bounds. |
| 2760 |
*/ |
| 2761 |
*merged_lb = (lb_cmpval < 0) ? *outer_lb : *inner_lb; |
| 2762 |
*merged_ub = (ub_cmpval > 0) ? *outer_ub : *inner_ub; |
| 2763 |
break; |
| 2764 |
|
| 2765 |
default: |
| 2766 |
elog(ERROR, "unrecognized join type: %d", (int) jointype); |
| 2767 |
} |
| 2768 |
} |
| 2769 |
|
| 2770 |
/* |
| 2771 |
* add_merged_range_bounds |
| 2772 |
* Add the bounds of a merged partition to the lists of range bounds |
| 2773 |
*/ |
| 2774 |
static void |
| 2775 |
add_merged_range_bounds(int partnatts, FmgrInfo *partsupfuncs, |
| 2776 |
Oid *partcollations, |
| 2777 |
PartitionRangeBound *merged_lb, |
| 2778 |
PartitionRangeBound *merged_ub, |
| 2779 |
int merged_index, |
| 2780 |
List **merged_datums, |
| 2781 |
List **merged_kinds, |
| 2782 |
List **merged_indexes) |
| 2783 |
{ |
| 2784 |
int cmpval; |
| 2785 |
|
| 2786 |
if (!*merged_datums) |
| 2787 |
{ |
| 2788 |
/* First merged partition */ |
| 2789 |
Assert(!*merged_kinds); |
| 2790 |
Assert(!*merged_indexes); |
| 2791 |
cmpval = 1; |
| 2792 |
} |
| 2793 |
else |
| 2794 |
{ |
| 2795 |
PartitionRangeBound prev_ub; |
| 2796 |
|
| 2797 |
Assert(*merged_datums); |
| 2798 |
Assert(*merged_kinds); |
| 2799 |
Assert(*merged_indexes); |
| 2800 |
|
| 2801 |
/* Get the last upper bound. */ |
| 2802 |
prev_ub.index = llast_int(*merged_indexes); |
| 2803 |
prev_ub.datums = (Datum *) llast(*merged_datums); |
| 2804 |
prev_ub.kind = (PartitionRangeDatumKind *) llast(*merged_kinds); |
| 2805 |
prev_ub.lower = false; |
| 2806 |
|
| 2807 |
/* |
| 2808 |
* We pass lower1 = false to partition_rbound_cmp() to prevent it from |
| 2809 |
* considering the last upper bound to be smaller than the lower bound |
| 2810 |
* of the merged partition when the values of the two range bounds |
| 2811 |
* compare equal. |
| 2812 |
*/ |
| 2813 |
cmpval = partition_rbound_cmp(partnatts, partsupfuncs, partcollations, |
| 2814 |
merged_lb->datums, merged_lb->kind, |
| 2815 |
false, &prev_ub); |
| 2816 |
Assert(cmpval >= 0); |
| 2817 |
} |
| 2818 |
|
| 2819 |
/* |
| 2820 |
* If the lower bound is higher than the last upper bound, add the lower |
| 2821 |
* bound with the index as -1 indicating that that is a lower bound; else, |
| 2822 |
* the last upper bound will be reused as the lower bound of the merged |
| 2823 |
* partition, so skip this. |
| 2824 |
*/ |
| 2825 |
if (cmpval > 0) |
| 2826 |
{ |
| 2827 |
*merged_datums = lappend(*merged_datums, merged_lb->datums); |
| 2828 |
*merged_kinds = lappend(*merged_kinds, merged_lb->kind); |
| 2829 |
*merged_indexes = lappend_int(*merged_indexes, -1); |
| 2830 |
} |
| 2831 |
|
| 2832 |
/* Add the upper bound and index of the merged partition. */ |
| 2833 |
*merged_datums = lappend(*merged_datums, merged_ub->datums); |
| 2834 |
*merged_kinds = lappend(*merged_kinds, merged_ub->kind); |
| 2835 |
*merged_indexes = lappend_int(*merged_indexes, merged_index); |
| 2836 |
} |
| 2837 |
|
| 2838 |
/* |
| 2839 |
* partitions_are_ordered |
| 2840 |
* Determine whether the partitions described by 'boundinfo' are ordered, |
| 2841 |
* that is partitions appearing earlier in the PartitionDesc sequence |
| 2842 |
* contain partition keys strictly less than those appearing later. |
| 2843 |
* Also, if NULL values are possible, they must come in the last |
| 2844 |
* partition defined in the PartitionDesc. 'live_parts' marks which |
| 2845 |
* partitions we should include when checking the ordering. Partitions |
| 2846 |
* that do not appear in 'live_parts' are ignored. |
| 2847 |
* |
| 2848 |
* If out of order, or there is insufficient info to know the order, |
| 2849 |
* then we return false. |
| 2850 |
*/ |
| 2851 |
bool |
| 2852 |
partitions_are_ordered(PartitionBoundInfo boundinfo, Bitmapset *live_parts) |
| 2853 |
{ |
| 2854 |
Assert(boundinfo != NULL); |
| 2855 |
|
| 2856 |
switch (boundinfo->strategy) |
| 2857 |
{ |
| 2858 |
case PARTITION_STRATEGY_RANGE: |
| 2859 |
|
| 2860 |
/* |
| 2861 |
* RANGE-type partitioning guarantees that the partitions can be |
| 2862 |
* scanned in the order that they're defined in the PartitionDesc |
| 2863 |
* to provide sequential, non-overlapping ranges of tuples. |
| 2864 |
* However, if a DEFAULT partition exists and it's contained |
| 2865 |
* within live_parts, then the partitions are not ordered. |
| 2866 |
*/ |
| 2867 |
if (!partition_bound_has_default(boundinfo) || |
| 2868 |
!bms_is_member(boundinfo->default_index, live_parts)) |
| 2869 |
return true; |
| 2870 |
break; |
| 2871 |
|
| 2872 |
case PARTITION_STRATEGY_LIST: |
| 2873 |
|
| 2874 |
/* |
| 2875 |
* LIST partitioned are ordered providing none of live_parts |
| 2876 |
* overlap with the partitioned table's interleaved partitions. |
| 2877 |
*/ |
| 2878 |
if (!bms_overlap(live_parts, boundinfo->interleaved_parts)) |
| 2879 |
return true; |
| 2880 |
|
| 2881 |
break; |
| 2882 |
case PARTITION_STRATEGY_HASH: |
| 2883 |
break; |
| 2884 |
} |
| 2885 |
|
| 2886 |
return false; |
| 2887 |
} |
| 2888 |
|
| 2889 |
/* |
| 2890 |
* check_new_partition_bound |
| 2891 |
* |
| 2892 |
* Checks if the new partition's bound overlaps any of the existing partitions |
| 2893 |
* of parent. Also performs additional checks as necessary per strategy. |
| 2894 |
*/ |
| 2895 |
void |
| 2896 |
check_new_partition_bound(char *relname, Relation parent, |
| 2897 |
PartitionBoundSpec *spec, ParseState *pstate) |
| 2898 |
{ |
| 2899 |
PartitionKey key = RelationGetPartitionKey(parent); |
| 2900 |
PartitionDesc partdesc = RelationGetPartitionDesc(parent, false); |
| 2901 |
PartitionBoundInfo boundinfo = partdesc->boundinfo; |
| 2902 |
int with = -1; |
| 2903 |
bool overlap = false; |
| 2904 |
int overlap_location = -1; |
| 2905 |
|
| 2906 |
if (spec->is_default) |
| 2907 |
{ |
| 2908 |
/* |
| 2909 |
* The default partition bound never conflicts with any other |
| 2910 |
* partition's; if that's what we're attaching, the only possible |
| 2911 |
* problem is that one already exists, so check for that and we're |
| 2912 |
* done. |
| 2913 |
*/ |
| 2914 |
if (boundinfo == NULL || !partition_bound_has_default(boundinfo)) |
| 2915 |
return; |
| 2916 |
|
| 2917 |
/* Default partition already exists, error out. */ |
| 2918 |
ereport(ERROR, |
| 2919 |
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 2920 |
errmsg("partition \"%s\" conflicts with existing default partition \"%s\"", |
| 2921 |
relname, get_rel_name(partdesc->oids[boundinfo->default_index])), |
| 2922 |
parser_errposition(pstate, spec->location))); |
| 2923 |
} |
| 2924 |
|
| 2925 |
switch (key->strategy) |
| 2926 |
{ |
| 2927 |
case PARTITION_STRATEGY_HASH: |
| 2928 |
{ |
| 2929 |
Assert(spec->strategy == PARTITION_STRATEGY_HASH); |
| 2930 |
Assert(spec->remainder >= 0 && spec->remainder < spec->modulus); |
| 2931 |
|
| 2932 |
if (partdesc->nparts > 0) |
| 2933 |
{ |
| 2934 |
int greatest_modulus; |
| 2935 |
int remainder; |
| 2936 |
int offset; |
| 2937 |
|
| 2938 |
/* |
| 2939 |
* Check rule that every modulus must be a factor of the |
| 2940 |
* next larger modulus. (For example, if you have a bunch |
| 2941 |
* of partitions that all have modulus 5, you can add a |
| 2942 |
* new partition with modulus 10 or a new partition with |
| 2943 |
* modulus 15, but you cannot add both a partition with |
| 2944 |
* modulus 10 and a partition with modulus 15, because 10 |
| 2945 |
* is not a factor of 15.) We need only check the next |
| 2946 |
* smaller and next larger existing moduli, relying on |
| 2947 |
* previous enforcement of this rule to be sure that the |
| 2948 |
* rest are in line. |
| 2949 |
*/ |
| 2950 |
|
| 2951 |
/* |
| 2952 |
* Get the greatest (modulus, remainder) pair contained in |
| 2953 |
* boundinfo->datums that is less than or equal to the |
| 2954 |
* (spec->modulus, spec->remainder) pair. |
| 2955 |
*/ |
| 2956 |
offset = partition_hash_bsearch(boundinfo, |
| 2957 |
spec->modulus, |
| 2958 |
spec->remainder); |
| 2959 |
if (offset < 0) |
| 2960 |
{ |
| 2961 |
int next_modulus; |
| 2962 |
|
| 2963 |
/* |
| 2964 |
* All existing moduli are greater or equal, so the |
| 2965 |
* new one must be a factor of the smallest one, which |
| 2966 |
* is first in the boundinfo. |
| 2967 |
*/ |
| 2968 |
next_modulus = DatumGetInt32(boundinfo->datums[0][0]); |
| 2969 |
if (next_modulus % spec->modulus != 0) |
| 2970 |
ereport(ERROR, |
| 2971 |
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 2972 |
errmsg("every hash partition modulus must be a factor of the next larger modulus"), |
| 2973 |
errdetail("The new modulus %d is not a factor of %d, the modulus of existing partition \"%s\".", |
| 2974 |
spec->modulus, next_modulus, |
| 2975 |
get_rel_name(partdesc->oids[0])))); |
| 2976 |
} |
| 2977 |
else |
| 2978 |
{ |
| 2979 |
int prev_modulus; |
| 2980 |
|
| 2981 |
/* |
| 2982 |
* We found the largest (modulus, remainder) pair less |
| 2983 |
* than or equal to the new one. That modulus must be |
| 2984 |
* a divisor of, or equal to, the new modulus. |
| 2985 |
*/ |
| 2986 |
prev_modulus = DatumGetInt32(boundinfo->datums[offset][0]); |
| 2987 |
|
| 2988 |
if (spec->modulus % prev_modulus != 0) |
| 2989 |
ereport(ERROR, |
| 2990 |
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 2991 |
errmsg("every hash partition modulus must be a factor of the next larger modulus"), |
| 2992 |
errdetail("The new modulus %d is not divisible by %d, the modulus of existing partition \"%s\".", |
| 2993 |
spec->modulus, |
| 2994 |
prev_modulus, |
| 2995 |
get_rel_name(partdesc->oids[offset])))); |
| 2996 |
|
| 2997 |
if (offset + 1 < boundinfo->ndatums) |
| 2998 |
{ |
| 2999 |
int next_modulus; |
| 3000 |
|
| 3001 |
/* |
| 3002 |
* Look at the next higher (modulus, remainder) |
| 3003 |
* pair. That could have the same modulus and a |
| 3004 |
* larger remainder than the new pair, in which |
| 3005 |
* case we're good. If it has a larger modulus, |
| 3006 |
* the new modulus must divide that one. |
| 3007 |
*/ |
| 3008 |
next_modulus = DatumGetInt32(boundinfo->datums[offset + 1][0]); |
| 3009 |
|
| 3010 |
if (next_modulus % spec->modulus != 0) |
| 3011 |
ereport(ERROR, |
| 3012 |
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 3013 |
errmsg("every hash partition modulus must be a factor of the next larger modulus"), |
| 3014 |
errdetail("The new modulus %d is not a factor of %d, the modulus of existing partition \"%s\".", |
| 3015 |
spec->modulus, next_modulus, |
| 3016 |
get_rel_name(partdesc->oids[offset + 1])))); |
| 3017 |
} |
| 3018 |
} |
| 3019 |
|
| 3020 |
greatest_modulus = boundinfo->nindexes; |
| 3021 |
remainder = spec->remainder; |
| 3022 |
|
| 3023 |
/* |
| 3024 |
* Normally, the lowest remainder that could conflict with |
| 3025 |
* the new partition is equal to the remainder specified |
| 3026 |
* for the new partition, but when the new partition has a |
| 3027 |
* modulus higher than any used so far, we need to adjust. |
| 3028 |
*/ |
| 3029 |
if (remainder >= greatest_modulus) |
| 3030 |
remainder = remainder % greatest_modulus; |
| 3031 |
|
| 3032 |
/* Check every potentially-conflicting remainder. */ |
| 3033 |
do |
| 3034 |
{ |
| 3035 |
if (boundinfo->indexes[remainder] != -1) |
| 3036 |
{ |
| 3037 |
overlap = true; |
| 3038 |
overlap_location = spec->location; |
| 3039 |
with = boundinfo->indexes[remainder]; |
| 3040 |
break; |
| 3041 |
} |
| 3042 |
remainder += spec->modulus; |
| 3043 |
} while (remainder < greatest_modulus); |
| 3044 |
} |
| 3045 |
|
| 3046 |
break; |
| 3047 |
} |
| 3048 |
|
| 3049 |
case PARTITION_STRATEGY_LIST: |
| 3050 |
{ |
| 3051 |
Assert(spec->strategy == PARTITION_STRATEGY_LIST); |
| 3052 |
|
| 3053 |
if (partdesc->nparts > 0) |
| 3054 |
{ |
| 3055 |
ListCell *cell; |
| 3056 |
|
| 3057 |
Assert(boundinfo && |
| 3058 |
boundinfo->strategy == PARTITION_STRATEGY_LIST && |
| 3059 |
(boundinfo->ndatums > 0 || |
| 3060 |
partition_bound_accepts_nulls(boundinfo) || |
| 3061 |
partition_bound_has_default(boundinfo))); |
| 3062 |
|
| 3063 |
foreach(cell, spec->listdatums) |
| 3064 |
{ |
| 3065 |
Const *val = lfirst_node(Const, cell); |
| 3066 |
|
| 3067 |
overlap_location = val->location; |
| 3068 |
if (!val->constisnull) |
| 3069 |
{ |
| 3070 |
int offset; |
| 3071 |
bool equal; |
| 3072 |
|
| 3073 |
offset = partition_list_bsearch(&key->partsupfunc[0], |
| 3074 |
key->partcollation, |
| 3075 |
boundinfo, |
| 3076 |
val->constvalue, |
| 3077 |
&equal); |
| 3078 |
if (offset >= 0 && equal) |
| 3079 |
{ |
| 3080 |
overlap = true; |
| 3081 |
with = boundinfo->indexes[offset]; |
| 3082 |
break; |
| 3083 |
} |
| 3084 |
} |
| 3085 |
else if (partition_bound_accepts_nulls(boundinfo)) |
| 3086 |
{ |
| 3087 |
overlap = true; |
| 3088 |
with = boundinfo->null_index; |
| 3089 |
break; |
| 3090 |
} |
| 3091 |
} |
| 3092 |
} |
| 3093 |
|
| 3094 |
break; |
| 3095 |
} |
| 3096 |
|
| 3097 |
case PARTITION_STRATEGY_RANGE: |
| 3098 |
{ |
| 3099 |
PartitionRangeBound *lower, |
| 3100 |
*upper; |
| 3101 |
int cmpval; |
| 3102 |
|
| 3103 |
Assert(spec->strategy == PARTITION_STRATEGY_RANGE); |
| 3104 |
lower = make_one_partition_rbound(key, -1, spec->lowerdatums, true); |
| 3105 |
upper = make_one_partition_rbound(key, -1, spec->upperdatums, false); |
| 3106 |
|
| 3107 |
/* |
| 3108 |
* First check if the resulting range would be empty with |
| 3109 |
* specified lower and upper bounds. partition_rbound_cmp |
| 3110 |
* cannot return zero here, since the lower-bound flags are |
| 3111 |
* different. |
| 3112 |
*/ |
| 3113 |
cmpval = partition_rbound_cmp(key->partnatts, |
| 3114 |
key->partsupfunc, |
| 3115 |
key->partcollation, |
| 3116 |
lower->datums, lower->kind, |
| 3117 |
true, upper); |
| 3118 |
Assert(cmpval != 0); |
| 3119 |
if (cmpval > 0) |
| 3120 |
{ |
| 3121 |
/* Point to problematic key in the lower datums list. */ |
| 3122 |
PartitionRangeDatum *datum = list_nth(spec->lowerdatums, |
| 3123 |
cmpval - 1); |
| 3124 |
|
| 3125 |
ereport(ERROR, |
| 3126 |
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 3127 |
errmsg("empty range bound specified for partition \"%s\"", |
| 3128 |
relname), |
| 3129 |
errdetail("Specified lower bound %s is greater than or equal to upper bound %s.", |
| 3130 |
get_range_partbound_string(spec->lowerdatums), |
| 3131 |
get_range_partbound_string(spec->upperdatums)), |
| 3132 |
parser_errposition(pstate, datum->location))); |
| 3133 |
} |
| 3134 |
|
| 3135 |
if (partdesc->nparts > 0) |
| 3136 |
{ |
| 3137 |
int offset; |
| 3138 |
|
| 3139 |
Assert(boundinfo && |
| 3140 |
boundinfo->strategy == PARTITION_STRATEGY_RANGE && |
| 3141 |
(boundinfo->ndatums > 0 || |
| 3142 |
partition_bound_has_default(boundinfo))); |
| 3143 |
|
| 3144 |
/* |
| 3145 |
* Test whether the new lower bound (which is treated |
| 3146 |
* inclusively as part of the new partition) lies inside |
| 3147 |
* an existing partition, or in a gap. |
| 3148 |
* |
| 3149 |
* If it's inside an existing partition, the bound at |
| 3150 |
* offset + 1 will be the upper bound of that partition, |
| 3151 |
* and its index will be >= 0. |
| 3152 |
* |
| 3153 |
* If it's in a gap, the bound at offset + 1 will be the |
| 3154 |
* lower bound of the next partition, and its index will |
| 3155 |
* be -1. This is also true if there is no next partition, |
| 3156 |
* since the index array is initialised with an extra -1 |
| 3157 |
* at the end. |
| 3158 |
*/ |
| 3159 |
offset = partition_range_bsearch(key->partnatts, |
| 3160 |
key->partsupfunc, |
| 3161 |
key->partcollation, |
| 3162 |
boundinfo, lower, |
| 3163 |
&cmpval); |
| 3164 |
|
| 3165 |
if (boundinfo->indexes[offset + 1] < 0) |
| 3166 |
{ |
| 3167 |
/* |
| 3168 |
* Check that the new partition will fit in the gap. |
| 3169 |
* For it to fit, the new upper bound must be less |
| 3170 |
* than or equal to the lower bound of the next |
| 3171 |
* partition, if there is one. |
| 3172 |
*/ |
| 3173 |
if (offset + 1 < boundinfo->ndatums) |
| 3174 |
{ |
| 3175 |
Datum *datums; |
| 3176 |
PartitionRangeDatumKind *kind; |
| 3177 |
bool is_lower; |
| 3178 |
|
| 3179 |
datums = boundinfo->datums[offset + 1]; |
| 3180 |
kind = boundinfo->kind[offset + 1]; |
| 3181 |
is_lower = (boundinfo->indexes[offset + 1] == -1); |
| 3182 |
|
| 3183 |
cmpval = partition_rbound_cmp(key->partnatts, |
| 3184 |
key->partsupfunc, |
| 3185 |
key->partcollation, |
| 3186 |
datums, kind, |
| 3187 |
is_lower, upper); |
| 3188 |
if (cmpval < 0) |
| 3189 |
{ |
| 3190 |
/* |
| 3191 |
* Point to problematic key in the upper |
| 3192 |
* datums list. |
| 3193 |
*/ |
| 3194 |
PartitionRangeDatum *datum = |
| 3195 |
list_nth(spec->upperdatums, abs(cmpval) - 1); |
| 3196 |
|
| 3197 |
/* |
| 3198 |
* The new partition overlaps with the |
| 3199 |
* existing partition between offset + 1 and |
| 3200 |
* offset + 2. |
| 3201 |
*/ |
| 3202 |
overlap = true; |
| 3203 |
overlap_location = datum->location; |
| 3204 |
with = boundinfo->indexes[offset + 2]; |
| 3205 |
} |
| 3206 |
} |
| 3207 |
} |
| 3208 |
else |
| 3209 |
{ |
| 3210 |
/* |
| 3211 |
* The new partition overlaps with the existing |
| 3212 |
* partition between offset and offset + 1. |
| 3213 |
*/ |
| 3214 |
PartitionRangeDatum *datum; |
| 3215 |
|
| 3216 |
/* |
| 3217 |
* Point to problematic key in the lower datums list; |
| 3218 |
* if we have equality, point to the first one. |
| 3219 |
*/ |
| 3220 |
datum = cmpval == 0 ? linitial(spec->lowerdatums) : |
| 3221 |
list_nth(spec->lowerdatums, abs(cmpval) - 1); |
| 3222 |
overlap = true; |
| 3223 |
overlap_location = datum->location; |
| 3224 |
with = boundinfo->indexes[offset + 1]; |
| 3225 |
} |
| 3226 |
} |
| 3227 |
|
| 3228 |
break; |
| 3229 |
} |
| 3230 |
} |
| 3231 |
|
| 3232 |
if (overlap) |
| 3233 |
{ |
| 3234 |
Assert(with >= 0); |
| 3235 |
ereport(ERROR, |
| 3236 |
(errcode(ERRCODE_INVALID_OBJECT_DEFINITION), |
| 3237 |
errmsg("partition \"%s\" would overlap partition \"%s\"", |
| 3238 |
relname, get_rel_name(partdesc->oids[with])), |
| 3239 |
parser_errposition(pstate, overlap_location))); |
| 3240 |
} |
| 3241 |
} |
| 3242 |
|
| 3243 |
/* |
| 3244 |
* check_default_partition_contents |
| 3245 |
* |
| 3246 |
* This function checks if there exists a row in the default partition that |
| 3247 |
* would properly belong to the new partition being added. If it finds one, |
| 3248 |
* it throws an error. |
| 3249 |
*/ |
| 3250 |
void |
| 3251 |
check_default_partition_contents(Relation parent, Relation default_rel, |
| 3252 |
PartitionBoundSpec *new_spec) |
| 3253 |
{ |
| 3254 |
List *new_part_constraints; |
| 3255 |
List *def_part_constraints; |
| 3256 |
List *all_parts; |
| 3257 |
ListCell *lc; |
| 3258 |
|
| 3259 |
new_part_constraints = (new_spec->strategy == PARTITION_STRATEGY_LIST) |
| 3260 |
? get_qual_for_list(parent, new_spec) |
| 3261 |
: get_qual_for_range(parent, new_spec, false); |
| 3262 |
def_part_constraints = |
| 3263 |
get_proposed_default_constraint(new_part_constraints); |
| 3264 |
|
| 3265 |
/* |
| 3266 |
* Map the Vars in the constraint expression from parent's attnos to |
| 3267 |
* default_rel's. |
| 3268 |
*/ |
| 3269 |
def_part_constraints = |
| 3270 |
map_partition_varattnos(def_part_constraints, 1, default_rel, |
| 3271 |
parent); |
| 3272 |
|
| 3273 |
/* |
| 3274 |
* If the existing constraints on the default partition imply that it will |
| 3275 |
* not contain any row that would belong to the new partition, we can |
| 3276 |
* avoid scanning the default partition. |
| 3277 |
*/ |
| 3278 |
if (PartConstraintImpliedByRelConstraint(default_rel, def_part_constraints)) |
| 3279 |
{ |
| 3280 |
ereport(DEBUG1, |
| 3281 |
(errmsg_internal("updated partition constraint for default partition \"%s\" is implied by existing constraints", |
| 3282 |
RelationGetRelationName(default_rel)))); |
| 3283 |
return; |
| 3284 |
} |
| 3285 |
|
| 3286 |
/* |
| 3287 |
* Scan the default partition and its subpartitions, and check for rows |
| 3288 |
* that do not satisfy the revised partition constraints. |
| 3289 |
*/ |
| 3290 |
if (default_rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE) |
| 3291 |
all_parts = find_all_inheritors(RelationGetRelid(default_rel), |
| 3292 |
AccessExclusiveLock, NULL); |
| 3293 |
else |
| 3294 |
all_parts = list_make1_oid(RelationGetRelid(default_rel)); |
| 3295 |
|
| 3296 |
foreach(lc, all_parts) |
| 3297 |
{ |
| 3298 |
Oid part_relid = lfirst_oid(lc); |
| 3299 |
Relation part_rel; |
| 3300 |
Expr *partition_constraint; |
| 3301 |
EState *estate; |
| 3302 |
ExprState *partqualstate = NULL; |
| 3303 |
Snapshot snapshot; |
| 3304 |
ExprContext *econtext; |
| 3305 |
TableScanDesc scan; |
| 3306 |
MemoryContext oldCxt; |
| 3307 |
TupleTableSlot *tupslot; |
| 3308 |
|
| 3309 |
/* Lock already taken above. */ |
| 3310 |
if (part_relid != RelationGetRelid(default_rel)) |
| 3311 |
{ |
| 3312 |
part_rel = table_open(part_relid, NoLock); |
| 3313 |
|
| 3314 |
/* |
| 3315 |
* Map the Vars in the constraint expression from default_rel's |
| 3316 |
* the sub-partition's. |
| 3317 |
*/ |
| 3318 |
partition_constraint = make_ands_explicit(def_part_constraints); |
| 3319 |
partition_constraint = (Expr *) |
| 3320 |
map_partition_varattnos((List *) partition_constraint, 1, |
| 3321 |
part_rel, default_rel); |
| 3322 |
|
| 3323 |
/* |
| 3324 |
* If the partition constraints on default partition child imply |
| 3325 |
* that it will not contain any row that would belong to the new |
| 3326 |
* partition, we can avoid scanning the child table. |
| 3327 |
*/ |
| 3328 |
if (PartConstraintImpliedByRelConstraint(part_rel, |
| 3329 |
def_part_constraints)) |
| 3330 |
{ |
| 3331 |
ereport(DEBUG1, |
| 3332 |
(errmsg_internal("updated partition constraint for default partition \"%s\" is implied by existing constraints", |
| 3333 |
RelationGetRelationName(part_rel)))); |
| 3334 |
|
| 3335 |
table_close(part_rel, NoLock); |
| 3336 |
continue; |
| 3337 |
} |
| 3338 |
} |
| 3339 |
else |
| 3340 |
{ |
| 3341 |
part_rel = default_rel; |
| 3342 |
partition_constraint = make_ands_explicit(def_part_constraints); |
| 3343 |
} |
| 3344 |
|
| 3345 |
/* |
| 3346 |
* Only RELKIND_RELATION relations (i.e. leaf partitions) need to be |
| 3347 |
* scanned. |
| 3348 |
*/ |
| 3349 |
if (part_rel->rd_rel->relkind != RELKIND_RELATION) |
| 3350 |
{ |
| 3351 |
if (part_rel->rd_rel->relkind == RELKIND_FOREIGN_TABLE) |
| 3352 |
ereport(WARNING, |
| 3353 |
(errcode(ERRCODE_CHECK_VIOLATION), |
| 3354 |
errmsg("skipped scanning foreign table \"%s\" which is a partition of default partition \"%s\"", |
| 3355 |
RelationGetRelationName(part_rel), |
| 3356 |
RelationGetRelationName(default_rel)))); |
| 3357 |
|
| 3358 |
if (RelationGetRelid(default_rel) != RelationGetRelid(part_rel)) |
| 3359 |
table_close(part_rel, NoLock); |
| 3360 |
|
| 3361 |
continue; |
| 3362 |
} |
| 3363 |
|
| 3364 |
estate = CreateExecutorState(); |
| 3365 |
|
| 3366 |
/* Build expression execution states for partition check quals */ |
| 3367 |
partqualstate = ExecPrepareExpr(partition_constraint, estate); |
| 3368 |
|
| 3369 |
econtext = GetPerTupleExprContext(estate); |
| 3370 |
snapshot = RegisterSnapshot(GetLatestSnapshot()); |
| 3371 |
tupslot = table_slot_create(part_rel, &estate->es_tupleTable); |
| 3372 |
scan = table_beginscan(part_rel, snapshot, 0, NULL); |
| 3373 |
|
| 3374 |
/* |
| 3375 |
* Switch to per-tuple memory context and reset it for each tuple |
| 3376 |
* produced, so we don't leak memory. |
| 3377 |
*/ |
| 3378 |
oldCxt = MemoryContextSwitchTo(GetPerTupleMemoryContext(estate)); |
| 3379 |
|
| 3380 |
while (table_scan_getnextslot(scan, ForwardScanDirection, tupslot)) |
| 3381 |
{ |
| 3382 |
econtext->ecxt_scantuple = tupslot; |
| 3383 |
|
| 3384 |
if (!ExecCheck(partqualstate, econtext)) |
| 3385 |
ereport(ERROR, |
| 3386 |
(errcode(ERRCODE_CHECK_VIOLATION), |
| 3387 |
errmsg("updated partition constraint for default partition \"%s\" would be violated by some row", |
| 3388 |
RelationGetRelationName(default_rel)), |
| 3389 |
errtable(default_rel))); |
| 3390 |
|
| 3391 |
ResetExprContext(econtext); |
| 3392 |
CHECK_FOR_INTERRUPTS(); |
| 3393 |
} |
| 3394 |
|
| 3395 |
MemoryContextSwitchTo(oldCxt); |
| 3396 |
table_endscan(scan); |
| 3397 |
UnregisterSnapshot(snapshot); |
| 3398 |
ExecDropSingleTupleTableSlot(tupslot); |
| 3399 |
FreeExecutorState(estate); |
| 3400 |
|
| 3401 |
if (RelationGetRelid(default_rel) != RelationGetRelid(part_rel)) |
| 3402 |
table_close(part_rel, NoLock); /* keep the lock until commit */ |
| 3403 |
} |
| 3404 |
} |
| 3405 |
|
| 3406 |
/* |
| 3407 |
* get_hash_partition_greatest_modulus |
| 3408 |
* |
| 3409 |
* Returns the greatest modulus of the hash partition bound. |
| 3410 |
* This is no longer used in the core code, but we keep it around |
| 3411 |
* in case external modules are using it. |
| 3412 |
*/ |
| 3413 |
int |
| 3414 |
get_hash_partition_greatest_modulus(PartitionBoundInfo bound) |
| 3415 |
{ |
| 3416 |
Assert(bound && bound->strategy == PARTITION_STRATEGY_HASH); |
| 3417 |
return bound->nindexes; |
| 3418 |
} |
| 3419 |
|
| 3420 |
/* |
| 3421 |
* make_one_partition_rbound |
| 3422 |
* |
| 3423 |
* Return a PartitionRangeBound given a list of PartitionRangeDatum elements |
| 3424 |
* and a flag telling whether the bound is lower or not. Made into a function |
| 3425 |
* because there are multiple sites that want to use this facility. |
| 3426 |
*/ |
| 3427 |
static PartitionRangeBound * |
| 3428 |
make_one_partition_rbound(PartitionKey key, int index, List *datums, bool lower) |
| 3429 |
{ |
| 3430 |
PartitionRangeBound *bound; |
| 3431 |
ListCell *lc; |
| 3432 |
int i; |
| 3433 |
|
| 3434 |
Assert(datums != NIL); |
| 3435 |
|
| 3436 |
bound = (PartitionRangeBound *) palloc0(sizeof(PartitionRangeBound)); |
| 3437 |
bound->index = index; |
| 3438 |
bound->datums = (Datum *) palloc0(key->partnatts * sizeof(Datum)); |
| 3439 |
bound->kind = (PartitionRangeDatumKind *) palloc0(key->partnatts * |
| 3440 |
sizeof(PartitionRangeDatumKind)); |
| 3441 |
bound->lower = lower; |
| 3442 |
|
| 3443 |
i = 0; |
| 3444 |
foreach(lc, datums) |
| 3445 |
{ |
| 3446 |
PartitionRangeDatum *datum = lfirst_node(PartitionRangeDatum, lc); |
| 3447 |
|
| 3448 |
/* What's contained in this range datum? */ |
| 3449 |
bound->kind[i] = datum->kind; |
| 3450 |
|
| 3451 |
if (datum->kind == PARTITION_RANGE_DATUM_VALUE) |
| 3452 |
{ |
| 3453 |
Const *val = castNode(Const, datum->value); |
| 3454 |
|
| 3455 |
if (val->constisnull) |
| 3456 |
elog(ERROR, "invalid range bound datum"); |
| 3457 |
bound->datums[i] = val->constvalue; |
| 3458 |
} |
| 3459 |
|
| 3460 |
i++; |
| 3461 |
} |
| 3462 |
|
| 3463 |
return bound; |
| 3464 |
} |
| 3465 |
|
| 3466 |
/* |
| 3467 |
* partition_rbound_cmp |
| 3468 |
* |
| 3469 |
* For two range bounds this decides whether the 1st one (specified by |
| 3470 |
* datums1, kind1, and lower1) is <, =, or > the bound specified in *b2. |
| 3471 |
* |
| 3472 |
* 0 is returned if they are equal, otherwise a non-zero integer whose sign |
| 3473 |
* indicates the ordering, and whose absolute value gives the 1-based |
| 3474 |
* partition key number of the first mismatching column. |
| 3475 |
* |
| 3476 |
* partnatts, partsupfunc and partcollation give the number of attributes in the |
| 3477 |
* bounds to be compared, comparison function to be used and the collations of |
| 3478 |
* attributes, respectively. |
| 3479 |
* |
| 3480 |
* Note that if the values of the two range bounds compare equal, then we take |
| 3481 |
* into account whether they are upper or lower bounds, and an upper bound is |
| 3482 |
* considered to be smaller than a lower bound. This is important to the way |
| 3483 |
* that RelationBuildPartitionDesc() builds the PartitionBoundInfoData |
| 3484 |
* structure, which only stores the upper bound of a common boundary between |
| 3485 |
* two contiguous partitions. |
| 3486 |
*/ |
| 3487 |
static int32 |
| 3488 |
partition_rbound_cmp(int partnatts, FmgrInfo *partsupfunc, |
| 3489 |
Oid *partcollation, |
| 3490 |
Datum *datums1, PartitionRangeDatumKind *kind1, |
| 3491 |
bool lower1, PartitionRangeBound *b2) |
| 3492 |
{ |
| 3493 |
int32 colnum = 0; |
| 3494 |
int32 cmpval = 0; /* placate compiler */ |
| 3495 |
int i; |
| 3496 |
Datum *datums2 = b2->datums; |
| 3497 |
PartitionRangeDatumKind *kind2 = b2->kind; |
| 3498 |
bool lower2 = b2->lower; |
| 3499 |
|
| 3500 |
for (i = 0; i < partnatts; i++) |
| 3501 |
{ |
| 3502 |
/* Track column number in case we need it for result */ |
| 3503 |
colnum++; |
| 3504 |
|
| 3505 |
/* |
| 3506 |
* First, handle cases where the column is unbounded, which should not |
| 3507 |
* invoke the comparison procedure, and should not consider any later |
| 3508 |
* columns. Note that the PartitionRangeDatumKind enum elements |
| 3509 |
* compare the same way as the values they represent. |
| 3510 |
*/ |
| 3511 |
if (kind1[i] < kind2[i]) |
| 3512 |
return -colnum; |
| 3513 |
else if (kind1[i] > kind2[i]) |
| 3514 |
return colnum; |
| 3515 |
else if (kind1[i] != PARTITION_RANGE_DATUM_VALUE) |
| 3516 |
{ |
| 3517 |
/* |
| 3518 |
* The column bounds are both MINVALUE or both MAXVALUE. No later |
| 3519 |
* columns should be considered, but we still need to compare |
| 3520 |
* whether they are upper or lower bounds. |
| 3521 |
*/ |
| 3522 |
break; |
| 3523 |
} |
| 3524 |
|
| 3525 |
cmpval = DatumGetInt32(FunctionCall2Coll(&partsupfunc[i], |
| 3526 |
partcollation[i], |
| 3527 |
datums1[i], |
| 3528 |
datums2[i])); |
| 3529 |
if (cmpval != 0) |
| 3530 |
break; |
| 3531 |
} |
| 3532 |
|
| 3533 |
/* |
| 3534 |
* If the comparison is anything other than equal, we're done. If they |
| 3535 |
* compare equal though, we still have to consider whether the boundaries |
| 3536 |
* are inclusive or exclusive. Exclusive one is considered smaller of the |
| 3537 |
* two. |
| 3538 |
*/ |
| 3539 |
if (cmpval == 0 && lower1 != lower2) |
| 3540 |
cmpval = lower1 ? 1 : -1; |
| 3541 |
|
| 3542 |
return cmpval == 0 ? 0 : (cmpval < 0 ? -colnum : colnum); |
| 3543 |
} |
| 3544 |
|
| 3545 |
/* |
| 3546 |
* partition_rbound_datum_cmp |
| 3547 |
* |
| 3548 |
* Return whether range bound (specified in rb_datums and rb_kind) |
| 3549 |
* is <, =, or > partition key of tuple (tuple_datums) |
| 3550 |
* |
| 3551 |
* n_tuple_datums, partsupfunc and partcollation give number of attributes in |
| 3552 |
* the bounds to be compared, comparison function to be used and the collations |
| 3553 |
* of attributes resp. |
| 3554 |
*/ |
| 3555 |
int32 |
| 3556 |
partition_rbound_datum_cmp(FmgrInfo *partsupfunc, Oid *partcollation, |
| 3557 |
Datum *rb_datums, PartitionRangeDatumKind *rb_kind, |
| 3558 |
Datum *tuple_datums, int n_tuple_datums) |
| 3559 |
{ |
| 3560 |
int i; |
| 3561 |
int32 cmpval = -1; |
| 3562 |
|
| 3563 |
for (i = 0; i < n_tuple_datums; i++) |
| 3564 |
{ |
| 3565 |
if (rb_kind[i] == PARTITION_RANGE_DATUM_MINVALUE) |
| 3566 |
return -1; |
| 3567 |
else if (rb_kind[i] == PARTITION_RANGE_DATUM_MAXVALUE) |
| 3568 |
return 1; |
| 3569 |
|
| 3570 |
cmpval = DatumGetInt32(FunctionCall2Coll(&partsupfunc[i], |
| 3571 |
partcollation[i], |
| 3572 |
rb_datums[i], |
| 3573 |
tuple_datums[i])); |
| 3574 |
if (cmpval != 0) |
| 3575 |
break; |
| 3576 |
} |
| 3577 |
|
| 3578 |
return cmpval; |
| 3579 |
} |
| 3580 |
|
| 3581 |
/* |
| 3582 |
* partition_hbound_cmp |
| 3583 |
* |
| 3584 |
* Compares modulus first, then remainder if modulus is equal. |
| 3585 |
*/ |
| 3586 |
static int32 |
| 3587 |
partition_hbound_cmp(int modulus1, int remainder1, int modulus2, int remainder2) |
| 3588 |
{ |
| 3589 |
if (modulus1 < modulus2) |
| 3590 |
return -1; |
| 3591 |
if (modulus1 > modulus2) |
| 3592 |
return 1; |
| 3593 |
if (modulus1 == modulus2 && remainder1 != remainder2) |
| 3594 |
return (remainder1 > remainder2) ? 1 : -1; |
| 3595 |
return 0; |
| 3596 |
} |
| 3597 |
|
| 3598 |
/* |
| 3599 |
* partition_list_bsearch |
| 3600 |
* Returns the index of the greatest bound datum that is less than equal |
| 3601 |
* to the given value or -1 if all of the bound datums are greater |
| 3602 |
* |
| 3603 |
* *is_equal is set to true if the bound datum at the returned index is equal |
| 3604 |
* to the input value. |
| 3605 |
*/ |
| 3606 |
int |
| 3607 |
partition_list_bsearch(FmgrInfo *partsupfunc, Oid *partcollation, |
| 3608 |
PartitionBoundInfo boundinfo, |
| 3609 |
Datum value, bool *is_equal) |
| 3610 |
{ |
| 3611 |
int lo, |
| 3612 |
hi, |
| 3613 |
mid; |
| 3614 |
|
| 3615 |
lo = -1; |
| 3616 |
hi = boundinfo->ndatums - 1; |
| 3617 |
while (lo < hi) |
| 3618 |
{ |
| 3619 |
int32 cmpval; |
| 3620 |
|
| 3621 |
mid = (lo + hi + 1) / 2; |
| 3622 |
cmpval = DatumGetInt32(FunctionCall2Coll(&partsupfunc[0], |
| 3623 |
partcollation[0], |
| 3624 |
boundinfo->datums[mid][0], |
| 3625 |
value)); |
| 3626 |
if (cmpval <= 0) |
| 3627 |
{ |
| 3628 |
lo = mid; |
| 3629 |
*is_equal = (cmpval == 0); |
| 3630 |
if (*is_equal) |
| 3631 |
break; |
| 3632 |
} |
| 3633 |
else |
| 3634 |
hi = mid - 1; |
| 3635 |
} |
| 3636 |
|
| 3637 |
return lo; |
| 3638 |
} |
| 3639 |
|
| 3640 |
/* |
| 3641 |
* partition_range_bsearch |
| 3642 |
* Returns the index of the greatest range bound that is less than or |
| 3643 |
* equal to the given range bound or -1 if all of the range bounds are |
| 3644 |
* greater |
| 3645 |
* |
| 3646 |
* Upon return from this function, *cmpval is set to 0 if the bound at the |
| 3647 |
* returned index matches the input range bound exactly, otherwise a |
| 3648 |
* non-zero integer whose sign indicates the ordering, and whose absolute |
| 3649 |
* value gives the 1-based partition key number of the first mismatching |
| 3650 |
* column. |
| 3651 |
*/ |
| 3652 |
static int |
| 3653 |
partition_range_bsearch(int partnatts, FmgrInfo *partsupfunc, |
| 3654 |
Oid *partcollation, |
| 3655 |
PartitionBoundInfo boundinfo, |
| 3656 |
PartitionRangeBound *probe, int32 *cmpval) |
| 3657 |
{ |
| 3658 |
int lo, |
| 3659 |
hi, |
| 3660 |
mid; |
| 3661 |
|
| 3662 |
lo = -1; |
| 3663 |
hi = boundinfo->ndatums - 1; |
| 3664 |
while (lo < hi) |
| 3665 |
{ |
| 3666 |
mid = (lo + hi + 1) / 2; |
| 3667 |
*cmpval = partition_rbound_cmp(partnatts, partsupfunc, |
| 3668 |
partcollation, |
| 3669 |
boundinfo->datums[mid], |
| 3670 |
boundinfo->kind[mid], |
| 3671 |
(boundinfo->indexes[mid] == -1), |
| 3672 |
probe); |
| 3673 |
if (*cmpval <= 0) |
| 3674 |
{ |
| 3675 |
lo = mid; |
| 3676 |
if (*cmpval == 0) |
| 3677 |
break; |
| 3678 |
} |
| 3679 |
else |
| 3680 |
hi = mid - 1; |
| 3681 |
} |
| 3682 |
|
| 3683 |
return lo; |
| 3684 |
} |
| 3685 |
|
| 3686 |
/* |
| 3687 |
* partition_range_datum_bsearch |
| 3688 |
* Returns the index of the greatest range bound that is less than or |
| 3689 |
* equal to the given tuple or -1 if all of the range bounds are greater |
| 3690 |
* |
| 3691 |
* *is_equal is set to true if the range bound at the returned index is equal |
| 3692 |
* to the input tuple. |
| 3693 |
*/ |
| 3694 |
int |
| 3695 |
partition_range_datum_bsearch(FmgrInfo *partsupfunc, Oid *partcollation, |
| 3696 |
PartitionBoundInfo boundinfo, |
| 3697 |
int nvalues, Datum *values, bool *is_equal) |
| 3698 |
{ |
| 3699 |
int lo, |
| 3700 |
hi, |
| 3701 |
mid; |
| 3702 |
|
| 3703 |
lo = -1; |
| 3704 |
hi = boundinfo->ndatums - 1; |
| 3705 |
while (lo < hi) |
| 3706 |
{ |
| 3707 |
int32 cmpval; |
| 3708 |
|
| 3709 |
mid = (lo + hi + 1) / 2; |
| 3710 |
cmpval = partition_rbound_datum_cmp(partsupfunc, |
| 3711 |
partcollation, |
| 3712 |
boundinfo->datums[mid], |
| 3713 |
boundinfo->kind[mid], |
| 3714 |
values, |
| 3715 |
nvalues); |
| 3716 |
if (cmpval <= 0) |
| 3717 |
{ |
| 3718 |
lo = mid; |
| 3719 |
*is_equal = (cmpval == 0); |
| 3720 |
|
| 3721 |
if (*is_equal) |
| 3722 |
break; |
| 3723 |
} |
| 3724 |
else |
| 3725 |
hi = mid - 1; |
| 3726 |
} |
| 3727 |
|
| 3728 |
return lo; |
| 3729 |
} |
| 3730 |
|
| 3731 |
/* |
| 3732 |
* partition_hash_bsearch |
| 3733 |
* Returns the index of the greatest (modulus, remainder) pair that is |
| 3734 |
* less than or equal to the given (modulus, remainder) pair or -1 if |
| 3735 |
* all of them are greater |
| 3736 |
*/ |
| 3737 |
int |
| 3738 |
partition_hash_bsearch(PartitionBoundInfo boundinfo, |
| 3739 |
int modulus, int remainder) |
| 3740 |
{ |
| 3741 |
int lo, |
| 3742 |
hi, |
| 3743 |
mid; |
| 3744 |
|
| 3745 |
lo = -1; |
| 3746 |
hi = boundinfo->ndatums - 1; |
| 3747 |
while (lo < hi) |
| 3748 |
{ |
| 3749 |
int32 cmpval, |
| 3750 |
bound_modulus, |
| 3751 |
bound_remainder; |
| 3752 |
|
| 3753 |
mid = (lo + hi + 1) / 2; |
| 3754 |
bound_modulus = DatumGetInt32(boundinfo->datums[mid][0]); |
| 3755 |
bound_remainder = DatumGetInt32(boundinfo->datums[mid][1]); |
| 3756 |
cmpval = partition_hbound_cmp(bound_modulus, bound_remainder, |
| 3757 |
modulus, remainder); |
| 3758 |
if (cmpval <= 0) |
| 3759 |
{ |
| 3760 |
lo = mid; |
| 3761 |
|
| 3762 |
if (cmpval == 0) |
| 3763 |
break; |
| 3764 |
} |
| 3765 |
else |
| 3766 |
hi = mid - 1; |
| 3767 |
} |
| 3768 |
|
| 3769 |
return lo; |
| 3770 |
} |
| 3771 |
|
| 3772 |
/* |
| 3773 |
* qsort_partition_hbound_cmp |
| 3774 |
* |
| 3775 |
* Hash bounds are sorted by modulus, then by remainder. |
| 3776 |
*/ |
| 3777 |
static int32 |
| 3778 |
qsort_partition_hbound_cmp(const void *a, const void *b) |
| 3779 |
{ |
| 3780 |
const PartitionHashBound *h1 = (const PartitionHashBound *) a; |
| 3781 |
const PartitionHashBound *h2 = (const PartitionHashBound *) b; |
| 3782 |
|
| 3783 |
return partition_hbound_cmp(h1->modulus, h1->remainder, |
| 3784 |
h2->modulus, h2->remainder); |
| 3785 |
} |
| 3786 |
|
| 3787 |
/* |
| 3788 |
* qsort_partition_list_value_cmp |
| 3789 |
* |
| 3790 |
* Compare two list partition bound datums. |
| 3791 |
*/ |
| 3792 |
static int32 |
| 3793 |
qsort_partition_list_value_cmp(const void *a, const void *b, void *arg) |
| 3794 |
{ |
| 3795 |
Datum val1 = ((const PartitionListValue *) a)->value, |
| 3796 |
val2 = ((const PartitionListValue *) b)->value; |
| 3797 |
PartitionKey key = (PartitionKey) arg; |
| 3798 |
|
| 3799 |
return DatumGetInt32(FunctionCall2Coll(&key->partsupfunc[0], |
| 3800 |
key->partcollation[0], |
| 3801 |
val1, val2)); |
| 3802 |
} |
| 3803 |
|
| 3804 |
/* |
| 3805 |
* qsort_partition_rbound_cmp |
| 3806 |
* |
| 3807 |
* Used when sorting range bounds across all range partitions. |
| 3808 |
*/ |
| 3809 |
static int32 |
| 3810 |
qsort_partition_rbound_cmp(const void *a, const void *b, void *arg) |
| 3811 |
{ |
| 3812 |
PartitionRangeBound *b1 = (*(PartitionRangeBound *const *) a); |
| 3813 |
PartitionRangeBound *b2 = (*(PartitionRangeBound *const *) b); |
| 3814 |
PartitionKey key = (PartitionKey) arg; |
| 3815 |
|
| 3816 |
return compare_range_bounds(key->partnatts, key->partsupfunc, |
| 3817 |
key->partcollation, |
| 3818 |
b1, b2); |
| 3819 |
} |
| 3820 |
|
| 3821 |
/* |
| 3822 |
* get_partition_operator |
| 3823 |
* |
| 3824 |
* Return oid of the operator of the given strategy for the given partition |
| 3825 |
* key column. It is assumed that the partitioning key is of the same type as |
| 3826 |
* the chosen partitioning opclass, or at least binary-compatible. In the |
| 3827 |
* latter case, *need_relabel is set to true if the opclass is not of a |
| 3828 |
* polymorphic type (indicating a RelabelType node needed on top), otherwise |
| 3829 |
* false. |
| 3830 |
*/ |
| 3831 |
static Oid |
| 3832 |
get_partition_operator(PartitionKey key, int col, StrategyNumber strategy, |
| 3833 |
bool *need_relabel) |
| 3834 |
{ |
| 3835 |
Oid operoid; |
| 3836 |
|
| 3837 |
/* |
| 3838 |
* Get the operator in the partitioning opfamily using the opclass' |
| 3839 |
* declared input type as both left- and righttype. |
| 3840 |
*/ |
| 3841 |
operoid = get_opfamily_member(key->partopfamily[col], |
| 3842 |
key->partopcintype[col], |
| 3843 |
key->partopcintype[col], |
| 3844 |
strategy); |
| 3845 |
if (!OidIsValid(operoid)) |
| 3846 |
elog(ERROR, "missing operator %d(%u,%u) in partition opfamily %u", |
| 3847 |
strategy, key->partopcintype[col], key->partopcintype[col], |
| 3848 |
key->partopfamily[col]); |
| 3849 |
|
| 3850 |
/* |
| 3851 |
* If the partition key column is not of the same type as the operator |
| 3852 |
* class and not polymorphic, tell caller to wrap the non-Const expression |
| 3853 |
* in a RelabelType. This matches what parse_coerce.c does. |
| 3854 |
*/ |
| 3855 |
*need_relabel = (key->parttypid[col] != key->partopcintype[col] && |
| 3856 |
key->partopcintype[col] != RECORDOID && |
| 3857 |
!IsPolymorphicType(key->partopcintype[col])); |
| 3858 |
|
| 3859 |
return operoid; |
| 3860 |
} |
| 3861 |
|
| 3862 |
/* |
| 3863 |
* make_partition_op_expr |
| 3864 |
* Returns an Expr for the given partition key column with arg1 and |
| 3865 |
* arg2 as its leftop and rightop, respectively |
| 3866 |
*/ |
| 3867 |
static Expr * |
| 3868 |
make_partition_op_expr(PartitionKey key, int keynum, |
| 3869 |
uint16 strategy, Expr *arg1, Expr *arg2) |
| 3870 |
{ |
| 3871 |
Oid operoid; |
| 3872 |
bool need_relabel = false; |
| 3873 |
Expr *result = NULL; |
| 3874 |
|
| 3875 |
/* Get the correct btree operator for this partitioning column */ |
| 3876 |
operoid = get_partition_operator(key, keynum, strategy, &need_relabel); |
| 3877 |
|
| 3878 |
/* |
| 3879 |
* Chosen operator may be such that the non-Const operand needs to be |
| 3880 |
* coerced, so apply the same; see the comment in |
| 3881 |
* get_partition_operator(). |
| 3882 |
*/ |
| 3883 |
if (!IsA(arg1, Const) && |
| 3884 |
(need_relabel || |
| 3885 |
key->partcollation[keynum] != key->parttypcoll[keynum])) |
| 3886 |
arg1 = (Expr *) makeRelabelType(arg1, |
| 3887 |
key->partopcintype[keynum], |
| 3888 |
-1, |
| 3889 |
key->partcollation[keynum], |
| 3890 |
COERCE_EXPLICIT_CAST); |
| 3891 |
|
| 3892 |
/* Generate the actual expression */ |
| 3893 |
switch (key->strategy) |
| 3894 |
{ |
| 3895 |
case PARTITION_STRATEGY_LIST: |
| 3896 |
{ |
| 3897 |
List *elems = (List *) arg2; |
| 3898 |
int nelems = list_length(elems); |
| 3899 |
|
| 3900 |
Assert(nelems >= 1); |
| 3901 |
Assert(keynum == 0); |
| 3902 |
|
| 3903 |
if (nelems > 1 && |
| 3904 |
!type_is_array(key->parttypid[keynum])) |
| 3905 |
{ |
| 3906 |
ArrayExpr *arrexpr; |
| 3907 |
ScalarArrayOpExpr *saopexpr; |
| 3908 |
|
| 3909 |
/* Construct an ArrayExpr for the right-hand inputs */ |
| 3910 |
arrexpr = makeNode(ArrayExpr); |
| 3911 |
arrexpr->array_typeid = |
| 3912 |
get_array_type(key->parttypid[keynum]); |
| 3913 |
arrexpr->array_collid = key->parttypcoll[keynum]; |
| 3914 |
arrexpr->element_typeid = key->parttypid[keynum]; |
| 3915 |
arrexpr->elements = elems; |
| 3916 |
arrexpr->multidims = false; |
| 3917 |
arrexpr->location = -1; |
| 3918 |
|
| 3919 |
/* Build leftop = ANY (rightop) */ |
| 3920 |
saopexpr = makeNode(ScalarArrayOpExpr); |
| 3921 |
saopexpr->opno = operoid; |
| 3922 |
saopexpr->opfuncid = get_opcode(operoid); |
| 3923 |
saopexpr->hashfuncid = InvalidOid; |
| 3924 |
saopexpr->negfuncid = InvalidOid; |
| 3925 |
saopexpr->useOr = true; |
| 3926 |
saopexpr->inputcollid = key->partcollation[keynum]; |
| 3927 |
saopexpr->args = list_make2(arg1, arrexpr); |
| 3928 |
saopexpr->location = -1; |
| 3929 |
|
| 3930 |
result = (Expr *) saopexpr; |
| 3931 |
} |
| 3932 |
else |
| 3933 |
{ |
| 3934 |
List *elemops = NIL; |
| 3935 |
ListCell *lc; |
| 3936 |
|
| 3937 |
foreach(lc, elems) |
| 3938 |
{ |
| 3939 |
Expr *elem = lfirst(lc), |
| 3940 |
*elemop; |
| 3941 |
|
| 3942 |
elemop = make_opclause(operoid, |
| 3943 |
BOOLOID, |
| 3944 |
false, |
| 3945 |
arg1, elem, |
| 3946 |
InvalidOid, |
| 3947 |
key->partcollation[keynum]); |
| 3948 |
elemops = lappend(elemops, elemop); |
| 3949 |
} |
| 3950 |
|
| 3951 |
result = nelems > 1 ? makeBoolExpr(OR_EXPR, elemops, -1) : linitial(elemops); |
| 3952 |
} |
| 3953 |
break; |
| 3954 |
} |
| 3955 |
|
| 3956 |
case PARTITION_STRATEGY_RANGE: |
| 3957 |
result = make_opclause(operoid, |
| 3958 |
BOOLOID, |
| 3959 |
false, |
| 3960 |
arg1, arg2, |
| 3961 |
InvalidOid, |
| 3962 |
key->partcollation[keynum]); |
| 3963 |
break; |
| 3964 |
|
| 3965 |
case PARTITION_STRATEGY_HASH: |
| 3966 |
Assert(false); |
| 3967 |
break; |
| 3968 |
} |
| 3969 |
|
| 3970 |
return result; |
| 3971 |
} |
| 3972 |
|
| 3973 |
/* |
| 3974 |
* get_qual_for_hash |
| 3975 |
* |
| 3976 |
* Returns a CHECK constraint expression to use as a hash partition's |
| 3977 |
* constraint, given the parent relation and partition bound structure. |
| 3978 |
* |
| 3979 |
* The partition constraint for a hash partition is always a call to the |
| 3980 |
* built-in function satisfies_hash_partition(). |
| 3981 |
*/ |
| 3982 |
static List * |
| 3983 |
get_qual_for_hash(Relation parent, PartitionBoundSpec *spec) |
| 3984 |
{ |
| 3985 |
PartitionKey key = RelationGetPartitionKey(parent); |
| 3986 |
FuncExpr *fexpr; |
| 3987 |
Node *relidConst; |
| 3988 |
Node *modulusConst; |
| 3989 |
Node *remainderConst; |
| 3990 |
List *args; |
| 3991 |
ListCell *partexprs_item; |
| 3992 |
int i; |
| 3993 |
|
| 3994 |
/* Fixed arguments. */ |
| 3995 |
relidConst = (Node *) makeConst(OIDOID, |
| 3996 |
-1, |
| 3997 |
InvalidOid, |
| 3998 |
sizeof(Oid), |
| 3999 |
ObjectIdGetDatum(RelationGetRelid(parent)), |
| 4000 |
false, |
| 4001 |
true); |
| 4002 |
|
| 4003 |
modulusConst = (Node *) makeConst(INT4OID, |
| 4004 |
-1, |
| 4005 |
InvalidOid, |
| 4006 |
sizeof(int32), |
| 4007 |
Int32GetDatum(spec->modulus), |
| 4008 |
false, |
| 4009 |
true); |
| 4010 |
|
| 4011 |
remainderConst = (Node *) makeConst(INT4OID, |
| 4012 |
-1, |
| 4013 |
InvalidOid, |
| 4014 |
sizeof(int32), |
| 4015 |
Int32GetDatum(spec->remainder), |
| 4016 |
false, |
| 4017 |
true); |
| 4018 |
|
| 4019 |
args = list_make3(relidConst, modulusConst, remainderConst); |
| 4020 |
partexprs_item = list_head(key->partexprs); |
| 4021 |
|
| 4022 |
/* Add an argument for each key column. */ |
| 4023 |
for (i = 0; i < key->partnatts; i++) |
| 4024 |
{ |
| 4025 |
Node *keyCol; |
| 4026 |
|
| 4027 |
/* Left operand */ |
| 4028 |
if (key->partattrs[i] != 0) |
| 4029 |
{ |
| 4030 |
keyCol = (Node *) makeVar(1, |
| 4031 |
key->partattrs[i], |
| 4032 |
key->parttypid[i], |
| 4033 |
key->parttypmod[i], |
| 4034 |
key->parttypcoll[i], |
| 4035 |
0); |
| 4036 |
} |
| 4037 |
else |
| 4038 |
{ |
| 4039 |
keyCol = (Node *) copyObject(lfirst(partexprs_item)); |
| 4040 |
partexprs_item = lnext(key->partexprs, partexprs_item); |
| 4041 |
} |
| 4042 |
|
| 4043 |
args = lappend(args, keyCol); |
| 4044 |
} |
| 4045 |
|
| 4046 |
fexpr = makeFuncExpr(F_SATISFIES_HASH_PARTITION, |
| 4047 |
BOOLOID, |
| 4048 |
args, |
| 4049 |
InvalidOid, |
| 4050 |
InvalidOid, |
| 4051 |
COERCE_EXPLICIT_CALL); |
| 4052 |
|
| 4053 |
return list_make1(fexpr); |
| 4054 |
} |
| 4055 |
|
| 4056 |
/* |
| 4057 |
* get_qual_for_list |
| 4058 |
* |
| 4059 |
* Returns an implicit-AND list of expressions to use as a list partition's |
| 4060 |
* constraint, given the parent relation and partition bound structure. |
| 4061 |
* |
| 4062 |
* The function returns NIL for a default partition when it's the only |
| 4063 |
* partition since in that case there is no constraint. |
| 4064 |
*/ |
| 4065 |
static List * |
| 4066 |
get_qual_for_list(Relation parent, PartitionBoundSpec *spec) |
| 4067 |
{ |
| 4068 |
PartitionKey key = RelationGetPartitionKey(parent); |
| 4069 |
List *result; |
| 4070 |
Expr *keyCol; |
| 4071 |
Expr *opexpr; |
| 4072 |
NullTest *nulltest; |
| 4073 |
ListCell *cell; |
| 4074 |
List *elems = NIL; |
| 4075 |
bool list_has_null = false; |
| 4076 |
|
| 4077 |
/* |
| 4078 |
* Only single-column list partitioning is supported, so we are worried |
| 4079 |
* only about the partition key with index 0. |
| 4080 |
*/ |
| 4081 |
Assert(key->partnatts == 1); |
| 4082 |
|
| 4083 |
/* Construct Var or expression representing the partition column */ |
| 4084 |
if (key->partattrs[0] != 0) |
| 4085 |
keyCol = (Expr *) makeVar(1, |
| 4086 |
key->partattrs[0], |
| 4087 |
key->parttypid[0], |
| 4088 |
key->parttypmod[0], |
| 4089 |
key->parttypcoll[0], |
| 4090 |
0); |
| 4091 |
else |
| 4092 |
keyCol = (Expr *) copyObject(linitial(key->partexprs)); |
| 4093 |
|
| 4094 |
/* |
| 4095 |
* For default list partition, collect datums for all the partitions. The |
| 4096 |
* default partition constraint should check that the partition key is |
| 4097 |
* equal to none of those. |
| 4098 |
*/ |
| 4099 |
if (spec->is_default) |
| 4100 |
{ |
| 4101 |
int i; |
| 4102 |
int ndatums = 0; |
| 4103 |
PartitionDesc pdesc = RelationGetPartitionDesc(parent, false); |
| 4104 |
PartitionBoundInfo boundinfo = pdesc->boundinfo; |
| 4105 |
|
| 4106 |
if (boundinfo) |
| 4107 |
{ |
| 4108 |
ndatums = boundinfo->ndatums; |
| 4109 |
|
| 4110 |
if (partition_bound_accepts_nulls(boundinfo)) |
| 4111 |
list_has_null = true; |
| 4112 |
} |
| 4113 |
|
| 4114 |
/* |
| 4115 |
* If default is the only partition, there need not be any partition |
| 4116 |
* constraint on it. |
| 4117 |
*/ |
| 4118 |
if (ndatums == 0 && !list_has_null) |
| 4119 |
return NIL; |
| 4120 |
|
| 4121 |
for (i = 0; i < ndatums; i++) |
| 4122 |
{ |
| 4123 |
Const *val; |
| 4124 |
|
| 4125 |
/* |
| 4126 |
* Construct Const from known-not-null datum. We must be careful |
| 4127 |
* to copy the value, because our result has to be able to outlive |
| 4128 |
* the relcache entry we're copying from. |
| 4129 |
*/ |
| 4130 |
val = makeConst(key->parttypid[0], |
| 4131 |
key->parttypmod[0], |
| 4132 |
key->parttypcoll[0], |
| 4133 |
key->parttyplen[0], |
| 4134 |
datumCopy(*boundinfo->datums[i], |
| 4135 |
key->parttypbyval[0], |
| 4136 |
key->parttyplen[0]), |
| 4137 |
false, /* isnull */ |
| 4138 |
key->parttypbyval[0]); |
| 4139 |
|
| 4140 |
elems = lappend(elems, val); |
| 4141 |
} |
| 4142 |
} |
| 4143 |
else |
| 4144 |
{ |
| 4145 |
/* |
| 4146 |
* Create list of Consts for the allowed values, excluding any nulls. |
| 4147 |
*/ |
| 4148 |
foreach(cell, spec->listdatums) |
| 4149 |
{ |
| 4150 |
Const *val = lfirst_node(Const, cell); |
| 4151 |
|
| 4152 |
if (val->constisnull) |
| 4153 |
list_has_null = true; |
| 4154 |
else |
| 4155 |
elems = lappend(elems, copyObject(val)); |
| 4156 |
} |
| 4157 |
} |
| 4158 |
|
| 4159 |
if (elems) |
| 4160 |
{ |
| 4161 |
/* |
| 4162 |
* Generate the operator expression from the non-null partition |
| 4163 |
* values. |
| 4164 |
*/ |
| 4165 |
opexpr = make_partition_op_expr(key, 0, BTEqualStrategyNumber, |
| 4166 |
keyCol, (Expr *) elems); |
| 4167 |
} |
| 4168 |
else |
| 4169 |
{ |
| 4170 |
/* |
| 4171 |
* If there are no partition values, we don't need an operator |
| 4172 |
* expression. |
| 4173 |
*/ |
| 4174 |
opexpr = NULL; |
| 4175 |
} |
| 4176 |
|
| 4177 |
if (!list_has_null) |
| 4178 |
{ |
| 4179 |
/* |
| 4180 |
* Gin up a "col IS NOT NULL" test that will be ANDed with the main |
| 4181 |
* expression. This might seem redundant, but the partition routing |
| 4182 |
* machinery needs it. |
| 4183 |
*/ |
| 4184 |
nulltest = makeNode(NullTest); |
| 4185 |
nulltest->arg = keyCol; |
| 4186 |
nulltest->nulltesttype = IS_NOT_NULL; |
| 4187 |
nulltest->argisrow = false; |
| 4188 |
nulltest->location = -1; |
| 4189 |
|
| 4190 |
result = opexpr ? list_make2(nulltest, opexpr) : list_make1(nulltest); |
| 4191 |
} |
| 4192 |
else |
| 4193 |
{ |
| 4194 |
/* |
| 4195 |
* Gin up a "col IS NULL" test that will be OR'd with the main |
| 4196 |
* expression. |
| 4197 |
*/ |
| 4198 |
nulltest = makeNode(NullTest); |
| 4199 |
nulltest->arg = keyCol; |
| 4200 |
nulltest->nulltesttype = IS_NULL; |
| 4201 |
nulltest->argisrow = false; |
| 4202 |
nulltest->location = -1; |
| 4203 |
|
| 4204 |
if (opexpr) |
| 4205 |
{ |
| 4206 |
Expr *or; |
| 4207 |
|
| 4208 |
or = makeBoolExpr(OR_EXPR, list_make2(nulltest, opexpr), -1); |
| 4209 |
result = list_make1(or); |
| 4210 |
} |
| 4211 |
else |
| 4212 |
result = list_make1(nulltest); |
| 4213 |
} |
| 4214 |
|
| 4215 |
/* |
| 4216 |
* Note that, in general, applying NOT to a constraint expression doesn't |
| 4217 |
* necessarily invert the set of rows it accepts, because NOT (NULL) is |
| 4218 |
* NULL. However, the partition constraints we construct here never |
| 4219 |
* evaluate to NULL, so applying NOT works as intended. |
| 4220 |
*/ |
| 4221 |
if (spec->is_default) |
| 4222 |
{ |
| 4223 |
result = list_make1(make_ands_explicit(result)); |
| 4224 |
result = list_make1(makeBoolExpr(NOT_EXPR, result, -1)); |
| 4225 |
} |
| 4226 |
|
| 4227 |
return result; |
| 4228 |
} |
| 4229 |
|
| 4230 |
/* |
| 4231 |
* get_qual_for_range |
| 4232 |
* |
| 4233 |
* Returns an implicit-AND list of expressions to use as a range partition's |
| 4234 |
* constraint, given the parent relation and partition bound structure. |
| 4235 |
* |
| 4236 |
* For a multi-column range partition key, say (a, b, c), with (al, bl, cl) |
| 4237 |
* as the lower bound tuple and (au, bu, cu) as the upper bound tuple, we |
| 4238 |
* generate an expression tree of the following form: |
| 4239 |
* |
| 4240 |
* (a IS NOT NULL) and (b IS NOT NULL) and (c IS NOT NULL) |
| 4241 |
* AND |
| 4242 |
* (a > al OR (a = al AND b > bl) OR (a = al AND b = bl AND c >= cl)) |
| 4243 |
* AND |
| 4244 |
* (a < au OR (a = au AND b < bu) OR (a = au AND b = bu AND c < cu)) |
| 4245 |
* |
| 4246 |
* It is often the case that a prefix of lower and upper bound tuples contains |
| 4247 |
* the same values, for example, (al = au), in which case, we will emit an |
| 4248 |
* expression tree of the following form: |
| 4249 |
* |
| 4250 |
* (a IS NOT NULL) and (b IS NOT NULL) and (c IS NOT NULL) |
| 4251 |
* AND |
| 4252 |
* (a = al) |
| 4253 |
* AND |
| 4254 |
* (b > bl OR (b = bl AND c >= cl)) |
| 4255 |
* AND |
| 4256 |
* (b < bu OR (b = bu AND c < cu)) |
| 4257 |
* |
| 4258 |
* If a bound datum is either MINVALUE or MAXVALUE, these expressions are |
| 4259 |
* simplified using the fact that any value is greater than MINVALUE and less |
| 4260 |
* than MAXVALUE. So, for example, if cu = MAXVALUE, c < cu is automatically |
| 4261 |
* true, and we need not emit any expression for it, and the last line becomes |
| 4262 |
* |
| 4263 |
* (b < bu) OR (b = bu), which is simplified to (b <= bu) |
| 4264 |
* |
| 4265 |
* In most common cases with only one partition column, say a, the following |
| 4266 |
* expression tree will be generated: a IS NOT NULL AND a >= al AND a < au |
| 4267 |
* |
| 4268 |
* For default partition, it returns the negation of the constraints of all |
| 4269 |
* the other partitions. |
| 4270 |
* |
| 4271 |
* External callers should pass for_default as false; we set it to true only |
| 4272 |
* when recursing. |
| 4273 |
*/ |
| 4274 |
static List * |
| 4275 |
get_qual_for_range(Relation parent, PartitionBoundSpec *spec, |
| 4276 |
bool for_default) |
| 4277 |
{ |
| 4278 |
List *result = NIL; |
| 4279 |
ListCell *cell1, |
| 4280 |
*cell2, |
| 4281 |
*partexprs_item, |
| 4282 |
*partexprs_item_saved; |
| 4283 |
int i, |
| 4284 |
j; |
| 4285 |
PartitionRangeDatum *ldatum, |
| 4286 |
*udatum; |
| 4287 |
PartitionKey key = RelationGetPartitionKey(parent); |
| 4288 |
Expr *keyCol; |
| 4289 |
Const *lower_val, |
| 4290 |
*upper_val; |
| 4291 |
List *lower_or_arms, |
| 4292 |
*upper_or_arms; |
| 4293 |
int num_or_arms, |
| 4294 |
current_or_arm; |
| 4295 |
ListCell *lower_or_start_datum, |
| 4296 |
*upper_or_start_datum; |
| 4297 |
bool need_next_lower_arm, |
| 4298 |
need_next_upper_arm; |
| 4299 |
|
| 4300 |
if (spec->is_default) |
| 4301 |
{ |
| 4302 |
List *or_expr_args = NIL; |
| 4303 |
PartitionDesc pdesc = RelationGetPartitionDesc(parent, false); |
| 4304 |
Oid *inhoids = pdesc->oids; |
| 4305 |
int nparts = pdesc->nparts, |
| 4306 |
k; |
| 4307 |
|
| 4308 |
for (k = 0; k < nparts; k++) |
| 4309 |
{ |
| 4310 |
Oid inhrelid = inhoids[k]; |
| 4311 |
HeapTuple tuple; |
| 4312 |
Datum datum; |
| 4313 |
PartitionBoundSpec *bspec; |
| 4314 |
|
| 4315 |
tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(inhrelid)); |
| 4316 |
if (!HeapTupleIsValid(tuple)) |
| 4317 |
elog(ERROR, "cache lookup failed for relation %u", inhrelid); |
| 4318 |
|
| 4319 |
datum = SysCacheGetAttrNotNull(RELOID, tuple, |
| 4320 |
Anum_pg_class_relpartbound); |
| 4321 |
bspec = (PartitionBoundSpec *) |
| 4322 |
stringToNode(TextDatumGetCString(datum)); |
| 4323 |
if (!IsA(bspec, PartitionBoundSpec)) |
| 4324 |
elog(ERROR, "expected PartitionBoundSpec"); |
| 4325 |
|
| 4326 |
if (!bspec->is_default) |
| 4327 |
{ |
| 4328 |
List *part_qual; |
| 4329 |
|
| 4330 |
part_qual = get_qual_for_range(parent, bspec, true); |
| 4331 |
|
| 4332 |
/* |
| 4333 |
* AND the constraints of the partition and add to |
| 4334 |
* or_expr_args |
| 4335 |
*/ |
| 4336 |
or_expr_args = lappend(or_expr_args, list_length(part_qual) > 1 |
| 4337 |
? makeBoolExpr(AND_EXPR, part_qual, -1) |
| 4338 |
: linitial(part_qual)); |
| 4339 |
} |
| 4340 |
ReleaseSysCache(tuple); |
| 4341 |
} |
| 4342 |
|
| 4343 |
if (or_expr_args != NIL) |
| 4344 |
{ |
| 4345 |
Expr *other_parts_constr; |
| 4346 |
|
| 4347 |
/* |
| 4348 |
* Combine the constraints obtained for non-default partitions |
| 4349 |
* using OR. As requested, each of the OR's args doesn't include |
| 4350 |
* the NOT NULL test for partition keys (which is to avoid its |
| 4351 |
* useless repetition). Add the same now. |
| 4352 |
*/ |
| 4353 |
other_parts_constr = |
| 4354 |
makeBoolExpr(AND_EXPR, |
| 4355 |
lappend(get_range_nulltest(key), |
| 4356 |
list_length(or_expr_args) > 1 |
| 4357 |
? makeBoolExpr(OR_EXPR, or_expr_args, |
| 4358 |
-1) |
| 4359 |
: linitial(or_expr_args)), |
| 4360 |
-1); |
| 4361 |
|
| 4362 |
/* |
| 4363 |
* Finally, the default partition contains everything *NOT* |
| 4364 |
* contained in the non-default partitions. |
| 4365 |
*/ |
| 4366 |
result = list_make1(makeBoolExpr(NOT_EXPR, |
| 4367 |
list_make1(other_parts_constr), -1)); |
| 4368 |
} |
| 4369 |
|
| 4370 |
return result; |
| 4371 |
} |
| 4372 |
|
| 4373 |
/* |
| 4374 |
* If it is the recursive call for default, we skip the get_range_nulltest |
| 4375 |
* to avoid accumulating the NullTest on the same keys for each partition. |
| 4376 |
*/ |
| 4377 |
if (!for_default) |
| 4378 |
result = get_range_nulltest(key); |
| 4379 |
|
| 4380 |
/* |
| 4381 |
* Iterate over the key columns and check if the corresponding lower and |
| 4382 |
* upper datums are equal using the btree equality operator for the |
| 4383 |
* column's type. If equal, we emit single keyCol = common_value |
| 4384 |
* expression. Starting from the first column for which the corresponding |
| 4385 |
* lower and upper bound datums are not equal, we generate OR expressions |
| 4386 |
* as shown in the function's header comment. |
| 4387 |
*/ |
| 4388 |
i = 0; |
| 4389 |
partexprs_item = list_head(key->partexprs); |
| 4390 |
partexprs_item_saved = partexprs_item; /* placate compiler */ |
| 4391 |
forboth(cell1, spec->lowerdatums, cell2, spec->upperdatums) |
| 4392 |
{ |
| 4393 |
EState *estate; |
| 4394 |
MemoryContext oldcxt; |
| 4395 |
Expr *test_expr; |
| 4396 |
ExprState *test_exprstate; |
| 4397 |
Datum test_result; |
| 4398 |
bool isNull; |
| 4399 |
|
| 4400 |
ldatum = lfirst_node(PartitionRangeDatum, cell1); |
| 4401 |
udatum = lfirst_node(PartitionRangeDatum, cell2); |
| 4402 |
|
| 4403 |
/* |
| 4404 |
* Since get_range_key_properties() modifies partexprs_item, and we |
| 4405 |
* might need to start over from the previous expression in the later |
| 4406 |
* part of this function, save away the current value. |
| 4407 |
*/ |
| 4408 |
partexprs_item_saved = partexprs_item; |
| 4409 |
|
| 4410 |
get_range_key_properties(key, i, ldatum, udatum, |
| 4411 |
&partexprs_item, |
| 4412 |
&keyCol, |
| 4413 |
&lower_val, &upper_val); |
| 4414 |
|
| 4415 |
/* |
| 4416 |
* If either value is NULL, the corresponding partition bound is |
| 4417 |
* either MINVALUE or MAXVALUE, and we treat them as unequal, because |
| 4418 |
* even if they're the same, there is no common value to equate the |
| 4419 |
* key column with. |
| 4420 |
*/ |
| 4421 |
if (!lower_val || !upper_val) |
| 4422 |
break; |
| 4423 |
|
| 4424 |
/* Create the test expression */ |
| 4425 |
estate = CreateExecutorState(); |
| 4426 |
oldcxt = MemoryContextSwitchTo(estate->es_query_cxt); |
| 4427 |
test_expr = make_partition_op_expr(key, i, BTEqualStrategyNumber, |
| 4428 |
(Expr *) lower_val, |
| 4429 |
(Expr *) upper_val); |
| 4430 |
fix_opfuncids((Node *) test_expr); |
| 4431 |
test_exprstate = ExecInitExpr(test_expr, NULL); |
| 4432 |
test_result = ExecEvalExprSwitchContext(test_exprstate, |
| 4433 |
GetPerTupleExprContext(estate), |
| 4434 |
&isNull); |
| 4435 |
MemoryContextSwitchTo(oldcxt); |
| 4436 |
FreeExecutorState(estate); |
| 4437 |
|
| 4438 |
/* If not equal, go generate the OR expressions */ |
| 4439 |
if (!DatumGetBool(test_result)) |
| 4440 |
break; |
| 4441 |
|
| 4442 |
/* |
| 4443 |
* The bounds for the last key column can't be equal, because such a |
| 4444 |
* range partition would never be allowed to be defined (it would have |
| 4445 |
* an empty range otherwise). |
| 4446 |
*/ |
| 4447 |
if (i == key->partnatts - 1) |
| 4448 |
elog(ERROR, "invalid range bound specification"); |
| 4449 |
|
| 4450 |
/* Equal, so generate keyCol = lower_val expression */ |
| 4451 |
result = lappend(result, |
| 4452 |
make_partition_op_expr(key, i, BTEqualStrategyNumber, |
| 4453 |
keyCol, (Expr *) lower_val)); |
| 4454 |
|
| 4455 |
i++; |
| 4456 |
} |
| 4457 |
|
| 4458 |
/* First pair of lower_val and upper_val that are not equal. */ |
| 4459 |
lower_or_start_datum = cell1; |
| 4460 |
upper_or_start_datum = cell2; |
| 4461 |
|
| 4462 |
/* OR will have as many arms as there are key columns left. */ |
| 4463 |
num_or_arms = key->partnatts - i; |
| 4464 |
current_or_arm = 0; |
| 4465 |
lower_or_arms = upper_or_arms = NIL; |
| 4466 |
need_next_lower_arm = need_next_upper_arm = true; |
| 4467 |
while (current_or_arm < num_or_arms) |
| 4468 |
{ |
| 4469 |
List *lower_or_arm_args = NIL, |
| 4470 |
*upper_or_arm_args = NIL; |
| 4471 |
|
| 4472 |
/* Restart scan of columns from the i'th one */ |
| 4473 |
j = i; |
| 4474 |
partexprs_item = partexprs_item_saved; |
| 4475 |
|
| 4476 |
for_both_cell(cell1, spec->lowerdatums, lower_or_start_datum, |
| 4477 |
cell2, spec->upperdatums, upper_or_start_datum) |
| 4478 |
{ |
| 4479 |
PartitionRangeDatum *ldatum_next = NULL, |
| 4480 |
*udatum_next = NULL; |
| 4481 |
|
| 4482 |
ldatum = lfirst_node(PartitionRangeDatum, cell1); |
| 4483 |
if (lnext(spec->lowerdatums, cell1)) |
| 4484 |
ldatum_next = castNode(PartitionRangeDatum, |
| 4485 |
lfirst(lnext(spec->lowerdatums, cell1))); |
| 4486 |
udatum = lfirst_node(PartitionRangeDatum, cell2); |
| 4487 |
if (lnext(spec->upperdatums, cell2)) |
| 4488 |
udatum_next = castNode(PartitionRangeDatum, |
| 4489 |
lfirst(lnext(spec->upperdatums, cell2))); |
| 4490 |
get_range_key_properties(key, j, ldatum, udatum, |
| 4491 |
&partexprs_item, |
| 4492 |
&keyCol, |
| 4493 |
&lower_val, &upper_val); |
| 4494 |
|
| 4495 |
if (need_next_lower_arm && lower_val) |
| 4496 |
{ |
| 4497 |
uint16 strategy; |
| 4498 |
|
| 4499 |
/* |
| 4500 |
* For the non-last columns of this arm, use the EQ operator. |
| 4501 |
* For the last column of this arm, use GT, unless this is the |
| 4502 |
* last column of the whole bound check, or the next bound |
| 4503 |
* datum is MINVALUE, in which case use GE. |
| 4504 |
*/ |
| 4505 |
if (j - i < current_or_arm) |
| 4506 |
strategy = BTEqualStrategyNumber; |
| 4507 |
else if (j == key->partnatts - 1 || |
| 4508 |
(ldatum_next && |
| 4509 |
ldatum_next->kind == PARTITION_RANGE_DATUM_MINVALUE)) |
| 4510 |
strategy = BTGreaterEqualStrategyNumber; |
| 4511 |
else |
| 4512 |
strategy = BTGreaterStrategyNumber; |
| 4513 |
|
| 4514 |
lower_or_arm_args = lappend(lower_or_arm_args, |
| 4515 |
make_partition_op_expr(key, j, |
| 4516 |
strategy, |
| 4517 |
keyCol, |
| 4518 |
(Expr *) lower_val)); |
| 4519 |
} |
| 4520 |
|
| 4521 |
if (need_next_upper_arm && upper_val) |
| 4522 |
{ |
| 4523 |
uint16 strategy; |
| 4524 |
|
| 4525 |
/* |
| 4526 |
* For the non-last columns of this arm, use the EQ operator. |
| 4527 |
* For the last column of this arm, use LT, unless the next |
| 4528 |
* bound datum is MAXVALUE, in which case use LE. |
| 4529 |
*/ |
| 4530 |
if (j - i < current_or_arm) |
| 4531 |
strategy = BTEqualStrategyNumber; |
| 4532 |
else if (udatum_next && |
| 4533 |
udatum_next->kind == PARTITION_RANGE_DATUM_MAXVALUE) |
| 4534 |
strategy = BTLessEqualStrategyNumber; |
| 4535 |
else |
| 4536 |
strategy = BTLessStrategyNumber; |
| 4537 |
|
| 4538 |
upper_or_arm_args = lappend(upper_or_arm_args, |
| 4539 |
make_partition_op_expr(key, j, |
| 4540 |
strategy, |
| 4541 |
keyCol, |
| 4542 |
(Expr *) upper_val)); |
| 4543 |
} |
| 4544 |
|
| 4545 |
/* |
| 4546 |
* Did we generate enough of OR's arguments? First arm considers |
| 4547 |
* the first of the remaining columns, second arm considers first |
| 4548 |
* two of the remaining columns, and so on. |
| 4549 |
*/ |
| 4550 |
++j; |
| 4551 |
if (j - i > current_or_arm) |
| 4552 |
{ |
| 4553 |
/* |
| 4554 |
* We must not emit any more arms if the new column that will |
| 4555 |
* be considered is unbounded, or this one was. |
| 4556 |
*/ |
| 4557 |
if (!lower_val || !ldatum_next || |
| 4558 |
ldatum_next->kind != PARTITION_RANGE_DATUM_VALUE) |
| 4559 |
need_next_lower_arm = false; |
| 4560 |
if (!upper_val || !udatum_next || |
| 4561 |
udatum_next->kind != PARTITION_RANGE_DATUM_VALUE) |
| 4562 |
need_next_upper_arm = false; |
| 4563 |
break; |
| 4564 |
} |
| 4565 |
} |
| 4566 |
|
| 4567 |
if (lower_or_arm_args != NIL) |
| 4568 |
lower_or_arms = lappend(lower_or_arms, |
| 4569 |
list_length(lower_or_arm_args) > 1 |
| 4570 |
? makeBoolExpr(AND_EXPR, lower_or_arm_args, -1) |
| 4571 |
: linitial(lower_or_arm_args)); |
| 4572 |
|
| 4573 |
if (upper_or_arm_args != NIL) |
| 4574 |
upper_or_arms = lappend(upper_or_arms, |
| 4575 |
list_length(upper_or_arm_args) > 1 |
| 4576 |
? makeBoolExpr(AND_EXPR, upper_or_arm_args, -1) |
| 4577 |
: linitial(upper_or_arm_args)); |
| 4578 |
|
| 4579 |
/* If no work to do in the next iteration, break away. */ |
| 4580 |
if (!need_next_lower_arm && !need_next_upper_arm) |
| 4581 |
break; |
| 4582 |
|
| 4583 |
++current_or_arm; |
| 4584 |
} |
| 4585 |
|
| 4586 |
/* |
| 4587 |
* Generate the OR expressions for each of lower and upper bounds (if |
| 4588 |
* required), and append to the list of implicitly ANDed list of |
| 4589 |
* expressions. |
| 4590 |
*/ |
| 4591 |
if (lower_or_arms != NIL) |
| 4592 |
result = lappend(result, |
| 4593 |
list_length(lower_or_arms) > 1 |
| 4594 |
? makeBoolExpr(OR_EXPR, lower_or_arms, -1) |
| 4595 |
: linitial(lower_or_arms)); |
| 4596 |
if (upper_or_arms != NIL) |
| 4597 |
result = lappend(result, |
| 4598 |
list_length(upper_or_arms) > 1 |
| 4599 |
? makeBoolExpr(OR_EXPR, upper_or_arms, -1) |
| 4600 |
: linitial(upper_or_arms)); |
| 4601 |
|
| 4602 |
/* |
| 4603 |
* As noted above, for non-default, we return list with constant TRUE. If |
| 4604 |
* the result is NIL during the recursive call for default, it implies |
| 4605 |
* this is the only other partition which can hold every value of the key |
| 4606 |
* except NULL. Hence we return the NullTest result skipped earlier. |
| 4607 |
*/ |
| 4608 |
if (result == NIL) |
| 4609 |
result = for_default |
| 4610 |
? get_range_nulltest(key) |
| 4611 |
: list_make1(makeBoolConst(true, false)); |
| 4612 |
|
| 4613 |
return result; |
| 4614 |
} |
| 4615 |
|
| 4616 |
/* |
| 4617 |
* get_range_key_properties |
| 4618 |
* Returns range partition key information for a given column |
| 4619 |
* |
| 4620 |
* This is a subroutine for get_qual_for_range, and its API is pretty |
| 4621 |
* specialized to that caller. |
| 4622 |
* |
| 4623 |
* Constructs an Expr for the key column (returned in *keyCol) and Consts |
| 4624 |
* for the lower and upper range limits (returned in *lower_val and |
| 4625 |
* *upper_val). For MINVALUE/MAXVALUE limits, NULL is returned instead of |
| 4626 |
* a Const. All of these structures are freshly palloc'd. |
| 4627 |
* |
| 4628 |
* *partexprs_item points to the cell containing the next expression in |
| 4629 |
* the key->partexprs list, or NULL. It may be advanced upon return. |
| 4630 |
*/ |
| 4631 |
static void |
| 4632 |
get_range_key_properties(PartitionKey key, int keynum, |
| 4633 |
PartitionRangeDatum *ldatum, |
| 4634 |
PartitionRangeDatum *udatum, |
| 4635 |
ListCell **partexprs_item, |
| 4636 |
Expr **keyCol, |
| 4637 |
Const **lower_val, Const **upper_val) |
| 4638 |
{ |
| 4639 |
/* Get partition key expression for this column */ |
| 4640 |
if (key->partattrs[keynum] != 0) |
| 4641 |
{ |
| 4642 |
*keyCol = (Expr *) makeVar(1, |
| 4643 |
key->partattrs[keynum], |
| 4644 |
key->parttypid[keynum], |
| 4645 |
key->parttypmod[keynum], |
| 4646 |
key->parttypcoll[keynum], |
| 4647 |
0); |
| 4648 |
} |
| 4649 |
else |
| 4650 |
{ |
| 4651 |
if (*partexprs_item == NULL) |
| 4652 |
elog(ERROR, "wrong number of partition key expressions"); |
| 4653 |
*keyCol = copyObject(lfirst(*partexprs_item)); |
| 4654 |
*partexprs_item = lnext(key->partexprs, *partexprs_item); |
| 4655 |
} |
| 4656 |
|
| 4657 |
/* Get appropriate Const nodes for the bounds */ |
| 4658 |
if (ldatum->kind == PARTITION_RANGE_DATUM_VALUE) |
| 4659 |
*lower_val = castNode(Const, copyObject(ldatum->value)); |
| 4660 |
else |
| 4661 |
*lower_val = NULL; |
| 4662 |
|
| 4663 |
if (udatum->kind == PARTITION_RANGE_DATUM_VALUE) |
| 4664 |
*upper_val = castNode(Const, copyObject(udatum->value)); |
| 4665 |
else |
| 4666 |
*upper_val = NULL; |
| 4667 |
} |
| 4668 |
|
| 4669 |
/* |
| 4670 |
* get_range_nulltest |
| 4671 |
* |
| 4672 |
* A non-default range partition table does not currently allow partition |
| 4673 |
* keys to be null, so emit an IS NOT NULL expression for each key column. |
| 4674 |
*/ |
| 4675 |
static List * |
| 4676 |
get_range_nulltest(PartitionKey key) |
| 4677 |
{ |
| 4678 |
List *result = NIL; |
| 4679 |
NullTest *nulltest; |
| 4680 |
ListCell *partexprs_item; |
| 4681 |
int i; |
| 4682 |
|
| 4683 |
partexprs_item = list_head(key->partexprs); |
| 4684 |
for (i = 0; i < key->partnatts; i++) |
| 4685 |
{ |
| 4686 |
Expr *keyCol; |
| 4687 |
|
| 4688 |
if (key->partattrs[i] != 0) |
| 4689 |
{ |
| 4690 |
keyCol = (Expr *) makeVar(1, |
| 4691 |
key->partattrs[i], |
| 4692 |
key->parttypid[i], |
| 4693 |
key->parttypmod[i], |
| 4694 |
key->parttypcoll[i], |
| 4695 |
0); |
| 4696 |
} |
| 4697 |
else |
| 4698 |
{ |
| 4699 |
if (partexprs_item == NULL) |
| 4700 |
elog(ERROR, "wrong number of partition key expressions"); |
| 4701 |
keyCol = copyObject(lfirst(partexprs_item)); |
| 4702 |
partexprs_item = lnext(key->partexprs, partexprs_item); |
| 4703 |
} |
| 4704 |
|
| 4705 |
nulltest = makeNode(NullTest); |
| 4706 |
nulltest->arg = keyCol; |
| 4707 |
nulltest->nulltesttype = IS_NOT_NULL; |
| 4708 |
nulltest->argisrow = false; |
| 4709 |
nulltest->location = -1; |
| 4710 |
result = lappend(result, nulltest); |
| 4711 |
} |
| 4712 |
|
| 4713 |
return result; |
| 4714 |
} |
| 4715 |
|
| 4716 |
/* |
| 4717 |
* compute_partition_hash_value |
| 4718 |
* |
| 4719 |
* Compute the hash value for given partition key values. |
| 4720 |
*/ |
| 4721 |
uint64 |
| 4722 |
compute_partition_hash_value(int partnatts, FmgrInfo *partsupfunc, const Oid *partcollation, |
| 4723 |
const Datum *values, const bool *isnull) |
| 4724 |
{ |
| 4725 |
int i; |
| 4726 |
uint64 rowHash = 0; |
| 4727 |
Datum seed = UInt64GetDatum(HASH_PARTITION_SEED); |
| 4728 |
|
| 4729 |
for (i = 0; i < partnatts; i++) |
| 4730 |
{ |
| 4731 |
/* Nulls are just ignored */ |
| 4732 |
if (!isnull[i]) |
| 4733 |
{ |
| 4734 |
Datum hash; |
| 4735 |
|
| 4736 |
Assert(OidIsValid(partsupfunc[i].fn_oid)); |
| 4737 |
|
| 4738 |
/* |
| 4739 |
* Compute hash for each datum value by calling respective |
| 4740 |
* datatype-specific hash functions of each partition key |
| 4741 |
* attribute. |
| 4742 |
*/ |
| 4743 |
hash = FunctionCall2Coll(&partsupfunc[i], partcollation[i], |
| 4744 |
values[i], seed); |
| 4745 |
|
| 4746 |
/* Form a single 64-bit hash value */ |
| 4747 |
rowHash = hash_combine64(rowHash, DatumGetUInt64(hash)); |
| 4748 |
} |
| 4749 |
} |
| 4750 |
|
| 4751 |
return rowHash; |
| 4752 |
} |
| 4753 |
|
| 4754 |
/* |
| 4755 |
* satisfies_hash_partition |
| 4756 |
* |
| 4757 |
* This is an SQL-callable function for use in hash partition constraints. |
| 4758 |
* The first three arguments are the parent table OID, modulus, and remainder. |
| 4759 |
* The remaining arguments are the value of the partitioning columns (or |
| 4760 |
* expressions); these are hashed and the results are combined into a single |
| 4761 |
* hash value by calling hash_combine64. |
| 4762 |
* |
| 4763 |
* Returns true if remainder produced when this computed single hash value is |
| 4764 |
* divided by the given modulus is equal to given remainder, otherwise false. |
| 4765 |
* NB: it's important that this never return null, as the constraint machinery |
| 4766 |
* would consider that to be a "pass". |
| 4767 |
* |
| 4768 |
* See get_qual_for_hash() for usage. |
| 4769 |
*/ |
| 4770 |
Datum |
| 4771 |
satisfies_hash_partition(PG_FUNCTION_ARGS) |
| 4772 |
{ |
| 4773 |
typedef struct ColumnsHashData |
| 4774 |
{ |
| 4775 |
Oid relid; |
| 4776 |
int nkeys; |
| 4777 |
Oid variadic_type; |
| 4778 |
int16 variadic_typlen; |
| 4779 |
bool variadic_typbyval; |
| 4780 |
char variadic_typalign; |
| 4781 |
Oid partcollid[PARTITION_MAX_KEYS]; |
| 4782 |
FmgrInfo partsupfunc[FLEXIBLE_ARRAY_MEMBER]; |
| 4783 |
} ColumnsHashData; |
| 4784 |
Oid parentId; |
| 4785 |
int modulus; |
| 4786 |
int remainder; |
| 4787 |
Datum seed = UInt64GetDatum(HASH_PARTITION_SEED); |
| 4788 |
ColumnsHashData *my_extra; |
| 4789 |
uint64 rowHash = 0; |
| 4790 |
|
| 4791 |
/* Return false if the parent OID, modulus, or remainder is NULL. */ |
| 4792 |
if (PG_ARGISNULL(0) || PG_ARGISNULL(1) || PG_ARGISNULL(2)) |
| 4793 |
PG_RETURN_BOOL(false); |
| 4794 |
parentId = PG_GETARG_OID(0); |
| 4795 |
modulus = PG_GETARG_INT32(1); |
| 4796 |
remainder = PG_GETARG_INT32(2); |
| 4797 |
|
| 4798 |
/* Sanity check modulus and remainder. */ |
| 4799 |
if (modulus <= 0) |
| 4800 |
ereport(ERROR, |
| 4801 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4802 |
errmsg("modulus for hash partition must be an integer value greater than zero"))); |
| 4803 |
if (remainder < 0) |
| 4804 |
ereport(ERROR, |
| 4805 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4806 |
errmsg("remainder for hash partition must be an integer value greater than or equal to zero"))); |
| 4807 |
if (remainder >= modulus) |
| 4808 |
ereport(ERROR, |
| 4809 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4810 |
errmsg("remainder for hash partition must be less than modulus"))); |
| 4811 |
|
| 4812 |
/* |
| 4813 |
* Cache hash function information. |
| 4814 |
*/ |
| 4815 |
my_extra = (ColumnsHashData *) fcinfo->flinfo->fn_extra; |
| 4816 |
if (my_extra == NULL || my_extra->relid != parentId) |
| 4817 |
{ |
| 4818 |
Relation parent; |
| 4819 |
PartitionKey key; |
| 4820 |
int j; |
| 4821 |
|
| 4822 |
/* Open parent relation and fetch partition key info */ |
| 4823 |
parent = relation_open(parentId, AccessShareLock); |
| 4824 |
key = RelationGetPartitionKey(parent); |
| 4825 |
|
| 4826 |
/* Reject parent table that is not hash-partitioned. */ |
| 4827 |
if (key == NULL || key->strategy != PARTITION_STRATEGY_HASH) |
| 4828 |
ereport(ERROR, |
| 4829 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4830 |
errmsg("\"%s\" is not a hash partitioned table", |
| 4831 |
get_rel_name(parentId)))); |
| 4832 |
|
| 4833 |
if (!get_fn_expr_variadic(fcinfo->flinfo)) |
| 4834 |
{ |
| 4835 |
int nargs = PG_NARGS() - 3; |
| 4836 |
|
| 4837 |
/* complain if wrong number of column values */ |
| 4838 |
if (key->partnatts != nargs) |
| 4839 |
ereport(ERROR, |
| 4840 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4841 |
errmsg("number of partitioning columns (%d) does not match number of partition keys provided (%d)", |
| 4842 |
key->partnatts, nargs))); |
| 4843 |
|
| 4844 |
/* allocate space for our cache */ |
| 4845 |
fcinfo->flinfo->fn_extra = |
| 4846 |
MemoryContextAllocZero(fcinfo->flinfo->fn_mcxt, |
| 4847 |
offsetof(ColumnsHashData, partsupfunc) + |
| 4848 |
sizeof(FmgrInfo) * nargs); |
| 4849 |
my_extra = (ColumnsHashData *) fcinfo->flinfo->fn_extra; |
| 4850 |
my_extra->relid = parentId; |
| 4851 |
my_extra->nkeys = key->partnatts; |
| 4852 |
memcpy(my_extra->partcollid, key->partcollation, |
| 4853 |
key->partnatts * sizeof(Oid)); |
| 4854 |
|
| 4855 |
/* check argument types and save fmgr_infos */ |
| 4856 |
for (j = 0; j < key->partnatts; ++j) |
| 4857 |
{ |
| 4858 |
Oid argtype = get_fn_expr_argtype(fcinfo->flinfo, j + 3); |
| 4859 |
|
| 4860 |
if (argtype != key->parttypid[j] && !IsBinaryCoercible(argtype, key->parttypid[j])) |
| 4861 |
ereport(ERROR, |
| 4862 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4863 |
errmsg("column %d of the partition key has type %s, but supplied value is of type %s", |
| 4864 |
j + 1, format_type_be(key->parttypid[j]), format_type_be(argtype)))); |
| 4865 |
|
| 4866 |
fmgr_info_copy(&my_extra->partsupfunc[j], |
| 4867 |
&key->partsupfunc[j], |
| 4868 |
fcinfo->flinfo->fn_mcxt); |
| 4869 |
} |
| 4870 |
} |
| 4871 |
else if (PG_ARGISNULL(3)) |
| 4872 |
{ |
| 4873 |
/* Special case for VARIADIC NULL::sometype[] */ |
| 4874 |
relation_close(parent, NoLock); |
| 4875 |
PG_RETURN_BOOL(false); |
| 4876 |
} |
| 4877 |
else |
| 4878 |
{ |
| 4879 |
ArrayType *variadic_array = PG_GETARG_ARRAYTYPE_P(3); |
| 4880 |
|
| 4881 |
/* allocate space for our cache -- just one FmgrInfo in this case */ |
| 4882 |
fcinfo->flinfo->fn_extra = |
| 4883 |
MemoryContextAllocZero(fcinfo->flinfo->fn_mcxt, |
| 4884 |
offsetof(ColumnsHashData, partsupfunc) + |
| 4885 |
sizeof(FmgrInfo)); |
| 4886 |
my_extra = (ColumnsHashData *) fcinfo->flinfo->fn_extra; |
| 4887 |
my_extra->relid = parentId; |
| 4888 |
my_extra->nkeys = key->partnatts; |
| 4889 |
my_extra->variadic_type = ARR_ELEMTYPE(variadic_array); |
| 4890 |
get_typlenbyvalalign(my_extra->variadic_type, |
| 4891 |
&my_extra->variadic_typlen, |
| 4892 |
&my_extra->variadic_typbyval, |
| 4893 |
&my_extra->variadic_typalign); |
| 4894 |
my_extra->partcollid[0] = key->partcollation[0]; |
| 4895 |
|
| 4896 |
/* check argument types */ |
| 4897 |
for (j = 0; j < key->partnatts; ++j) |
| 4898 |
if (key->parttypid[j] != my_extra->variadic_type) |
| 4899 |
ereport(ERROR, |
| 4900 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4901 |
errmsg("column %d of the partition key has type \"%s\", but supplied value is of type \"%s\"", |
| 4902 |
j + 1, |
| 4903 |
format_type_be(key->parttypid[j]), |
| 4904 |
format_type_be(my_extra->variadic_type)))); |
| 4905 |
|
| 4906 |
fmgr_info_copy(&my_extra->partsupfunc[0], |
| 4907 |
&key->partsupfunc[0], |
| 4908 |
fcinfo->flinfo->fn_mcxt); |
| 4909 |
} |
| 4910 |
|
| 4911 |
/* Hold lock until commit */ |
| 4912 |
relation_close(parent, NoLock); |
| 4913 |
} |
| 4914 |
|
| 4915 |
if (!OidIsValid(my_extra->variadic_type)) |
| 4916 |
{ |
| 4917 |
int nkeys = my_extra->nkeys; |
| 4918 |
int i; |
| 4919 |
|
| 4920 |
/* |
| 4921 |
* For a non-variadic call, neither the number of arguments nor their |
| 4922 |
* types can change across calls, so avoid the expense of rechecking |
| 4923 |
* here. |
| 4924 |
*/ |
| 4925 |
|
| 4926 |
for (i = 0; i < nkeys; i++) |
| 4927 |
{ |
| 4928 |
Datum hash; |
| 4929 |
|
| 4930 |
/* keys start from fourth argument of function. */ |
| 4931 |
int argno = i + 3; |
| 4932 |
|
| 4933 |
if (PG_ARGISNULL(argno)) |
| 4934 |
continue; |
| 4935 |
|
| 4936 |
hash = FunctionCall2Coll(&my_extra->partsupfunc[i], |
| 4937 |
my_extra->partcollid[i], |
| 4938 |
PG_GETARG_DATUM(argno), |
| 4939 |
seed); |
| 4940 |
|
| 4941 |
/* Form a single 64-bit hash value */ |
| 4942 |
rowHash = hash_combine64(rowHash, DatumGetUInt64(hash)); |
| 4943 |
} |
| 4944 |
} |
| 4945 |
else |
| 4946 |
{ |
| 4947 |
ArrayType *variadic_array; |
| 4948 |
int i; |
| 4949 |
int nelems; |
| 4950 |
Datum *datum; |
| 4951 |
bool *isnull; |
| 4952 |
|
| 4953 |
/* Special case for VARIADIC NULL::sometype[] */ |
| 4954 |
if (PG_ARGISNULL(3)) |
| 4955 |
PG_RETURN_BOOL(false); |
| 4956 |
|
| 4957 |
variadic_array = PG_GETARG_ARRAYTYPE_P(3); |
| 4958 |
|
| 4959 |
deconstruct_array(variadic_array, |
| 4960 |
my_extra->variadic_type, |
| 4961 |
my_extra->variadic_typlen, |
| 4962 |
my_extra->variadic_typbyval, |
| 4963 |
my_extra->variadic_typalign, |
| 4964 |
&datum, &isnull, &nelems); |
| 4965 |
|
| 4966 |
/* complain if wrong number of column values */ |
| 4967 |
if (nelems != my_extra->nkeys) |
| 4968 |
ereport(ERROR, |
| 4969 |
(errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
| 4970 |
errmsg("number of partitioning columns (%d) does not match number of partition keys provided (%d)", |
| 4971 |
my_extra->nkeys, nelems))); |
| 4972 |
|
| 4973 |
for (i = 0; i < nelems; i++) |
| 4974 |
{ |
| 4975 |
Datum hash; |
| 4976 |
|
| 4977 |
if (isnull[i]) |
| 4978 |
continue; |
| 4979 |
|
| 4980 |
hash = FunctionCall2Coll(&my_extra->partsupfunc[0], |
| 4981 |
my_extra->partcollid[0], |
| 4982 |
datum[i], |
| 4983 |
seed); |
| 4984 |
|
| 4985 |
/* Form a single 64-bit hash value */ |
| 4986 |
rowHash = hash_combine64(rowHash, DatumGetUInt64(hash)); |
| 4987 |
} |
| 4988 |
} |
| 4989 |
|
| 4990 |
PG_RETURN_BOOL(rowHash % modulus == remainder); |
| 4991 |
} |