| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* cluster.c |
| 4 |
* CLUSTER a table on an index. This is now also used for VACUUM FULL. |
| 5 |
* |
| 6 |
* There is hardly anything left of Paul Brown's original implementation... |
| 7 |
* |
| 8 |
* |
| 9 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 10 |
* Portions Copyright (c) 1994-5, Regents of the University of California |
| 11 |
* |
| 12 |
* |
| 13 |
* IDENTIFICATION |
| 14 |
* src/backend/commands/cluster.c |
| 15 |
* |
| 16 |
*------------------------------------------------------------------------- |
| 17 |
*/ |
| 18 |
#include "postgres.h" |
| 19 |
|
| 20 |
#include "access/amapi.h" |
| 21 |
#include "access/heapam.h" |
| 22 |
#include "access/multixact.h" |
| 23 |
#include "access/relscan.h" |
| 24 |
#include "access/tableam.h" |
| 25 |
#include "access/toast_internals.h" |
| 26 |
#include "access/transam.h" |
| 27 |
#include "access/xact.h" |
| 28 |
#include "catalog/catalog.h" |
| 29 |
#include "catalog/dependency.h" |
| 30 |
#include "catalog/heap.h" |
| 31 |
#include "catalog/index.h" |
| 32 |
#include "catalog/namespace.h" |
| 33 |
#include "catalog/objectaccess.h" |
| 34 |
#include "catalog/pg_am.h" |
| 35 |
#include "catalog/pg_inherits.h" |
| 36 |
#include "catalog/toasting.h" |
| 37 |
#include "commands/cluster.h" |
| 38 |
#include "commands/defrem.h" |
| 39 |
#include "commands/progress.h" |
| 40 |
#include "commands/tablecmds.h" |
| 41 |
#include "commands/vacuum.h" |
| 42 |
#include "miscadmin.h" |
| 43 |
#include "optimizer/optimizer.h" |
| 44 |
#include "pgstat.h" |
| 45 |
#include "storage/bufmgr.h" |
| 46 |
#include "storage/lmgr.h" |
| 47 |
#include "storage/predicate.h" |
| 48 |
#include "utils/acl.h" |
| 49 |
#include "utils/fmgroids.h" |
| 50 |
#include "utils/guc.h" |
| 51 |
#include "utils/inval.h" |
| 52 |
#include "utils/lsyscache.h" |
| 53 |
#include "utils/memutils.h" |
| 54 |
#include "utils/pg_rusage.h" |
| 55 |
#include "utils/relmapper.h" |
| 56 |
#include "utils/snapmgr.h" |
| 57 |
#include "utils/syscache.h" |
| 58 |
|
| 59 |
/* |
| 60 |
* This struct is used to pass around the information on tables to be |
| 61 |
* clustered. We need this so we can make a list of them when invoked without |
| 62 |
* a specific table/index pair. |
| 63 |
*/ |
| 64 |
typedef struct |
| 65 |
{ |
| 66 |
Oid tableOid; |
| 67 |
Oid indexOid; |
| 68 |
} RelToCluster; |
| 69 |
|
| 70 |
|
| 71 |
static void cluster_multiple_rels(List *rtcs, ClusterParams *params); |
| 72 |
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose); |
| 73 |
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex, |
| 74 |
bool verbose, bool *pSwapToastByContent, |
| 75 |
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti); |
| 76 |
static List *get_tables_to_cluster(MemoryContext cluster_context); |
| 77 |
static List *get_tables_to_cluster_partitioned(MemoryContext cluster_context, |
| 78 |
Oid indexOid); |
| 79 |
static bool cluster_is_permitted_for_relation(Oid relid, Oid userid); |
| 80 |
|
| 81 |
|
| 82 |
/*--------------------------------------------------------------------------- |
| 83 |
* This cluster code allows for clustering multiple tables at once. Because |
| 84 |
* of this, we cannot just run everything on a single transaction, or we |
| 85 |
* would be forced to acquire exclusive locks on all the tables being |
| 86 |
* clustered, simultaneously --- very likely leading to deadlock. |
| 87 |
* |
| 88 |
* To solve this we follow a similar strategy to VACUUM code, |
| 89 |
* clustering each relation in a separate transaction. For this to work, |
| 90 |
* we need to: |
| 91 |
* - provide a separate memory context so that we can pass information in |
| 92 |
* a way that survives across transactions |
| 93 |
* - start a new transaction every time a new relation is clustered |
| 94 |
* - check for validity of the information on to-be-clustered relations, |
| 95 |
* as someone might have deleted a relation behind our back, or |
| 96 |
* clustered one on a different index |
| 97 |
* - end the transaction |
| 98 |
* |
| 99 |
* The single-relation case does not have any such overhead. |
| 100 |
* |
| 101 |
* We also allow a relation to be specified without index. In that case, |
| 102 |
* the indisclustered bit will be looked up, and an ERROR will be thrown |
| 103 |
* if there is no index with the bit set. |
| 104 |
*--------------------------------------------------------------------------- |
| 105 |
*/ |
| 106 |
void |
| 107 |
cluster(ParseState *pstate, ClusterStmt *stmt, bool isTopLevel) |
| 108 |
{ |
| 109 |
ListCell *lc; |
| 110 |
ClusterParams params = {0}; |
| 111 |
bool verbose = false; |
| 112 |
Relation rel = NULL; |
| 113 |
Oid indexOid = InvalidOid; |
| 114 |
MemoryContext cluster_context; |
| 115 |
List *rtcs; |
| 116 |
|
| 117 |
/* Parse option list */ |
| 118 |
foreach(lc, stmt->params) |
| 119 |
{ |
| 120 |
DefElem *opt = (DefElem *) lfirst(lc); |
| 121 |
|
| 122 |
if (strcmp(opt->defname, "verbose") == 0) |
| 123 |
verbose = defGetBoolean(opt); |
| 124 |
else |
| 125 |
ereport(ERROR, |
| 126 |
(errcode(ERRCODE_SYNTAX_ERROR), |
| 127 |
errmsg("unrecognized %s option \"%s\"", |
| 128 |
"CLUSTER", opt->defname), |
| 129 |
parser_errposition(pstate, opt->location))); |
| 130 |
} |
| 131 |
|
| 132 |
params.options = (verbose ? CLUOPT_VERBOSE : 0); |
| 133 |
|
| 134 |
if (stmt->relation != NULL) |
| 135 |
{ |
| 136 |
/* This is the single-relation case. */ |
| 137 |
Oid tableOid; |
| 138 |
|
| 139 |
/* |
| 140 |
* Find, lock, and check permissions on the table. We obtain |
| 141 |
* AccessExclusiveLock right away to avoid lock-upgrade hazard in the |
| 142 |
* single-transaction case. |
| 143 |
*/ |
| 144 |
tableOid = RangeVarGetRelidExtended(stmt->relation, |
| 145 |
AccessExclusiveLock, |
| 146 |
0, |
| 147 |
RangeVarCallbackMaintainsTable, |
| 148 |
NULL); |
| 149 |
rel = table_open(tableOid, NoLock); |
| 150 |
|
| 151 |
/* |
| 152 |
* Reject clustering a remote temp table ... their local buffer |
| 153 |
* manager is not going to cope. |
| 154 |
*/ |
| 155 |
if (RELATION_IS_OTHER_TEMP(rel)) |
| 156 |
ereport(ERROR, |
| 157 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 158 |
errmsg("cannot cluster temporary tables of other sessions"))); |
| 159 |
|
| 160 |
if (stmt->indexname == NULL) |
| 161 |
{ |
| 162 |
ListCell *index; |
| 163 |
|
| 164 |
/* We need to find the index that has indisclustered set. */ |
| 165 |
foreach(index, RelationGetIndexList(rel)) |
| 166 |
{ |
| 167 |
indexOid = lfirst_oid(index); |
| 168 |
if (get_index_isclustered(indexOid)) |
| 169 |
break; |
| 170 |
indexOid = InvalidOid; |
| 171 |
} |
| 172 |
|
| 173 |
if (!OidIsValid(indexOid)) |
| 174 |
ereport(ERROR, |
| 175 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 176 |
errmsg("there is no previously clustered index for table \"%s\"", |
| 177 |
stmt->relation->relname))); |
| 178 |
} |
| 179 |
else |
| 180 |
{ |
| 181 |
/* |
| 182 |
* The index is expected to be in the same namespace as the |
| 183 |
* relation. |
| 184 |
*/ |
| 185 |
indexOid = get_relname_relid(stmt->indexname, |
| 186 |
rel->rd_rel->relnamespace); |
| 187 |
if (!OidIsValid(indexOid)) |
| 188 |
ereport(ERROR, |
| 189 |
(errcode(ERRCODE_UNDEFINED_OBJECT), |
| 190 |
errmsg("index \"%s\" for table \"%s\" does not exist", |
| 191 |
stmt->indexname, stmt->relation->relname))); |
| 192 |
} |
| 193 |
|
| 194 |
/* For non-partitioned tables, do what we came here to do. */ |
| 195 |
if (rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE) |
| 196 |
{ |
| 197 |
cluster_rel(rel, indexOid, ¶ms); |
| 198 |
/* cluster_rel closes the relation, but keeps lock */ |
| 199 |
|
| 200 |
return; |
| 201 |
} |
| 202 |
} |
| 203 |
|
| 204 |
/* |
| 205 |
* By here, we know we are in a multi-table situation. In order to avoid |
| 206 |
* holding locks for too long, we want to process each table in its own |
| 207 |
* transaction. This forces us to disallow running inside a user |
| 208 |
* transaction block. |
| 209 |
*/ |
| 210 |
PreventInTransactionBlock(isTopLevel, "CLUSTER"); |
| 211 |
|
| 212 |
/* Also, we need a memory context to hold our list of relations */ |
| 213 |
cluster_context = AllocSetContextCreate(PortalContext, |
| 214 |
"Cluster", |
| 215 |
ALLOCSET_DEFAULT_SIZES); |
| 216 |
|
| 217 |
/* |
| 218 |
* Either we're processing a partitioned table, or we were not given any |
| 219 |
* table name at all. In either case, obtain a list of relations to |
| 220 |
* process. |
| 221 |
* |
| 222 |
* In the former case, an index name must have been given, so we don't |
| 223 |
* need to recheck its "indisclustered" bit, but we have to check that it |
| 224 |
* is an index that we can cluster on. In the latter case, we set the |
| 225 |
* option bit to have indisclustered verified. |
| 226 |
* |
| 227 |
* Rechecking the relation itself is necessary here in all cases. |
| 228 |
*/ |
| 229 |
params.options |= CLUOPT_RECHECK; |
| 230 |
if (rel != NULL) |
| 231 |
{ |
| 232 |
Assert(rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE); |
| 233 |
check_index_is_clusterable(rel, indexOid, AccessShareLock); |
| 234 |
rtcs = get_tables_to_cluster_partitioned(cluster_context, indexOid); |
| 235 |
|
| 236 |
/* close relation, releasing lock on parent table */ |
| 237 |
table_close(rel, AccessExclusiveLock); |
| 238 |
} |
| 239 |
else |
| 240 |
{ |
| 241 |
rtcs = get_tables_to_cluster(cluster_context); |
| 242 |
params.options |= CLUOPT_RECHECK_ISCLUSTERED; |
| 243 |
} |
| 244 |
|
| 245 |
/* Do the job. */ |
| 246 |
cluster_multiple_rels(rtcs, ¶ms); |
| 247 |
|
| 248 |
/* Start a new transaction for the cleanup work. */ |
| 249 |
StartTransactionCommand(); |
| 250 |
|
| 251 |
/* Clean up working storage */ |
| 252 |
MemoryContextDelete(cluster_context); |
| 253 |
} |
| 254 |
|
| 255 |
/* |
| 256 |
* Given a list of relations to cluster, process each of them in a separate |
| 257 |
* transaction. |
| 258 |
* |
| 259 |
* We expect to be in a transaction at start, but there isn't one when we |
| 260 |
* return. |
| 261 |
*/ |
| 262 |
static void |
| 263 |
cluster_multiple_rels(List *rtcs, ClusterParams *params) |
| 264 |
{ |
| 265 |
ListCell *lc; |
| 266 |
|
| 267 |
/* Commit to get out of starting transaction */ |
| 268 |
PopActiveSnapshot(); |
| 269 |
CommitTransactionCommand(); |
| 270 |
|
| 271 |
/* Cluster the tables, each in a separate transaction */ |
| 272 |
foreach(lc, rtcs) |
| 273 |
{ |
| 274 |
RelToCluster *rtc = (RelToCluster *) lfirst(lc); |
| 275 |
Relation rel; |
| 276 |
|
| 277 |
/* Start a new transaction for each relation. */ |
| 278 |
StartTransactionCommand(); |
| 279 |
|
| 280 |
/* functions in indexes may want a snapshot set */ |
| 281 |
PushActiveSnapshot(GetTransactionSnapshot()); |
| 282 |
|
| 283 |
rel = table_open(rtc->tableOid, AccessExclusiveLock); |
| 284 |
|
| 285 |
/* Process this table */ |
| 286 |
cluster_rel(rel, rtc->indexOid, params); |
| 287 |
/* cluster_rel closes the relation, but keeps lock */ |
| 288 |
|
| 289 |
PopActiveSnapshot(); |
| 290 |
CommitTransactionCommand(); |
| 291 |
} |
| 292 |
} |
| 293 |
|
| 294 |
/* |
| 295 |
* cluster_rel |
| 296 |
* |
| 297 |
* This clusters the table by creating a new, clustered table and |
| 298 |
* swapping the relfilenumbers of the new table and the old table, so |
| 299 |
* the OID of the original table is preserved. Thus we do not lose |
| 300 |
* GRANT, inheritance nor references to this table. |
| 301 |
* |
| 302 |
* Indexes are rebuilt too, via REINDEX. Since we are effectively bulk-loading |
| 303 |
* the new table, it's better to create the indexes afterwards than to fill |
| 304 |
* them incrementally while we load the table. |
| 305 |
* |
| 306 |
* If indexOid is InvalidOid, the table will be rewritten in physical order |
| 307 |
* instead of index order. This is the new implementation of VACUUM FULL, |
| 308 |
* and error messages should refer to the operation as VACUUM not CLUSTER. |
| 309 |
*/ |
| 310 |
void |
| 311 |
cluster_rel(Relation OldHeap, Oid indexOid, ClusterParams *params) |
| 312 |
{ |
| 313 |
Oid tableOid = RelationGetRelid(OldHeap); |
| 314 |
Oid save_userid; |
| 315 |
int save_sec_context; |
| 316 |
int save_nestlevel; |
| 317 |
bool verbose = ((params->options & CLUOPT_VERBOSE) != 0); |
| 318 |
bool recheck = ((params->options & CLUOPT_RECHECK) != 0); |
| 319 |
Relation index; |
| 320 |
|
| 321 |
Assert(CheckRelationLockedByMe(OldHeap, AccessExclusiveLock, false)); |
| 322 |
|
| 323 |
/* Check for user-requested abort. */ |
| 324 |
CHECK_FOR_INTERRUPTS(); |
| 325 |
|
| 326 |
pgstat_progress_start_command(PROGRESS_COMMAND_CLUSTER, tableOid); |
| 327 |
if (OidIsValid(indexOid)) |
| 328 |
pgstat_progress_update_param(PROGRESS_CLUSTER_COMMAND, |
| 329 |
PROGRESS_CLUSTER_COMMAND_CLUSTER); |
| 330 |
else |
| 331 |
pgstat_progress_update_param(PROGRESS_CLUSTER_COMMAND, |
| 332 |
PROGRESS_CLUSTER_COMMAND_VACUUM_FULL); |
| 333 |
|
| 334 |
/* |
| 335 |
* Switch to the table owner's userid, so that any index functions are run |
| 336 |
* as that user. Also lock down security-restricted operations and |
| 337 |
* arrange to make GUC variable changes local to this command. |
| 338 |
*/ |
| 339 |
GetUserIdAndSecContext(&save_userid, &save_sec_context); |
| 340 |
SetUserIdAndSecContext(OldHeap->rd_rel->relowner, |
| 341 |
save_sec_context | SECURITY_RESTRICTED_OPERATION); |
| 342 |
save_nestlevel = NewGUCNestLevel(); |
| 343 |
RestrictSearchPath(); |
| 344 |
|
| 345 |
/* |
| 346 |
* Since we may open a new transaction for each relation, we have to check |
| 347 |
* that the relation still is what we think it is. |
| 348 |
* |
| 349 |
* If this is a single-transaction CLUSTER, we can skip these tests. We |
| 350 |
* *must* skip the one on indisclustered since it would reject an attempt |
| 351 |
* to cluster a not-previously-clustered index. |
| 352 |
*/ |
| 353 |
if (recheck) |
| 354 |
{ |
| 355 |
/* Check that the user still has privileges for the relation */ |
| 356 |
if (!cluster_is_permitted_for_relation(tableOid, save_userid)) |
| 357 |
{ |
| 358 |
relation_close(OldHeap, AccessExclusiveLock); |
| 359 |
goto out; |
| 360 |
} |
| 361 |
|
| 362 |
/* |
| 363 |
* Silently skip a temp table for a remote session. Only doing this |
| 364 |
* check in the "recheck" case is appropriate (which currently means |
| 365 |
* somebody is executing a database-wide CLUSTER or on a partitioned |
| 366 |
* table), because there is another check in cluster() which will stop |
| 367 |
* any attempt to cluster remote temp tables by name. There is |
| 368 |
* another check in cluster_rel which is redundant, but we leave it |
| 369 |
* for extra safety. |
| 370 |
*/ |
| 371 |
if (RELATION_IS_OTHER_TEMP(OldHeap)) |
| 372 |
{ |
| 373 |
relation_close(OldHeap, AccessExclusiveLock); |
| 374 |
goto out; |
| 375 |
} |
| 376 |
|
| 377 |
if (OidIsValid(indexOid)) |
| 378 |
{ |
| 379 |
/* |
| 380 |
* Check that the index still exists |
| 381 |
*/ |
| 382 |
if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(indexOid))) |
| 383 |
{ |
| 384 |
relation_close(OldHeap, AccessExclusiveLock); |
| 385 |
goto out; |
| 386 |
} |
| 387 |
|
| 388 |
/* |
| 389 |
* Check that the index is still the one with indisclustered set, |
| 390 |
* if needed. |
| 391 |
*/ |
| 392 |
if ((params->options & CLUOPT_RECHECK_ISCLUSTERED) != 0 && |
| 393 |
!get_index_isclustered(indexOid)) |
| 394 |
{ |
| 395 |
relation_close(OldHeap, AccessExclusiveLock); |
| 396 |
goto out; |
| 397 |
} |
| 398 |
} |
| 399 |
} |
| 400 |
|
| 401 |
/* |
| 402 |
* We allow VACUUM FULL, but not CLUSTER, on shared catalogs. CLUSTER |
| 403 |
* would work in most respects, but the index would only get marked as |
| 404 |
* indisclustered in the current database, leading to unexpected behavior |
| 405 |
* if CLUSTER were later invoked in another database. |
| 406 |
*/ |
| 407 |
if (OidIsValid(indexOid) && OldHeap->rd_rel->relisshared) |
| 408 |
ereport(ERROR, |
| 409 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 410 |
errmsg("cannot cluster a shared catalog"))); |
| 411 |
|
| 412 |
/* |
| 413 |
* Don't process temp tables of other backends ... their local buffer |
| 414 |
* manager is not going to cope. |
| 415 |
*/ |
| 416 |
if (RELATION_IS_OTHER_TEMP(OldHeap)) |
| 417 |
{ |
| 418 |
if (OidIsValid(indexOid)) |
| 419 |
ereport(ERROR, |
| 420 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 421 |
errmsg("cannot cluster temporary tables of other sessions"))); |
| 422 |
else |
| 423 |
ereport(ERROR, |
| 424 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 425 |
errmsg("cannot vacuum temporary tables of other sessions"))); |
| 426 |
} |
| 427 |
|
| 428 |
/* |
| 429 |
* Also check for active uses of the relation in the current transaction, |
| 430 |
* including open scans and pending AFTER trigger events. |
| 431 |
*/ |
| 432 |
CheckTableNotInUse(OldHeap, OidIsValid(indexOid) ? "CLUSTER" : "VACUUM"); |
| 433 |
|
| 434 |
/* Check heap and index are valid to cluster on */ |
| 435 |
if (OidIsValid(indexOid)) |
| 436 |
{ |
| 437 |
/* verify the index is good and lock it */ |
| 438 |
check_index_is_clusterable(OldHeap, indexOid, AccessExclusiveLock); |
| 439 |
/* also open it */ |
| 440 |
index = index_open(indexOid, NoLock); |
| 441 |
} |
| 442 |
else |
| 443 |
index = NULL; |
| 444 |
|
| 445 |
/* |
| 446 |
* Quietly ignore the request if this is a materialized view which has not |
| 447 |
* been populated from its query. No harm is done because there is no data |
| 448 |
* to deal with, and we don't want to throw an error if this is part of a |
| 449 |
* multi-relation request -- for example, CLUSTER was run on the entire |
| 450 |
* database. |
| 451 |
*/ |
| 452 |
if (OldHeap->rd_rel->relkind == RELKIND_MATVIEW && |
| 453 |
!RelationIsPopulated(OldHeap)) |
| 454 |
{ |
| 455 |
relation_close(OldHeap, AccessExclusiveLock); |
| 456 |
goto out; |
| 457 |
} |
| 458 |
|
| 459 |
Assert(OldHeap->rd_rel->relkind == RELKIND_RELATION || |
| 460 |
OldHeap->rd_rel->relkind == RELKIND_MATVIEW || |
| 461 |
OldHeap->rd_rel->relkind == RELKIND_TOASTVALUE); |
| 462 |
|
| 463 |
/* |
| 464 |
* All predicate locks on the tuples or pages are about to be made |
| 465 |
* invalid, because we move tuples around. Promote them to relation |
| 466 |
* locks. Predicate locks on indexes will be promoted when they are |
| 467 |
* reindexed. |
| 468 |
*/ |
| 469 |
TransferPredicateLocksToHeapRelation(OldHeap); |
| 470 |
|
| 471 |
/* rebuild_relation does all the dirty work */ |
| 472 |
rebuild_relation(OldHeap, index, verbose); |
| 473 |
/* rebuild_relation closes OldHeap, and index if valid */ |
| 474 |
|
| 475 |
out: |
| 476 |
/* Roll back any GUC changes executed by index functions */ |
| 477 |
AtEOXact_GUC(false, save_nestlevel); |
| 478 |
|
| 479 |
/* Restore userid and security context */ |
| 480 |
SetUserIdAndSecContext(save_userid, save_sec_context); |
| 481 |
|
| 482 |
pgstat_progress_end_command(); |
| 483 |
} |
| 484 |
|
| 485 |
/* |
| 486 |
* Verify that the specified heap and index are valid to cluster on |
| 487 |
* |
| 488 |
* Side effect: obtains lock on the index. The caller may |
| 489 |
* in some cases already have AccessExclusiveLock on the table, but |
| 490 |
* not in all cases so we can't rely on the table-level lock for |
| 491 |
* protection here. |
| 492 |
*/ |
| 493 |
void |
| 494 |
check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode) |
| 495 |
{ |
| 496 |
Relation OldIndex; |
| 497 |
|
| 498 |
OldIndex = index_open(indexOid, lockmode); |
| 499 |
|
| 500 |
/* |
| 501 |
* Check that index is in fact an index on the given relation |
| 502 |
*/ |
| 503 |
if (OldIndex->rd_index == NULL || |
| 504 |
OldIndex->rd_index->indrelid != RelationGetRelid(OldHeap)) |
| 505 |
ereport(ERROR, |
| 506 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 507 |
errmsg("\"%s\" is not an index for table \"%s\"", |
| 508 |
RelationGetRelationName(OldIndex), |
| 509 |
RelationGetRelationName(OldHeap)))); |
| 510 |
|
| 511 |
/* Index AM must allow clustering */ |
| 512 |
if (!OldIndex->rd_indam->amclusterable) |
| 513 |
ereport(ERROR, |
| 514 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 515 |
errmsg("cannot cluster on index \"%s\" because access method does not support clustering", |
| 516 |
RelationGetRelationName(OldIndex)))); |
| 517 |
|
| 518 |
/* |
| 519 |
* Disallow clustering on incomplete indexes (those that might not index |
| 520 |
* every row of the relation). We could relax this by making a separate |
| 521 |
* seqscan pass over the table to copy the missing rows, but that seems |
| 522 |
* expensive and tedious. |
| 523 |
*/ |
| 524 |
if (!heap_attisnull(OldIndex->rd_indextuple, Anum_pg_index_indpred, NULL)) |
| 525 |
ereport(ERROR, |
| 526 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 527 |
errmsg("cannot cluster on partial index \"%s\"", |
| 528 |
RelationGetRelationName(OldIndex)))); |
| 529 |
|
| 530 |
/* |
| 531 |
* Disallow if index is left over from a failed CREATE INDEX CONCURRENTLY; |
| 532 |
* it might well not contain entries for every heap row, or might not even |
| 533 |
* be internally consistent. (But note that we don't check indcheckxmin; |
| 534 |
* the worst consequence of following broken HOT chains would be that we |
| 535 |
* might put recently-dead tuples out-of-order in the new table, and there |
| 536 |
* is little harm in that.) |
| 537 |
*/ |
| 538 |
if (!OldIndex->rd_index->indisvalid) |
| 539 |
ereport(ERROR, |
| 540 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 541 |
errmsg("cannot cluster on invalid index \"%s\"", |
| 542 |
RelationGetRelationName(OldIndex)))); |
| 543 |
|
| 544 |
/* Drop relcache refcnt on OldIndex, but keep lock */ |
| 545 |
index_close(OldIndex, NoLock); |
| 546 |
} |
| 547 |
|
| 548 |
/* |
| 549 |
* mark_index_clustered: mark the specified index as the one clustered on |
| 550 |
* |
| 551 |
* With indexOid == InvalidOid, will mark all indexes of rel not-clustered. |
| 552 |
*/ |
| 553 |
void |
| 554 |
mark_index_clustered(Relation rel, Oid indexOid, bool is_internal) |
| 555 |
{ |
| 556 |
HeapTuple indexTuple; |
| 557 |
Form_pg_index indexForm; |
| 558 |
Relation pg_index; |
| 559 |
ListCell *index; |
| 560 |
|
| 561 |
/* Disallow applying to a partitioned table */ |
| 562 |
if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE) |
| 563 |
ereport(ERROR, |
| 564 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 565 |
errmsg("cannot mark index clustered in partitioned table"))); |
| 566 |
|
| 567 |
/* |
| 568 |
* If the index is already marked clustered, no need to do anything. |
| 569 |
*/ |
| 570 |
if (OidIsValid(indexOid)) |
| 571 |
{ |
| 572 |
if (get_index_isclustered(indexOid)) |
| 573 |
return; |
| 574 |
} |
| 575 |
|
| 576 |
/* |
| 577 |
* Check each index of the relation and set/clear the bit as needed. |
| 578 |
*/ |
| 579 |
pg_index = table_open(IndexRelationId, RowExclusiveLock); |
| 580 |
|
| 581 |
foreach(index, RelationGetIndexList(rel)) |
| 582 |
{ |
| 583 |
Oid thisIndexOid = lfirst_oid(index); |
| 584 |
|
| 585 |
indexTuple = SearchSysCacheCopy1(INDEXRELID, |
| 586 |
ObjectIdGetDatum(thisIndexOid)); |
| 587 |
if (!HeapTupleIsValid(indexTuple)) |
| 588 |
elog(ERROR, "cache lookup failed for index %u", thisIndexOid); |
| 589 |
indexForm = (Form_pg_index) GETSTRUCT(indexTuple); |
| 590 |
|
| 591 |
/* |
| 592 |
* Unset the bit if set. We know it's wrong because we checked this |
| 593 |
* earlier. |
| 594 |
*/ |
| 595 |
if (indexForm->indisclustered) |
| 596 |
{ |
| 597 |
indexForm->indisclustered = false; |
| 598 |
CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple); |
| 599 |
} |
| 600 |
else if (thisIndexOid == indexOid) |
| 601 |
{ |
| 602 |
/* this was checked earlier, but let's be real sure */ |
| 603 |
if (!indexForm->indisvalid) |
| 604 |
elog(ERROR, "cannot cluster on invalid index %u", indexOid); |
| 605 |
indexForm->indisclustered = true; |
| 606 |
CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple); |
| 607 |
} |
| 608 |
|
| 609 |
InvokeObjectPostAlterHookArg(IndexRelationId, thisIndexOid, 0, |
| 610 |
InvalidOid, is_internal); |
| 611 |
|
| 612 |
heap_freetuple(indexTuple); |
| 613 |
} |
| 614 |
|
| 615 |
table_close(pg_index, RowExclusiveLock); |
| 616 |
} |
| 617 |
|
| 618 |
/* |
| 619 |
* rebuild_relation: rebuild an existing relation in index or physical order |
| 620 |
* |
| 621 |
* OldHeap: table to rebuild. |
| 622 |
* index: index to cluster by, or NULL to rewrite in physical order. |
| 623 |
* |
| 624 |
* On entry, heap and index (if one is given) must be open, and |
| 625 |
* AccessExclusiveLock held on them. |
| 626 |
* On exit, they are closed, but locks on them are not released. |
| 627 |
*/ |
| 628 |
static void |
| 629 |
rebuild_relation(Relation OldHeap, Relation index, bool verbose) |
| 630 |
{ |
| 631 |
Oid tableOid = RelationGetRelid(OldHeap); |
| 632 |
Oid accessMethod = OldHeap->rd_rel->relam; |
| 633 |
Oid tableSpace = OldHeap->rd_rel->reltablespace; |
| 634 |
Oid OIDNewHeap; |
| 635 |
Relation NewHeap; |
| 636 |
char relpersistence; |
| 637 |
bool is_system_catalog; |
| 638 |
bool swap_toast_by_content; |
| 639 |
TransactionId frozenXid; |
| 640 |
MultiXactId cutoffMulti; |
| 641 |
|
| 642 |
Assert(CheckRelationLockedByMe(OldHeap, AccessExclusiveLock, false) && |
| 643 |
(index == NULL || CheckRelationLockedByMe(index, AccessExclusiveLock, false))); |
| 644 |
|
| 645 |
if (index) |
| 646 |
/* Mark the correct index as clustered */ |
| 647 |
mark_index_clustered(OldHeap, RelationGetRelid(index), true); |
| 648 |
|
| 649 |
/* Remember info about rel before closing OldHeap */ |
| 650 |
relpersistence = OldHeap->rd_rel->relpersistence; |
| 651 |
is_system_catalog = IsSystemRelation(OldHeap); |
| 652 |
|
| 653 |
/* |
| 654 |
* Create the transient table that will receive the re-ordered data. |
| 655 |
* |
| 656 |
* OldHeap is already locked, so no need to lock it again. make_new_heap |
| 657 |
* obtains AccessExclusiveLock on the new heap and its toast table. |
| 658 |
*/ |
| 659 |
OIDNewHeap = make_new_heap(tableOid, tableSpace, |
| 660 |
accessMethod, |
| 661 |
relpersistence, |
| 662 |
NoLock); |
| 663 |
Assert(CheckRelationOidLockedByMe(OIDNewHeap, AccessExclusiveLock, false)); |
| 664 |
NewHeap = table_open(OIDNewHeap, NoLock); |
| 665 |
|
| 666 |
/* Copy the heap data into the new table in the desired order */ |
| 667 |
copy_table_data(NewHeap, OldHeap, index, verbose, |
| 668 |
&swap_toast_by_content, &frozenXid, &cutoffMulti); |
| 669 |
|
| 670 |
|
| 671 |
/* Close relcache entries, but keep lock until transaction commit */ |
| 672 |
table_close(OldHeap, NoLock); |
| 673 |
if (index) |
| 674 |
index_close(index, NoLock); |
| 675 |
|
| 676 |
/* |
| 677 |
* Close the new relation so it can be dropped as soon as the storage is |
| 678 |
* swapped. The relation is not visible to others, so no need to unlock it |
| 679 |
* explicitly. |
| 680 |
*/ |
| 681 |
table_close(NewHeap, NoLock); |
| 682 |
|
| 683 |
/* |
| 684 |
* Swap the physical files of the target and transient tables, then |
| 685 |
* rebuild the target's indexes and throw away the transient table. |
| 686 |
*/ |
| 687 |
finish_heap_swap(tableOid, OIDNewHeap, is_system_catalog, |
| 688 |
swap_toast_by_content, false, true, |
| 689 |
frozenXid, cutoffMulti, |
| 690 |
relpersistence); |
| 691 |
} |
| 692 |
|
| 693 |
|
| 694 |
/* |
| 695 |
* Create the transient table that will be filled with new data during |
| 696 |
* CLUSTER, ALTER TABLE, and similar operations. The transient table |
| 697 |
* duplicates the logical structure of the OldHeap; but will have the |
| 698 |
* specified physical storage properties NewTableSpace, NewAccessMethod, and |
| 699 |
* relpersistence. |
| 700 |
* |
| 701 |
* After this, the caller should load the new heap with transferred/modified |
| 702 |
* data, then call finish_heap_swap to complete the operation. |
| 703 |
*/ |
| 704 |
Oid |
| 705 |
make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod, |
| 706 |
char relpersistence, LOCKMODE lockmode) |
| 707 |
{ |
| 708 |
TupleDesc OldHeapDesc; |
| 709 |
char NewHeapName[NAMEDATALEN]; |
| 710 |
Oid OIDNewHeap; |
| 711 |
Oid toastid; |
| 712 |
Relation OldHeap; |
| 713 |
HeapTuple tuple; |
| 714 |
Datum reloptions; |
| 715 |
bool isNull; |
| 716 |
Oid namespaceid; |
| 717 |
|
| 718 |
OldHeap = table_open(OIDOldHeap, lockmode); |
| 719 |
OldHeapDesc = RelationGetDescr(OldHeap); |
| 720 |
|
| 721 |
/* |
| 722 |
* Note that the NewHeap will not receive any of the defaults or |
| 723 |
* constraints associated with the OldHeap; we don't need 'em, and there's |
| 724 |
* no reason to spend cycles inserting them into the catalogs only to |
| 725 |
* delete them. |
| 726 |
*/ |
| 727 |
|
| 728 |
/* |
| 729 |
* But we do want to use reloptions of the old heap for new heap. |
| 730 |
*/ |
| 731 |
tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(OIDOldHeap)); |
| 732 |
if (!HeapTupleIsValid(tuple)) |
| 733 |
elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap); |
| 734 |
reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions, |
| 735 |
&isNull); |
| 736 |
if (isNull) |
| 737 |
reloptions = (Datum) 0; |
| 738 |
|
| 739 |
if (relpersistence == RELPERSISTENCE_TEMP) |
| 740 |
namespaceid = LookupCreationNamespace("pg_temp"); |
| 741 |
else |
| 742 |
namespaceid = RelationGetNamespace(OldHeap); |
| 743 |
|
| 744 |
/* |
| 745 |
* Create the new heap, using a temporary name in the same namespace as |
| 746 |
* the existing table. NOTE: there is some risk of collision with user |
| 747 |
* relnames. Working around this seems more trouble than it's worth; in |
| 748 |
* particular, we can't create the new heap in a different namespace from |
| 749 |
* the old, or we will have problems with the TEMP status of temp tables. |
| 750 |
* |
| 751 |
* Note: the new heap is not a shared relation, even if we are rebuilding |
| 752 |
* a shared rel. However, we do make the new heap mapped if the source is |
| 753 |
* mapped. This simplifies swap_relation_files, and is absolutely |
| 754 |
* necessary for rebuilding pg_class, for reasons explained there. |
| 755 |
*/ |
| 756 |
snprintf(NewHeapName, sizeof(NewHeapName), "pg_temp_%u", OIDOldHeap); |
| 757 |
|
| 758 |
OIDNewHeap = heap_create_with_catalog(NewHeapName, |
| 759 |
namespaceid, |
| 760 |
NewTableSpace, |
| 761 |
InvalidOid, |
| 762 |
InvalidOid, |
| 763 |
InvalidOid, |
| 764 |
OldHeap->rd_rel->relowner, |
| 765 |
NewAccessMethod, |
| 766 |
OldHeapDesc, |
| 767 |
NIL, |
| 768 |
RELKIND_RELATION, |
| 769 |
relpersistence, |
| 770 |
false, |
| 771 |
RelationIsMapped(OldHeap), |
| 772 |
ONCOMMIT_NOOP, |
| 773 |
reloptions, |
| 774 |
false, |
| 775 |
true, |
| 776 |
true, |
| 777 |
OIDOldHeap, |
| 778 |
NULL); |
| 779 |
Assert(OIDNewHeap != InvalidOid); |
| 780 |
|
| 781 |
ReleaseSysCache(tuple); |
| 782 |
|
| 783 |
/* |
| 784 |
* Advance command counter so that the newly-created relation's catalog |
| 785 |
* tuples will be visible to table_open. |
| 786 |
*/ |
| 787 |
CommandCounterIncrement(); |
| 788 |
|
| 789 |
/* |
| 790 |
* If necessary, create a TOAST table for the new relation. |
| 791 |
* |
| 792 |
* If the relation doesn't have a TOAST table already, we can't need one |
| 793 |
* for the new relation. The other way around is possible though: if some |
| 794 |
* wide columns have been dropped, NewHeapCreateToastTable can decide that |
| 795 |
* no TOAST table is needed for the new table. |
| 796 |
* |
| 797 |
* Note that NewHeapCreateToastTable ends with CommandCounterIncrement, so |
| 798 |
* that the TOAST table will be visible for insertion. |
| 799 |
*/ |
| 800 |
toastid = OldHeap->rd_rel->reltoastrelid; |
| 801 |
if (OidIsValid(toastid)) |
| 802 |
{ |
| 803 |
/* keep the existing toast table's reloptions, if any */ |
| 804 |
tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(toastid)); |
| 805 |
if (!HeapTupleIsValid(tuple)) |
| 806 |
elog(ERROR, "cache lookup failed for relation %u", toastid); |
| 807 |
reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions, |
| 808 |
&isNull); |
| 809 |
if (isNull) |
| 810 |
reloptions = (Datum) 0; |
| 811 |
|
| 812 |
NewHeapCreateToastTable(OIDNewHeap, reloptions, lockmode, toastid); |
| 813 |
|
| 814 |
ReleaseSysCache(tuple); |
| 815 |
} |
| 816 |
|
| 817 |
table_close(OldHeap, NoLock); |
| 818 |
|
| 819 |
return OIDNewHeap; |
| 820 |
} |
| 821 |
|
| 822 |
/* |
| 823 |
* Do the physical copying of table data. |
| 824 |
* |
| 825 |
* There are three output parameters: |
| 826 |
* *pSwapToastByContent is set true if toast tables must be swapped by content. |
| 827 |
* *pFreezeXid receives the TransactionId used as freeze cutoff point. |
| 828 |
* *pCutoffMulti receives the MultiXactId used as a cutoff point. |
| 829 |
*/ |
| 830 |
static void |
| 831 |
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex, bool verbose, |
| 832 |
bool *pSwapToastByContent, TransactionId *pFreezeXid, |
| 833 |
MultiXactId *pCutoffMulti) |
| 834 |
{ |
| 835 |
Relation relRelation; |
| 836 |
HeapTuple reltup; |
| 837 |
Form_pg_class relform; |
| 838 |
TupleDesc oldTupDesc PG_USED_FOR_ASSERTS_ONLY; |
| 839 |
TupleDesc newTupDesc PG_USED_FOR_ASSERTS_ONLY; |
| 840 |
VacuumParams params; |
| 841 |
struct VacuumCutoffs cutoffs; |
| 842 |
bool use_sort; |
| 843 |
double num_tuples = 0, |
| 844 |
tups_vacuumed = 0, |
| 845 |
tups_recently_dead = 0; |
| 846 |
BlockNumber num_pages; |
| 847 |
int elevel = verbose ? INFO : DEBUG2; |
| 848 |
PGRUsage ru0; |
| 849 |
char *nspname; |
| 850 |
|
| 851 |
pg_rusage_init(&ru0); |
| 852 |
|
| 853 |
/* Store a copy of the namespace name for logging purposes */ |
| 854 |
nspname = get_namespace_name(RelationGetNamespace(OldHeap)); |
| 855 |
|
| 856 |
/* |
| 857 |
* Their tuple descriptors should be exactly alike, but here we only need |
| 858 |
* assume that they have the same number of columns. |
| 859 |
*/ |
| 860 |
oldTupDesc = RelationGetDescr(OldHeap); |
| 861 |
newTupDesc = RelationGetDescr(NewHeap); |
| 862 |
Assert(newTupDesc->natts == oldTupDesc->natts); |
| 863 |
|
| 864 |
/* |
| 865 |
* If the OldHeap has a toast table, get lock on the toast table to keep |
| 866 |
* it from being vacuumed. This is needed because autovacuum processes |
| 867 |
* toast tables independently of their main tables, with no lock on the |
| 868 |
* latter. If an autovacuum were to start on the toast table after we |
| 869 |
* compute our OldestXmin below, it would use a later OldestXmin, and then |
| 870 |
* possibly remove as DEAD toast tuples belonging to main tuples we think |
| 871 |
* are only RECENTLY_DEAD. Then we'd fail while trying to copy those |
| 872 |
* tuples. |
| 873 |
* |
| 874 |
* We don't need to open the toast relation here, just lock it. The lock |
| 875 |
* will be held till end of transaction. |
| 876 |
*/ |
| 877 |
if (OldHeap->rd_rel->reltoastrelid) |
| 878 |
LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock); |
| 879 |
|
| 880 |
/* |
| 881 |
* If both tables have TOAST tables, perform toast swap by content. It is |
| 882 |
* possible that the old table has a toast table but the new one doesn't, |
| 883 |
* if toastable columns have been dropped. In that case we have to do |
| 884 |
* swap by links. This is okay because swap by content is only essential |
| 885 |
* for system catalogs, and we don't support schema changes for them. |
| 886 |
*/ |
| 887 |
if (OldHeap->rd_rel->reltoastrelid && NewHeap->rd_rel->reltoastrelid) |
| 888 |
{ |
| 889 |
*pSwapToastByContent = true; |
| 890 |
|
| 891 |
/* |
| 892 |
* When doing swap by content, any toast pointers written into NewHeap |
| 893 |
* must use the old toast table's OID, because that's where the toast |
| 894 |
* data will eventually be found. Set this up by setting rd_toastoid. |
| 895 |
* This also tells toast_save_datum() to preserve the toast value |
| 896 |
* OIDs, which we want so as not to invalidate toast pointers in |
| 897 |
* system catalog caches, and to avoid making multiple copies of a |
| 898 |
* single toast value. |
| 899 |
* |
| 900 |
* Note that we must hold NewHeap open until we are done writing data, |
| 901 |
* since the relcache will not guarantee to remember this setting once |
| 902 |
* the relation is closed. Also, this technique depends on the fact |
| 903 |
* that no one will try to read from the NewHeap until after we've |
| 904 |
* finished writing it and swapping the rels --- otherwise they could |
| 905 |
* follow the toast pointers to the wrong place. (It would actually |
| 906 |
* work for values copied over from the old toast table, but not for |
| 907 |
* any values that we toast which were previously not toasted.) |
| 908 |
*/ |
| 909 |
NewHeap->rd_toastoid = OldHeap->rd_rel->reltoastrelid; |
| 910 |
} |
| 911 |
else |
| 912 |
*pSwapToastByContent = false; |
| 913 |
|
| 914 |
/* |
| 915 |
* Compute xids used to freeze and weed out dead tuples and multixacts. |
| 916 |
* Since we're going to rewrite the whole table anyway, there's no reason |
| 917 |
* not to be aggressive about this. |
| 918 |
*/ |
| 919 |
memset(¶ms, 0, sizeof(VacuumParams)); |
| 920 |
vacuum_get_cutoffs(OldHeap, ¶ms, &cutoffs); |
| 921 |
|
| 922 |
/* |
| 923 |
* FreezeXid will become the table's new relfrozenxid, and that mustn't go |
| 924 |
* backwards, so take the max. |
| 925 |
*/ |
| 926 |
{ |
| 927 |
TransactionId relfrozenxid = OldHeap->rd_rel->relfrozenxid; |
| 928 |
|
| 929 |
if (TransactionIdIsValid(relfrozenxid) && |
| 930 |
TransactionIdPrecedes(cutoffs.FreezeLimit, relfrozenxid)) |
| 931 |
cutoffs.FreezeLimit = relfrozenxid; |
| 932 |
} |
| 933 |
|
| 934 |
/* |
| 935 |
* MultiXactCutoff, similarly, shouldn't go backwards either. |
| 936 |
*/ |
| 937 |
{ |
| 938 |
MultiXactId relminmxid = OldHeap->rd_rel->relminmxid; |
| 939 |
|
| 940 |
if (MultiXactIdIsValid(relminmxid) && |
| 941 |
MultiXactIdPrecedes(cutoffs.MultiXactCutoff, relminmxid)) |
| 942 |
cutoffs.MultiXactCutoff = relminmxid; |
| 943 |
} |
| 944 |
|
| 945 |
/* |
| 946 |
* Decide whether to use an indexscan or seqscan-and-optional-sort to scan |
| 947 |
* the OldHeap. We know how to use a sort to duplicate the ordering of a |
| 948 |
* btree index, and will use seqscan-and-sort for that case if the planner |
| 949 |
* tells us it's cheaper. Otherwise, always indexscan if an index is |
| 950 |
* provided, else plain seqscan. |
| 951 |
*/ |
| 952 |
if (OldIndex != NULL && OldIndex->rd_rel->relam == BTREE_AM_OID) |
| 953 |
use_sort = plan_cluster_use_sort(RelationGetRelid(OldHeap), |
| 954 |
RelationGetRelid(OldIndex)); |
| 955 |
else |
| 956 |
use_sort = false; |
| 957 |
|
| 958 |
/* Log what we're doing */ |
| 959 |
if (OldIndex != NULL && !use_sort) |
| 960 |
ereport(elevel, |
| 961 |
(errmsg("clustering \"%s.%s\" using index scan on \"%s\"", |
| 962 |
nspname, |
| 963 |
RelationGetRelationName(OldHeap), |
| 964 |
RelationGetRelationName(OldIndex)))); |
| 965 |
else if (use_sort) |
| 966 |
ereport(elevel, |
| 967 |
(errmsg("clustering \"%s.%s\" using sequential scan and sort", |
| 968 |
nspname, |
| 969 |
RelationGetRelationName(OldHeap)))); |
| 970 |
else |
| 971 |
ereport(elevel, |
| 972 |
(errmsg("vacuuming \"%s.%s\"", |
| 973 |
nspname, |
| 974 |
RelationGetRelationName(OldHeap)))); |
| 975 |
|
| 976 |
/* |
| 977 |
* Hand off the actual copying to AM specific function, the generic code |
| 978 |
* cannot know how to deal with visibility across AMs. Note that this |
| 979 |
* routine is allowed to set FreezeXid / MultiXactCutoff to different |
| 980 |
* values (e.g. because the AM doesn't use freezing). |
| 981 |
*/ |
| 982 |
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort, |
| 983 |
cutoffs.OldestXmin, &cutoffs.FreezeLimit, |
| 984 |
&cutoffs.MultiXactCutoff, |
| 985 |
&num_tuples, &tups_vacuumed, |
| 986 |
&tups_recently_dead); |
| 987 |
|
| 988 |
/* return selected values to caller, get set as relfrozenxid/minmxid */ |
| 989 |
*pFreezeXid = cutoffs.FreezeLimit; |
| 990 |
*pCutoffMulti = cutoffs.MultiXactCutoff; |
| 991 |
|
| 992 |
/* Reset rd_toastoid just to be tidy --- it shouldn't be looked at again */ |
| 993 |
NewHeap->rd_toastoid = InvalidOid; |
| 994 |
|
| 995 |
num_pages = RelationGetNumberOfBlocks(NewHeap); |
| 996 |
|
| 997 |
/* Log what we did */ |
| 998 |
ereport(elevel, |
| 999 |
(errmsg("\"%s.%s\": found %.0f removable, %.0f nonremovable row versions in %u pages", |
| 1000 |
nspname, |
| 1001 |
RelationGetRelationName(OldHeap), |
| 1002 |
tups_vacuumed, num_tuples, |
| 1003 |
RelationGetNumberOfBlocks(OldHeap)), |
| 1004 |
errdetail("%.0f dead row versions cannot be removed yet.\n" |
| 1005 |
"%s.", |
| 1006 |
tups_recently_dead, |
| 1007 |
pg_rusage_show(&ru0)))); |
| 1008 |
|
| 1009 |
/* Update pg_class to reflect the correct values of pages and tuples. */ |
| 1010 |
relRelation = table_open(RelationRelationId, RowExclusiveLock); |
| 1011 |
|
| 1012 |
reltup = SearchSysCacheCopy1(RELOID, |
| 1013 |
ObjectIdGetDatum(RelationGetRelid(NewHeap))); |
| 1014 |
if (!HeapTupleIsValid(reltup)) |
| 1015 |
elog(ERROR, "cache lookup failed for relation %u", |
| 1016 |
RelationGetRelid(NewHeap)); |
| 1017 |
relform = (Form_pg_class) GETSTRUCT(reltup); |
| 1018 |
|
| 1019 |
relform->relpages = num_pages; |
| 1020 |
relform->reltuples = num_tuples; |
| 1021 |
|
| 1022 |
/* Don't update the stats for pg_class. See swap_relation_files. */ |
| 1023 |
if (RelationGetRelid(OldHeap) != RelationRelationId) |
| 1024 |
CatalogTupleUpdate(relRelation, &reltup->t_self, reltup); |
| 1025 |
else |
| 1026 |
CacheInvalidateRelcacheByTuple(reltup); |
| 1027 |
|
| 1028 |
/* Clean up. */ |
| 1029 |
heap_freetuple(reltup); |
| 1030 |
table_close(relRelation, RowExclusiveLock); |
| 1031 |
|
| 1032 |
/* Make the update visible */ |
| 1033 |
CommandCounterIncrement(); |
| 1034 |
} |
| 1035 |
|
| 1036 |
/* |
| 1037 |
* Swap the physical files of two given relations. |
| 1038 |
* |
| 1039 |
* We swap the physical identity (reltablespace, relfilenumber) while keeping |
| 1040 |
* the same logical identities of the two relations. relpersistence is also |
| 1041 |
* swapped, which is critical since it determines where buffers live for each |
| 1042 |
* relation. |
| 1043 |
* |
| 1044 |
* We can swap associated TOAST data in either of two ways: recursively swap |
| 1045 |
* the physical content of the toast tables (and their indexes), or swap the |
| 1046 |
* TOAST links in the given relations' pg_class entries. The former is needed |
| 1047 |
* to manage rewrites of shared catalogs (where we cannot change the pg_class |
| 1048 |
* links) while the latter is the only way to handle cases in which a toast |
| 1049 |
* table is added or removed altogether. |
| 1050 |
* |
| 1051 |
* Additionally, the first relation is marked with relfrozenxid set to |
| 1052 |
* frozenXid. It seems a bit ugly to have this here, but the caller would |
| 1053 |
* have to do it anyway, so having it here saves a heap_update. Note: in |
| 1054 |
* the swap-toast-links case, we assume we don't need to change the toast |
| 1055 |
* table's relfrozenxid: the new version of the toast table should already |
| 1056 |
* have relfrozenxid set to RecentXmin, which is good enough. |
| 1057 |
* |
| 1058 |
* Lastly, if r2 and its toast table and toast index (if any) are mapped, |
| 1059 |
* their OIDs are emitted into mapped_tables[]. This is hacky but beats |
| 1060 |
* having to look the information up again later in finish_heap_swap. |
| 1061 |
*/ |
| 1062 |
static void |
| 1063 |
swap_relation_files(Oid r1, Oid r2, bool target_is_pg_class, |
| 1064 |
bool swap_toast_by_content, |
| 1065 |
bool is_internal, |
| 1066 |
TransactionId frozenXid, |
| 1067 |
MultiXactId cutoffMulti, |
| 1068 |
Oid *mapped_tables) |
| 1069 |
{ |
| 1070 |
Relation relRelation; |
| 1071 |
HeapTuple reltup1, |
| 1072 |
reltup2; |
| 1073 |
Form_pg_class relform1, |
| 1074 |
relform2; |
| 1075 |
RelFileNumber relfilenumber1, |
| 1076 |
relfilenumber2; |
| 1077 |
RelFileNumber swaptemp; |
| 1078 |
char swptmpchr; |
| 1079 |
Oid relam1, |
| 1080 |
relam2; |
| 1081 |
|
| 1082 |
/* We need writable copies of both pg_class tuples. */ |
| 1083 |
relRelation = table_open(RelationRelationId, RowExclusiveLock); |
| 1084 |
|
| 1085 |
reltup1 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r1)); |
| 1086 |
if (!HeapTupleIsValid(reltup1)) |
| 1087 |
elog(ERROR, "cache lookup failed for relation %u", r1); |
| 1088 |
relform1 = (Form_pg_class) GETSTRUCT(reltup1); |
| 1089 |
|
| 1090 |
reltup2 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r2)); |
| 1091 |
if (!HeapTupleIsValid(reltup2)) |
| 1092 |
elog(ERROR, "cache lookup failed for relation %u", r2); |
| 1093 |
relform2 = (Form_pg_class) GETSTRUCT(reltup2); |
| 1094 |
|
| 1095 |
relfilenumber1 = relform1->relfilenode; |
| 1096 |
relfilenumber2 = relform2->relfilenode; |
| 1097 |
relam1 = relform1->relam; |
| 1098 |
relam2 = relform2->relam; |
| 1099 |
|
| 1100 |
if (RelFileNumberIsValid(relfilenumber1) && |
| 1101 |
RelFileNumberIsValid(relfilenumber2)) |
| 1102 |
{ |
| 1103 |
/* |
| 1104 |
* Normal non-mapped relations: swap relfilenumbers, reltablespaces, |
| 1105 |
* relpersistence |
| 1106 |
*/ |
| 1107 |
Assert(!target_is_pg_class); |
| 1108 |
|
| 1109 |
swaptemp = relform1->relfilenode; |
| 1110 |
relform1->relfilenode = relform2->relfilenode; |
| 1111 |
relform2->relfilenode = swaptemp; |
| 1112 |
|
| 1113 |
swaptemp = relform1->reltablespace; |
| 1114 |
relform1->reltablespace = relform2->reltablespace; |
| 1115 |
relform2->reltablespace = swaptemp; |
| 1116 |
|
| 1117 |
swaptemp = relform1->relam; |
| 1118 |
relform1->relam = relform2->relam; |
| 1119 |
relform2->relam = swaptemp; |
| 1120 |
|
| 1121 |
swptmpchr = relform1->relpersistence; |
| 1122 |
relform1->relpersistence = relform2->relpersistence; |
| 1123 |
relform2->relpersistence = swptmpchr; |
| 1124 |
|
| 1125 |
/* Also swap toast links, if we're swapping by links */ |
| 1126 |
if (!swap_toast_by_content) |
| 1127 |
{ |
| 1128 |
swaptemp = relform1->reltoastrelid; |
| 1129 |
relform1->reltoastrelid = relform2->reltoastrelid; |
| 1130 |
relform2->reltoastrelid = swaptemp; |
| 1131 |
} |
| 1132 |
} |
| 1133 |
else |
| 1134 |
{ |
| 1135 |
/* |
| 1136 |
* Mapped-relation case. Here we have to swap the relation mappings |
| 1137 |
* instead of modifying the pg_class columns. Both must be mapped. |
| 1138 |
*/ |
| 1139 |
if (RelFileNumberIsValid(relfilenumber1) || |
| 1140 |
RelFileNumberIsValid(relfilenumber2)) |
| 1141 |
elog(ERROR, "cannot swap mapped relation \"%s\" with non-mapped relation", |
| 1142 |
NameStr(relform1->relname)); |
| 1143 |
|
| 1144 |
/* |
| 1145 |
* We can't change the tablespace nor persistence of a mapped rel, and |
| 1146 |
* we can't handle toast link swapping for one either, because we must |
| 1147 |
* not apply any critical changes to its pg_class row. These cases |
| 1148 |
* should be prevented by upstream permissions tests, so these checks |
| 1149 |
* are non-user-facing emergency backstop. |
| 1150 |
*/ |
| 1151 |
if (relform1->reltablespace != relform2->reltablespace) |
| 1152 |
elog(ERROR, "cannot change tablespace of mapped relation \"%s\"", |
| 1153 |
NameStr(relform1->relname)); |
| 1154 |
if (relform1->relpersistence != relform2->relpersistence) |
| 1155 |
elog(ERROR, "cannot change persistence of mapped relation \"%s\"", |
| 1156 |
NameStr(relform1->relname)); |
| 1157 |
if (relform1->relam != relform2->relam) |
| 1158 |
elog(ERROR, "cannot change access method of mapped relation \"%s\"", |
| 1159 |
NameStr(relform1->relname)); |
| 1160 |
if (!swap_toast_by_content && |
| 1161 |
(relform1->reltoastrelid || relform2->reltoastrelid)) |
| 1162 |
elog(ERROR, "cannot swap toast by links for mapped relation \"%s\"", |
| 1163 |
NameStr(relform1->relname)); |
| 1164 |
|
| 1165 |
/* |
| 1166 |
* Fetch the mappings --- shouldn't fail, but be paranoid |
| 1167 |
*/ |
| 1168 |
relfilenumber1 = RelationMapOidToFilenumber(r1, relform1->relisshared); |
| 1169 |
if (!RelFileNumberIsValid(relfilenumber1)) |
| 1170 |
elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u", |
| 1171 |
NameStr(relform1->relname), r1); |
| 1172 |
relfilenumber2 = RelationMapOidToFilenumber(r2, relform2->relisshared); |
| 1173 |
if (!RelFileNumberIsValid(relfilenumber2)) |
| 1174 |
elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u", |
| 1175 |
NameStr(relform2->relname), r2); |
| 1176 |
|
| 1177 |
/* |
| 1178 |
* Send replacement mappings to relmapper. Note these won't actually |
| 1179 |
* take effect until CommandCounterIncrement. |
| 1180 |
*/ |
| 1181 |
RelationMapUpdateMap(r1, relfilenumber2, relform1->relisshared, false); |
| 1182 |
RelationMapUpdateMap(r2, relfilenumber1, relform2->relisshared, false); |
| 1183 |
|
| 1184 |
/* Pass OIDs of mapped r2 tables back to caller */ |
| 1185 |
*mapped_tables++ = r2; |
| 1186 |
} |
| 1187 |
|
| 1188 |
/* |
| 1189 |
* Recognize that rel1's relfilenumber (swapped from rel2) is new in this |
| 1190 |
* subtransaction. The rel2 storage (swapped from rel1) may or may not be |
| 1191 |
* new. |
| 1192 |
*/ |
| 1193 |
{ |
| 1194 |
Relation rel1, |
| 1195 |
rel2; |
| 1196 |
|
| 1197 |
rel1 = relation_open(r1, NoLock); |
| 1198 |
rel2 = relation_open(r2, NoLock); |
| 1199 |
rel2->rd_createSubid = rel1->rd_createSubid; |
| 1200 |
rel2->rd_newRelfilelocatorSubid = rel1->rd_newRelfilelocatorSubid; |
| 1201 |
rel2->rd_firstRelfilelocatorSubid = rel1->rd_firstRelfilelocatorSubid; |
| 1202 |
RelationAssumeNewRelfilelocator(rel1); |
| 1203 |
relation_close(rel1, NoLock); |
| 1204 |
relation_close(rel2, NoLock); |
| 1205 |
} |
| 1206 |
|
| 1207 |
/* |
| 1208 |
* In the case of a shared catalog, these next few steps will only affect |
| 1209 |
* our own database's pg_class row; but that's okay, because they are all |
| 1210 |
* noncritical updates. That's also an important fact for the case of a |
| 1211 |
* mapped catalog, because it's possible that we'll commit the map change |
| 1212 |
* and then fail to commit the pg_class update. |
| 1213 |
*/ |
| 1214 |
|
| 1215 |
/* set rel1's frozen Xid and minimum MultiXid */ |
| 1216 |
if (relform1->relkind != RELKIND_INDEX) |
| 1217 |
{ |
| 1218 |
Assert(!TransactionIdIsValid(frozenXid) || |
| 1219 |
TransactionIdIsNormal(frozenXid)); |
| 1220 |
relform1->relfrozenxid = frozenXid; |
| 1221 |
relform1->relminmxid = cutoffMulti; |
| 1222 |
} |
| 1223 |
|
| 1224 |
/* swap size statistics too, since new rel has freshly-updated stats */ |
| 1225 |
{ |
| 1226 |
int32 swap_pages; |
| 1227 |
float4 swap_tuples; |
| 1228 |
int32 swap_allvisible; |
| 1229 |
int32 swap_allfrozen; |
| 1230 |
|
| 1231 |
swap_pages = relform1->relpages; |
| 1232 |
relform1->relpages = relform2->relpages; |
| 1233 |
relform2->relpages = swap_pages; |
| 1234 |
|
| 1235 |
swap_tuples = relform1->reltuples; |
| 1236 |
relform1->reltuples = relform2->reltuples; |
| 1237 |
relform2->reltuples = swap_tuples; |
| 1238 |
|
| 1239 |
swap_allvisible = relform1->relallvisible; |
| 1240 |
relform1->relallvisible = relform2->relallvisible; |
| 1241 |
relform2->relallvisible = swap_allvisible; |
| 1242 |
|
| 1243 |
swap_allfrozen = relform1->relallfrozen; |
| 1244 |
relform1->relallfrozen = relform2->relallfrozen; |
| 1245 |
relform2->relallfrozen = swap_allfrozen; |
| 1246 |
} |
| 1247 |
|
| 1248 |
/* |
| 1249 |
* Update the tuples in pg_class --- unless the target relation of the |
| 1250 |
* swap is pg_class itself. In that case, there is zero point in making |
| 1251 |
* changes because we'd be updating the old data that we're about to throw |
| 1252 |
* away. Because the real work being done here for a mapped relation is |
| 1253 |
* just to change the relation map settings, it's all right to not update |
| 1254 |
* the pg_class rows in this case. The most important changes will instead |
| 1255 |
* performed later, in finish_heap_swap() itself. |
| 1256 |
*/ |
| 1257 |
if (!target_is_pg_class) |
| 1258 |
{ |
| 1259 |
CatalogIndexState indstate; |
| 1260 |
|
| 1261 |
indstate = CatalogOpenIndexes(relRelation); |
| 1262 |
CatalogTupleUpdateWithInfo(relRelation, &reltup1->t_self, reltup1, |
| 1263 |
indstate); |
| 1264 |
CatalogTupleUpdateWithInfo(relRelation, &reltup2->t_self, reltup2, |
| 1265 |
indstate); |
| 1266 |
CatalogCloseIndexes(indstate); |
| 1267 |
} |
| 1268 |
else |
| 1269 |
{ |
| 1270 |
/* no update ... but we do still need relcache inval */ |
| 1271 |
CacheInvalidateRelcacheByTuple(reltup1); |
| 1272 |
CacheInvalidateRelcacheByTuple(reltup2); |
| 1273 |
} |
| 1274 |
|
| 1275 |
/* |
| 1276 |
* Now that pg_class has been updated with its relevant information for |
| 1277 |
* the swap, update the dependency of the relations to point to their new |
| 1278 |
* table AM, if it has changed. |
| 1279 |
*/ |
| 1280 |
if (relam1 != relam2) |
| 1281 |
{ |
| 1282 |
if (changeDependencyFor(RelationRelationId, |
| 1283 |
r1, |
| 1284 |
AccessMethodRelationId, |
| 1285 |
relam1, |
| 1286 |
relam2) != 1) |
| 1287 |
elog(ERROR, "could not change access method dependency for relation \"%s.%s\"", |
| 1288 |
get_namespace_name(get_rel_namespace(r1)), |
| 1289 |
get_rel_name(r1)); |
| 1290 |
if (changeDependencyFor(RelationRelationId, |
| 1291 |
r2, |
| 1292 |
AccessMethodRelationId, |
| 1293 |
relam2, |
| 1294 |
relam1) != 1) |
| 1295 |
elog(ERROR, "could not change access method dependency for relation \"%s.%s\"", |
| 1296 |
get_namespace_name(get_rel_namespace(r2)), |
| 1297 |
get_rel_name(r2)); |
| 1298 |
} |
| 1299 |
|
| 1300 |
/* |
| 1301 |
* Post alter hook for modified relations. The change to r2 is always |
| 1302 |
* internal, but r1 depends on the invocation context. |
| 1303 |
*/ |
| 1304 |
InvokeObjectPostAlterHookArg(RelationRelationId, r1, 0, |
| 1305 |
InvalidOid, is_internal); |
| 1306 |
InvokeObjectPostAlterHookArg(RelationRelationId, r2, 0, |
| 1307 |
InvalidOid, true); |
| 1308 |
|
| 1309 |
/* |
| 1310 |
* If we have toast tables associated with the relations being swapped, |
| 1311 |
* deal with them too. |
| 1312 |
*/ |
| 1313 |
if (relform1->reltoastrelid || relform2->reltoastrelid) |
| 1314 |
{ |
| 1315 |
if (swap_toast_by_content) |
| 1316 |
{ |
| 1317 |
if (relform1->reltoastrelid && relform2->reltoastrelid) |
| 1318 |
{ |
| 1319 |
/* Recursively swap the contents of the toast tables */ |
| 1320 |
swap_relation_files(relform1->reltoastrelid, |
| 1321 |
relform2->reltoastrelid, |
| 1322 |
target_is_pg_class, |
| 1323 |
swap_toast_by_content, |
| 1324 |
is_internal, |
| 1325 |
frozenXid, |
| 1326 |
cutoffMulti, |
| 1327 |
mapped_tables); |
| 1328 |
} |
| 1329 |
else |
| 1330 |
{ |
| 1331 |
/* caller messed up */ |
| 1332 |
elog(ERROR, "cannot swap toast files by content when there's only one"); |
| 1333 |
} |
| 1334 |
} |
| 1335 |
else |
| 1336 |
{ |
| 1337 |
/* |
| 1338 |
* We swapped the ownership links, so we need to change dependency |
| 1339 |
* data to match. |
| 1340 |
* |
| 1341 |
* NOTE: it is possible that only one table has a toast table. |
| 1342 |
* |
| 1343 |
* NOTE: at present, a TOAST table's only dependency is the one on |
| 1344 |
* its owning table. If more are ever created, we'd need to use |
| 1345 |
* something more selective than deleteDependencyRecordsFor() to |
| 1346 |
* get rid of just the link we want. |
| 1347 |
*/ |
| 1348 |
ObjectAddress baseobject, |
| 1349 |
toastobject; |
| 1350 |
long count; |
| 1351 |
|
| 1352 |
/* |
| 1353 |
* We disallow this case for system catalogs, to avoid the |
| 1354 |
* possibility that the catalog we're rebuilding is one of the |
| 1355 |
* ones the dependency changes would change. It's too late to be |
| 1356 |
* making any data changes to the target catalog. |
| 1357 |
*/ |
| 1358 |
if (IsSystemClass(r1, relform1)) |
| 1359 |
elog(ERROR, "cannot swap toast files by links for system catalogs"); |
| 1360 |
|
| 1361 |
/* Delete old dependencies */ |
| 1362 |
if (relform1->reltoastrelid) |
| 1363 |
{ |
| 1364 |
count = deleteDependencyRecordsFor(RelationRelationId, |
| 1365 |
relform1->reltoastrelid, |
| 1366 |
false); |
| 1367 |
if (count != 1) |
| 1368 |
elog(ERROR, "expected one dependency record for TOAST table, found %ld", |
| 1369 |
count); |
| 1370 |
} |
| 1371 |
if (relform2->reltoastrelid) |
| 1372 |
{ |
| 1373 |
count = deleteDependencyRecordsFor(RelationRelationId, |
| 1374 |
relform2->reltoastrelid, |
| 1375 |
false); |
| 1376 |
if (count != 1) |
| 1377 |
elog(ERROR, "expected one dependency record for TOAST table, found %ld", |
| 1378 |
count); |
| 1379 |
} |
| 1380 |
|
| 1381 |
/* Register new dependencies */ |
| 1382 |
baseobject.classId = RelationRelationId; |
| 1383 |
baseobject.objectSubId = 0; |
| 1384 |
toastobject.classId = RelationRelationId; |
| 1385 |
toastobject.objectSubId = 0; |
| 1386 |
|
| 1387 |
if (relform1->reltoastrelid) |
| 1388 |
{ |
| 1389 |
baseobject.objectId = r1; |
| 1390 |
toastobject.objectId = relform1->reltoastrelid; |
| 1391 |
recordDependencyOn(&toastobject, &baseobject, |
| 1392 |
DEPENDENCY_INTERNAL); |
| 1393 |
} |
| 1394 |
|
| 1395 |
if (relform2->reltoastrelid) |
| 1396 |
{ |
| 1397 |
baseobject.objectId = r2; |
| 1398 |
toastobject.objectId = relform2->reltoastrelid; |
| 1399 |
recordDependencyOn(&toastobject, &baseobject, |
| 1400 |
DEPENDENCY_INTERNAL); |
| 1401 |
} |
| 1402 |
} |
| 1403 |
} |
| 1404 |
|
| 1405 |
/* |
| 1406 |
* If we're swapping two toast tables by content, do the same for their |
| 1407 |
* valid index. The swap can actually be safely done only if the relations |
| 1408 |
* have indexes. |
| 1409 |
*/ |
| 1410 |
if (swap_toast_by_content && |
| 1411 |
relform1->relkind == RELKIND_TOASTVALUE && |
| 1412 |
relform2->relkind == RELKIND_TOASTVALUE) |
| 1413 |
{ |
| 1414 |
Oid toastIndex1, |
| 1415 |
toastIndex2; |
| 1416 |
|
| 1417 |
/* Get valid index for each relation */ |
| 1418 |
toastIndex1 = toast_get_valid_index(r1, |
| 1419 |
AccessExclusiveLock); |
| 1420 |
toastIndex2 = toast_get_valid_index(r2, |
| 1421 |
AccessExclusiveLock); |
| 1422 |
|
| 1423 |
swap_relation_files(toastIndex1, |
| 1424 |
toastIndex2, |
| 1425 |
target_is_pg_class, |
| 1426 |
swap_toast_by_content, |
| 1427 |
is_internal, |
| 1428 |
InvalidTransactionId, |
| 1429 |
InvalidMultiXactId, |
| 1430 |
mapped_tables); |
| 1431 |
} |
| 1432 |
|
| 1433 |
/* Clean up. */ |
| 1434 |
heap_freetuple(reltup1); |
| 1435 |
heap_freetuple(reltup2); |
| 1436 |
|
| 1437 |
table_close(relRelation, RowExclusiveLock); |
| 1438 |
} |
| 1439 |
|
| 1440 |
/* |
| 1441 |
* Remove the transient table that was built by make_new_heap, and finish |
| 1442 |
* cleaning up (including rebuilding all indexes on the old heap). |
| 1443 |
*/ |
| 1444 |
void |
| 1445 |
finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap, |
| 1446 |
bool is_system_catalog, |
| 1447 |
bool swap_toast_by_content, |
| 1448 |
bool check_constraints, |
| 1449 |
bool is_internal, |
| 1450 |
TransactionId frozenXid, |
| 1451 |
MultiXactId cutoffMulti, |
| 1452 |
char newrelpersistence) |
| 1453 |
{ |
| 1454 |
ObjectAddress object; |
| 1455 |
Oid mapped_tables[4]; |
| 1456 |
int reindex_flags; |
| 1457 |
ReindexParams reindex_params = {0}; |
| 1458 |
int i; |
| 1459 |
|
| 1460 |
/* Report that we are now swapping relation files */ |
| 1461 |
pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, |
| 1462 |
PROGRESS_CLUSTER_PHASE_SWAP_REL_FILES); |
| 1463 |
|
| 1464 |
/* Zero out possible results from swapped_relation_files */ |
| 1465 |
memset(mapped_tables, 0, sizeof(mapped_tables)); |
| 1466 |
|
| 1467 |
/* |
| 1468 |
* Swap the contents of the heap relations (including any toast tables). |
| 1469 |
* Also set old heap's relfrozenxid to frozenXid. |
| 1470 |
*/ |
| 1471 |
swap_relation_files(OIDOldHeap, OIDNewHeap, |
| 1472 |
(OIDOldHeap == RelationRelationId), |
| 1473 |
swap_toast_by_content, is_internal, |
| 1474 |
frozenXid, cutoffMulti, mapped_tables); |
| 1475 |
|
| 1476 |
/* |
| 1477 |
* If it's a system catalog, queue a sinval message to flush all catcaches |
| 1478 |
* on the catalog when we reach CommandCounterIncrement. |
| 1479 |
*/ |
| 1480 |
if (is_system_catalog) |
| 1481 |
CacheInvalidateCatalog(OIDOldHeap); |
| 1482 |
|
| 1483 |
/* |
| 1484 |
* Rebuild each index on the relation (but not the toast table, which is |
| 1485 |
* all-new at this point). It is important to do this before the DROP |
| 1486 |
* step because if we are processing a system catalog that will be used |
| 1487 |
* during DROP, we want to have its indexes available. There is no |
| 1488 |
* advantage to the other order anyway because this is all transactional, |
| 1489 |
* so no chance to reclaim disk space before commit. We do not need a |
| 1490 |
* final CommandCounterIncrement() because reindex_relation does it. |
| 1491 |
* |
| 1492 |
* Note: because index_build is called via reindex_relation, it will never |
| 1493 |
* set indcheckxmin true for the indexes. This is OK even though in some |
| 1494 |
* sense we are building new indexes rather than rebuilding existing ones, |
| 1495 |
* because the new heap won't contain any HOT chains at all, let alone |
| 1496 |
* broken ones, so it can't be necessary to set indcheckxmin. |
| 1497 |
*/ |
| 1498 |
reindex_flags = REINDEX_REL_SUPPRESS_INDEX_USE; |
| 1499 |
if (check_constraints) |
| 1500 |
reindex_flags |= REINDEX_REL_CHECK_CONSTRAINTS; |
| 1501 |
|
| 1502 |
/* |
| 1503 |
* Ensure that the indexes have the same persistence as the parent |
| 1504 |
* relation. |
| 1505 |
*/ |
| 1506 |
if (newrelpersistence == RELPERSISTENCE_UNLOGGED) |
| 1507 |
reindex_flags |= REINDEX_REL_FORCE_INDEXES_UNLOGGED; |
| 1508 |
else if (newrelpersistence == RELPERSISTENCE_PERMANENT) |
| 1509 |
reindex_flags |= REINDEX_REL_FORCE_INDEXES_PERMANENT; |
| 1510 |
|
| 1511 |
/* Report that we are now reindexing relations */ |
| 1512 |
pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, |
| 1513 |
PROGRESS_CLUSTER_PHASE_REBUILD_INDEX); |
| 1514 |
|
| 1515 |
reindex_relation(NULL, OIDOldHeap, reindex_flags, &reindex_params); |
| 1516 |
|
| 1517 |
/* Report that we are now doing clean up */ |
| 1518 |
pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, |
| 1519 |
PROGRESS_CLUSTER_PHASE_FINAL_CLEANUP); |
| 1520 |
|
| 1521 |
/* |
| 1522 |
* If the relation being rebuilt is pg_class, swap_relation_files() |
| 1523 |
* couldn't update pg_class's own pg_class entry (check comments in |
| 1524 |
* swap_relation_files()), thus relfrozenxid was not updated. That's |
| 1525 |
* annoying because a potential reason for doing a VACUUM FULL is a |
| 1526 |
* imminent or actual anti-wraparound shutdown. So, now that we can |
| 1527 |
* access the new relation using its indices, update relfrozenxid. |
| 1528 |
* pg_class doesn't have a toast relation, so we don't need to update the |
| 1529 |
* corresponding toast relation. Not that there's little point moving all |
| 1530 |
* relfrozenxid updates here since swap_relation_files() needs to write to |
| 1531 |
* pg_class for non-mapped relations anyway. |
| 1532 |
*/ |
| 1533 |
if (OIDOldHeap == RelationRelationId) |
| 1534 |
{ |
| 1535 |
Relation relRelation; |
| 1536 |
HeapTuple reltup; |
| 1537 |
Form_pg_class relform; |
| 1538 |
|
| 1539 |
relRelation = table_open(RelationRelationId, RowExclusiveLock); |
| 1540 |
|
| 1541 |
reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(OIDOldHeap)); |
| 1542 |
if (!HeapTupleIsValid(reltup)) |
| 1543 |
elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap); |
| 1544 |
relform = (Form_pg_class) GETSTRUCT(reltup); |
| 1545 |
|
| 1546 |
relform->relfrozenxid = frozenXid; |
| 1547 |
relform->relminmxid = cutoffMulti; |
| 1548 |
|
| 1549 |
CatalogTupleUpdate(relRelation, &reltup->t_self, reltup); |
| 1550 |
|
| 1551 |
table_close(relRelation, RowExclusiveLock); |
| 1552 |
} |
| 1553 |
|
| 1554 |
/* Destroy new heap with old filenumber */ |
| 1555 |
object.classId = RelationRelationId; |
| 1556 |
object.objectId = OIDNewHeap; |
| 1557 |
object.objectSubId = 0; |
| 1558 |
|
| 1559 |
/* |
| 1560 |
* The new relation is local to our transaction and we know nothing |
| 1561 |
* depends on it, so DROP_RESTRICT should be OK. |
| 1562 |
*/ |
| 1563 |
performDeletion(&object, DROP_RESTRICT, PERFORM_DELETION_INTERNAL); |
| 1564 |
|
| 1565 |
/* performDeletion does CommandCounterIncrement at end */ |
| 1566 |
|
| 1567 |
/* |
| 1568 |
* Now we must remove any relation mapping entries that we set up for the |
| 1569 |
* transient table, as well as its toast table and toast index if any. If |
| 1570 |
* we fail to do this before commit, the relmapper will complain about new |
| 1571 |
* permanent map entries being added post-bootstrap. |
| 1572 |
*/ |
| 1573 |
for (i = 0; OidIsValid(mapped_tables[i]); i++) |
| 1574 |
RelationMapRemoveMapping(mapped_tables[i]); |
| 1575 |
|
| 1576 |
/* |
| 1577 |
* At this point, everything is kosher except that, if we did toast swap |
| 1578 |
* by links, the toast table's name corresponds to the transient table. |
| 1579 |
* The name is irrelevant to the backend because it's referenced by OID, |
| 1580 |
* but users looking at the catalogs could be confused. Rename it to |
| 1581 |
* prevent this problem. |
| 1582 |
* |
| 1583 |
* Note no lock required on the relation, because we already hold an |
| 1584 |
* exclusive lock on it. |
| 1585 |
*/ |
| 1586 |
if (!swap_toast_by_content) |
| 1587 |
{ |
| 1588 |
Relation newrel; |
| 1589 |
|
| 1590 |
newrel = table_open(OIDOldHeap, NoLock); |
| 1591 |
if (OidIsValid(newrel->rd_rel->reltoastrelid)) |
| 1592 |
{ |
| 1593 |
Oid toastidx; |
| 1594 |
char NewToastName[NAMEDATALEN]; |
| 1595 |
|
| 1596 |
/* Get the associated valid index to be renamed */ |
| 1597 |
toastidx = toast_get_valid_index(newrel->rd_rel->reltoastrelid, |
| 1598 |
NoLock); |
| 1599 |
|
| 1600 |
/* rename the toast table ... */ |
| 1601 |
snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u", |
| 1602 |
OIDOldHeap); |
| 1603 |
RenameRelationInternal(newrel->rd_rel->reltoastrelid, |
| 1604 |
NewToastName, true, false); |
| 1605 |
|
| 1606 |
/* ... and its valid index too. */ |
| 1607 |
snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u_index", |
| 1608 |
OIDOldHeap); |
| 1609 |
|
| 1610 |
RenameRelationInternal(toastidx, |
| 1611 |
NewToastName, true, true); |
| 1612 |
|
| 1613 |
/* |
| 1614 |
* Reset the relrewrite for the toast. The command-counter |
| 1615 |
* increment is required here as we are about to update the tuple |
| 1616 |
* that is updated as part of RenameRelationInternal. |
| 1617 |
*/ |
| 1618 |
CommandCounterIncrement(); |
| 1619 |
ResetRelRewrite(newrel->rd_rel->reltoastrelid); |
| 1620 |
} |
| 1621 |
relation_close(newrel, NoLock); |
| 1622 |
} |
| 1623 |
|
| 1624 |
/* if it's not a catalog table, clear any missing attribute settings */ |
| 1625 |
if (!is_system_catalog) |
| 1626 |
{ |
| 1627 |
Relation newrel; |
| 1628 |
|
| 1629 |
newrel = table_open(OIDOldHeap, NoLock); |
| 1630 |
RelationClearMissing(newrel); |
| 1631 |
relation_close(newrel, NoLock); |
| 1632 |
} |
| 1633 |
} |
| 1634 |
|
| 1635 |
|
| 1636 |
/* |
| 1637 |
* Get a list of tables that the current user has privileges on and |
| 1638 |
* have indisclustered set. Return the list in a List * of RelToCluster |
| 1639 |
* (stored in the specified memory context), each one giving the tableOid |
| 1640 |
* and the indexOid on which the table is already clustered. |
| 1641 |
*/ |
| 1642 |
static List * |
| 1643 |
get_tables_to_cluster(MemoryContext cluster_context) |
| 1644 |
{ |
| 1645 |
Relation indRelation; |
| 1646 |
TableScanDesc scan; |
| 1647 |
ScanKeyData entry; |
| 1648 |
HeapTuple indexTuple; |
| 1649 |
Form_pg_index index; |
| 1650 |
MemoryContext old_context; |
| 1651 |
List *rtcs = NIL; |
| 1652 |
|
| 1653 |
/* |
| 1654 |
* Get all indexes that have indisclustered set and that the current user |
| 1655 |
* has the appropriate privileges for. |
| 1656 |
*/ |
| 1657 |
indRelation = table_open(IndexRelationId, AccessShareLock); |
| 1658 |
ScanKeyInit(&entry, |
| 1659 |
Anum_pg_index_indisclustered, |
| 1660 |
BTEqualStrategyNumber, F_BOOLEQ, |
| 1661 |
BoolGetDatum(true)); |
| 1662 |
scan = table_beginscan_catalog(indRelation, 1, &entry); |
| 1663 |
while ((indexTuple = heap_getnext(scan, ForwardScanDirection)) != NULL) |
| 1664 |
{ |
| 1665 |
RelToCluster *rtc; |
| 1666 |
|
| 1667 |
index = (Form_pg_index) GETSTRUCT(indexTuple); |
| 1668 |
|
| 1669 |
if (!cluster_is_permitted_for_relation(index->indrelid, GetUserId())) |
| 1670 |
continue; |
| 1671 |
|
| 1672 |
/* Use a permanent memory context for the result list */ |
| 1673 |
old_context = MemoryContextSwitchTo(cluster_context); |
| 1674 |
|
| 1675 |
rtc = (RelToCluster *) palloc(sizeof(RelToCluster)); |
| 1676 |
rtc->tableOid = index->indrelid; |
| 1677 |
rtc->indexOid = index->indexrelid; |
| 1678 |
rtcs = lappend(rtcs, rtc); |
| 1679 |
|
| 1680 |
MemoryContextSwitchTo(old_context); |
| 1681 |
} |
| 1682 |
table_endscan(scan); |
| 1683 |
|
| 1684 |
relation_close(indRelation, AccessShareLock); |
| 1685 |
|
| 1686 |
return rtcs; |
| 1687 |
} |
| 1688 |
|
| 1689 |
/* |
| 1690 |
* Given an index on a partitioned table, return a list of RelToCluster for |
| 1691 |
* all the children leaves tables/indexes. |
| 1692 |
* |
| 1693 |
* Like expand_vacuum_rel, but here caller must hold AccessExclusiveLock |
| 1694 |
* on the table containing the index. |
| 1695 |
*/ |
| 1696 |
static List * |
| 1697 |
get_tables_to_cluster_partitioned(MemoryContext cluster_context, Oid indexOid) |
| 1698 |
{ |
| 1699 |
List *inhoids; |
| 1700 |
ListCell *lc; |
| 1701 |
List *rtcs = NIL; |
| 1702 |
MemoryContext old_context; |
| 1703 |
|
| 1704 |
/* Do not lock the children until they're processed */ |
| 1705 |
inhoids = find_all_inheritors(indexOid, NoLock, NULL); |
| 1706 |
|
| 1707 |
foreach(lc, inhoids) |
| 1708 |
{ |
| 1709 |
Oid indexrelid = lfirst_oid(lc); |
| 1710 |
Oid relid = IndexGetRelation(indexrelid, false); |
| 1711 |
RelToCluster *rtc; |
| 1712 |
|
| 1713 |
/* consider only leaf indexes */ |
| 1714 |
if (get_rel_relkind(indexrelid) != RELKIND_INDEX) |
| 1715 |
continue; |
| 1716 |
|
| 1717 |
/* |
| 1718 |
* It's possible that the user does not have privileges to CLUSTER the |
| 1719 |
* leaf partition despite having such privileges on the partitioned |
| 1720 |
* table. We skip any partitions which the user is not permitted to |
| 1721 |
* CLUSTER. |
| 1722 |
*/ |
| 1723 |
if (!cluster_is_permitted_for_relation(relid, GetUserId())) |
| 1724 |
continue; |
| 1725 |
|
| 1726 |
/* Use a permanent memory context for the result list */ |
| 1727 |
old_context = MemoryContextSwitchTo(cluster_context); |
| 1728 |
|
| 1729 |
rtc = (RelToCluster *) palloc(sizeof(RelToCluster)); |
| 1730 |
rtc->tableOid = relid; |
| 1731 |
rtc->indexOid = indexrelid; |
| 1732 |
rtcs = lappend(rtcs, rtc); |
| 1733 |
|
| 1734 |
MemoryContextSwitchTo(old_context); |
| 1735 |
} |
| 1736 |
|
| 1737 |
return rtcs; |
| 1738 |
} |
| 1739 |
|
| 1740 |
/* |
| 1741 |
* Return whether userid has privileges to CLUSTER relid. If not, this |
| 1742 |
* function emits a WARNING. |
| 1743 |
*/ |
| 1744 |
static bool |
| 1745 |
cluster_is_permitted_for_relation(Oid relid, Oid userid) |
| 1746 |
{ |
| 1747 |
if (pg_class_aclcheck(relid, userid, ACL_MAINTAIN) == ACLCHECK_OK) |
| 1748 |
return true; |
| 1749 |
|
| 1750 |
ereport(WARNING, |
| 1751 |
(errmsg("permission denied to cluster \"%s\", skipping it", |
| 1752 |
get_rel_name(relid)))); |
| 1753 |
return false; |
| 1754 |
} |