PostgreSQL Source Code  git master
cluster.c File Reference
#include "postgres.h"
#include "access/amapi.h"
#include "access/heapam.h"
#include "access/multixact.h"
#include "access/relscan.h"
#include "access/tableam.h"
#include "access/toast_internals.h"
#include "access/transam.h"
#include "access/xact.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
#include "catalog/index.h"
#include "catalog/namespace.h"
#include "catalog/objectaccess.h"
#include "catalog/pg_am.h"
#include "catalog/pg_database.h"
#include "catalog/pg_inherits.h"
#include "catalog/toasting.h"
#include "commands/cluster.h"
#include "commands/defrem.h"
#include "commands/progress.h"
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
#include "utils/inval.h"
#include "utils/lsyscache.h"
#include "utils/memutils.h"
#include "utils/pg_rusage.h"
#include "utils/relmapper.h"
#include "utils/snapmgr.h"
#include "utils/syscache.h"
Include dependency graph for cluster.c:

Go to the source code of this file.

Data Structures

struct  RelToCluster
 

Functions

static void cluster_multiple_rels (List *rtcs, ClusterParams *params)
 
static void rebuild_relation (Relation OldHeap, Oid indexOid, bool verbose)
 
static void copy_table_data (Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex, bool verbose, bool *pSwapToastByContent, TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
 
static Listget_tables_to_cluster (MemoryContext cluster_context)
 
static Listget_tables_to_cluster_partitioned (MemoryContext cluster_context, Oid indexOid)
 
static bool cluster_is_permitted_for_relation (Oid relid, Oid userid)
 
void cluster (ParseState *pstate, ClusterStmt *stmt, bool isTopLevel)
 
void cluster_rel (Oid tableOid, Oid indexOid, ClusterParams *params)
 
void check_index_is_clusterable (Relation OldHeap, Oid indexOid, LOCKMODE lockmode)
 
void mark_index_clustered (Relation rel, Oid indexOid, bool is_internal)
 
Oid make_new_heap (Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod, char relpersistence, LOCKMODE lockmode)
 
static void swap_relation_files (Oid r1, Oid r2, bool target_is_pg_class, bool swap_toast_by_content, bool is_internal, TransactionId frozenXid, MultiXactId cutoffMulti, Oid *mapped_tables)
 
void finish_heap_swap (Oid OIDOldHeap, Oid OIDNewHeap, bool is_system_catalog, bool swap_toast_by_content, bool check_constraints, bool is_internal, TransactionId frozenXid, MultiXactId cutoffMulti, char newrelpersistence)
 

Function Documentation

◆ check_index_is_clusterable()

void check_index_is_clusterable ( Relation  OldHeap,
Oid  indexOid,
LOCKMODE  lockmode 
)

Definition at line 500 of file cluster.c.

501 {
502  Relation OldIndex;
503 
504  OldIndex = index_open(indexOid, lockmode);
505 
506  /*
507  * Check that index is in fact an index on the given relation
508  */
509  if (OldIndex->rd_index == NULL ||
510  OldIndex->rd_index->indrelid != RelationGetRelid(OldHeap))
511  ereport(ERROR,
512  (errcode(ERRCODE_WRONG_OBJECT_TYPE),
513  errmsg("\"%s\" is not an index for table \"%s\"",
514  RelationGetRelationName(OldIndex),
515  RelationGetRelationName(OldHeap))));
516 
517  /* Index AM must allow clustering */
518  if (!OldIndex->rd_indam->amclusterable)
519  ereport(ERROR,
520  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
521  errmsg("cannot cluster on index \"%s\" because access method does not support clustering",
522  RelationGetRelationName(OldIndex))));
523 
524  /*
525  * Disallow clustering on incomplete indexes (those that might not index
526  * every row of the relation). We could relax this by making a separate
527  * seqscan pass over the table to copy the missing rows, but that seems
528  * expensive and tedious.
529  */
530  if (!heap_attisnull(OldIndex->rd_indextuple, Anum_pg_index_indpred, NULL))
531  ereport(ERROR,
532  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
533  errmsg("cannot cluster on partial index \"%s\"",
534  RelationGetRelationName(OldIndex))));
535 
536  /*
537  * Disallow if index is left over from a failed CREATE INDEX CONCURRENTLY;
538  * it might well not contain entries for every heap row, or might not even
539  * be internally consistent. (But note that we don't check indcheckxmin;
540  * the worst consequence of following broken HOT chains would be that we
541  * might put recently-dead tuples out-of-order in the new table, and there
542  * is little harm in that.)
543  */
544  if (!OldIndex->rd_index->indisvalid)
545  ereport(ERROR,
546  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
547  errmsg("cannot cluster on invalid index \"%s\"",
548  RelationGetRelationName(OldIndex))));
549 
550  /* Drop relcache refcnt on OldIndex, but keep lock */
551  index_close(OldIndex, NoLock);
552 }
int errcode(int sqlerrcode)
Definition: elog.c:857
int errmsg(const char *fmt,...)
Definition: elog.c:1070
#define ERROR
Definition: elog.h:39
#define ereport(elevel,...)
Definition: elog.h:149
bool heap_attisnull(HeapTuple tup, int attnum, TupleDesc tupleDesc)
Definition: heaptuple.c:455
void index_close(Relation relation, LOCKMODE lockmode)
Definition: indexam.c:177
Relation index_open(Oid relationId, LOCKMODE lockmode)
Definition: indexam.c:133
#define NoLock
Definition: lockdefs.h:34
#define RelationGetRelid(relation)
Definition: rel.h:505
#define RelationGetRelationName(relation)
Definition: rel.h:539
bool amclusterable
Definition: amapi.h:242
struct IndexAmRoutine * rd_indam
Definition: rel.h:206
struct HeapTupleData * rd_indextuple
Definition: rel.h:194
Form_pg_index rd_index
Definition: rel.h:192

References IndexAmRoutine::amclusterable, ereport, errcode(), errmsg(), ERROR, heap_attisnull(), index_close(), index_open(), NoLock, RelationData::rd_indam, RelationData::rd_index, RelationData::rd_indextuple, RelationGetRelationName, and RelationGetRelid.

Referenced by ATExecClusterOn(), cluster(), and cluster_rel().

◆ cluster()

void cluster ( ParseState pstate,
ClusterStmt stmt,
bool  isTopLevel 
)

Definition at line 108 of file cluster.c.

109 {
110  ListCell *lc;
111  ClusterParams params = {0};
112  bool verbose = false;
113  Relation rel = NULL;
114  Oid indexOid = InvalidOid;
115  MemoryContext cluster_context;
116  List *rtcs;
117 
118  /* Parse option list */
119  foreach(lc, stmt->params)
120  {
121  DefElem *opt = (DefElem *) lfirst(lc);
122 
123  if (strcmp(opt->defname, "verbose") == 0)
124  verbose = defGetBoolean(opt);
125  else
126  ereport(ERROR,
127  (errcode(ERRCODE_SYNTAX_ERROR),
128  errmsg("unrecognized CLUSTER option \"%s\"",
129  opt->defname),
130  parser_errposition(pstate, opt->location)));
131  }
132 
133  params.options = (verbose ? CLUOPT_VERBOSE : 0);
134 
135  if (stmt->relation != NULL)
136  {
137  /* This is the single-relation case. */
138  Oid tableOid;
139 
140  /*
141  * Find, lock, and check permissions on the table. We obtain
142  * AccessExclusiveLock right away to avoid lock-upgrade hazard in the
143  * single-transaction case.
144  */
145  tableOid = RangeVarGetRelidExtended(stmt->relation,
147  0,
149  NULL);
150  rel = table_open(tableOid, NoLock);
151 
152  /*
153  * Reject clustering a remote temp table ... their local buffer
154  * manager is not going to cope.
155  */
156  if (RELATION_IS_OTHER_TEMP(rel))
157  ereport(ERROR,
158  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
159  errmsg("cannot cluster temporary tables of other sessions")));
160 
161  if (stmt->indexname == NULL)
162  {
163  ListCell *index;
164 
165  /* We need to find the index that has indisclustered set. */
166  foreach(index, RelationGetIndexList(rel))
167  {
168  indexOid = lfirst_oid(index);
169  if (get_index_isclustered(indexOid))
170  break;
171  indexOid = InvalidOid;
172  }
173 
174  if (!OidIsValid(indexOid))
175  ereport(ERROR,
176  (errcode(ERRCODE_UNDEFINED_OBJECT),
177  errmsg("there is no previously clustered index for table \"%s\"",
178  stmt->relation->relname)));
179  }
180  else
181  {
182  /*
183  * The index is expected to be in the same namespace as the
184  * relation.
185  */
186  indexOid = get_relname_relid(stmt->indexname,
187  rel->rd_rel->relnamespace);
188  if (!OidIsValid(indexOid))
189  ereport(ERROR,
190  (errcode(ERRCODE_UNDEFINED_OBJECT),
191  errmsg("index \"%s\" for table \"%s\" does not exist",
192  stmt->indexname, stmt->relation->relname)));
193  }
194 
195  if (rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
196  {
197  /* close relation, keep lock till commit */
198  table_close(rel, NoLock);
199 
200  /* Do the job. */
201  cluster_rel(tableOid, indexOid, &params);
202 
203  return;
204  }
205  }
206 
207  /*
208  * By here, we know we are in a multi-table situation. In order to avoid
209  * holding locks for too long, we want to process each table in its own
210  * transaction. This forces us to disallow running inside a user
211  * transaction block.
212  */
213  PreventInTransactionBlock(isTopLevel, "CLUSTER");
214 
215  /* Also, we need a memory context to hold our list of relations */
216  cluster_context = AllocSetContextCreate(PortalContext,
217  "Cluster",
219 
220  /*
221  * Either we're processing a partitioned table, or we were not given any
222  * table name at all. In either case, obtain a list of relations to
223  * process.
224  *
225  * In the former case, an index name must have been given, so we don't
226  * need to recheck its "indisclustered" bit, but we have to check that it
227  * is an index that we can cluster on. In the latter case, we set the
228  * option bit to have indisclustered verified.
229  *
230  * Rechecking the relation itself is necessary here in all cases.
231  */
232  params.options |= CLUOPT_RECHECK;
233  if (rel != NULL)
234  {
235  Assert(rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE);
237  rtcs = get_tables_to_cluster_partitioned(cluster_context, indexOid);
238 
239  /* close relation, releasing lock on parent table */
241  }
242  else
243  {
244  rtcs = get_tables_to_cluster(cluster_context);
246  }
247 
248  /* Do the job. */
249  cluster_multiple_rels(rtcs, &params);
250 
251  /* Start a new transaction for the cleanup work. */
253 
254  /* Clean up working storage */
255  MemoryContextDelete(cluster_context);
256 }
#define Assert(condition)
Definition: c.h:858
#define OidIsValid(objectId)
Definition: c.h:775
void cluster_rel(Oid tableOid, Oid indexOid, ClusterParams *params)
Definition: cluster.c:311
void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode)
Definition: cluster.c:500
static List * get_tables_to_cluster(MemoryContext cluster_context)
Definition: cluster.c:1636
static List * get_tables_to_cluster_partitioned(MemoryContext cluster_context, Oid indexOid)
Definition: cluster.c:1690
static void cluster_multiple_rels(List *rtcs, ClusterParams *params)
Definition: cluster.c:266
#define CLUOPT_VERBOSE
Definition: cluster.h:23
#define CLUOPT_RECHECK_ISCLUSTERED
Definition: cluster.h:25
#define CLUOPT_RECHECK
Definition: cluster.h:24
bool defGetBoolean(DefElem *def)
Definition: define.c:107
#define stmt
Definition: indent_codes.h:59
int verbose
#define AccessExclusiveLock
Definition: lockdefs.h:43
#define AccessShareLock
Definition: lockdefs.h:36
bool get_index_isclustered(Oid index_oid)
Definition: lsyscache.c:3601
Oid get_relname_relid(const char *relname, Oid relnamespace)
Definition: lsyscache.c:1885
void MemoryContextDelete(MemoryContext context)
Definition: mcxt.c:454
MemoryContext PortalContext
Definition: mcxt.c:158
#define AllocSetContextCreate
Definition: memutils.h:129
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:160
Oid RangeVarGetRelidExtended(const RangeVar *relation, LOCKMODE lockmode, uint32 flags, RangeVarGetRelidCallback callback, void *callback_arg)
Definition: namespace.c:426
int parser_errposition(ParseState *pstate, int location)
Definition: parse_node.c:106
#define lfirst(lc)
Definition: pg_list.h:172
#define lfirst_oid(lc)
Definition: pg_list.h:174
#define InvalidOid
Definition: postgres_ext.h:36
unsigned int Oid
Definition: postgres_ext.h:31
#define RELATION_IS_OTHER_TEMP(relation)
Definition: rel.h:658
List * RelationGetIndexList(Relation relation)
Definition: relcache.c:4760
bits32 options
Definition: cluster.h:30
char * defname
Definition: parsenodes.h:815
ParseLoc location
Definition: parsenodes.h:819
Definition: pg_list.h:54
Form_pg_class rd_rel
Definition: rel.h:111
Definition: type.h:95
void table_close(Relation relation, LOCKMODE lockmode)
Definition: table.c:126
Relation table_open(Oid relationId, LOCKMODE lockmode)
Definition: table.c:40
void RangeVarCallbackMaintainsTable(const RangeVar *relation, Oid relId, Oid oldRelId, void *arg)
Definition: tablecmds.c:17550
void PreventInTransactionBlock(bool isTopLevel, const char *stmtType)
Definition: xact.c:3584
void StartTransactionCommand(void)
Definition: xact.c:2995

References AccessExclusiveLock, AccessShareLock, ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate, Assert, check_index_is_clusterable(), CLUOPT_RECHECK, CLUOPT_RECHECK_ISCLUSTERED, CLUOPT_VERBOSE, cluster_multiple_rels(), cluster_rel(), defGetBoolean(), DefElem::defname, ereport, errcode(), errmsg(), ERROR, get_index_isclustered(), get_relname_relid(), get_tables_to_cluster(), get_tables_to_cluster_partitioned(), InvalidOid, lfirst, lfirst_oid, DefElem::location, MemoryContextDelete(), NoLock, OidIsValid, ClusterParams::options, parser_errposition(), PortalContext, PreventInTransactionBlock(), RangeVarCallbackMaintainsTable(), RangeVarGetRelidExtended(), RelationData::rd_rel, RELATION_IS_OTHER_TEMP, RelationGetIndexList(), StartTransactionCommand(), stmt, table_close(), table_open(), and verbose.

Referenced by adjust_data_dir(), check_bin_dir(), check_data_dir(), check_for_data_types_usage(), check_for_incompatible_polymorphics(), check_for_isn_and_int8_passing_mismatch(), check_for_pg_role_prefix(), check_for_prepared_transactions(), check_for_tables_with_oids(), check_for_user_defined_encoding_conversions(), check_for_user_defined_postfix_ops(), check_is_install_user(), check_proper_datallowconn(), cluster_conn_opts(), connectToServer(), get_bin_version(), get_control_data(), get_db_conn(), get_db_infos(), get_db_rel_and_slot_infos(), get_major_server_version(), get_rel_infos(), get_sock_dir(), get_template0_info(), jsonb_9_4_check_applicable(), old_9_6_invalidate_hash_indexes(), report_extension_updates(), set_tablespace_directory_suffix(), standard_ProcessUtility(), start_postmaster(), and stop_postmaster().

◆ cluster_is_permitted_for_relation()

static bool cluster_is_permitted_for_relation ( Oid  relid,
Oid  userid 
)
static

Definition at line 1738 of file cluster.c.

1739 {
1740  if (pg_class_aclcheck(relid, userid, ACL_MAINTAIN) == ACLCHECK_OK)
1741  return true;
1742 
1743  ereport(WARNING,
1744  (errmsg("permission denied to cluster \"%s\", skipping it",
1745  get_rel_name(relid))));
1746  return false;
1747 }
@ ACLCHECK_OK
Definition: acl.h:183
AclResult pg_class_aclcheck(Oid table_oid, Oid roleid, AclMode mode)
Definition: aclchk.c:4093
#define WARNING
Definition: elog.h:36
char * get_rel_name(Oid relid)
Definition: lsyscache.c:1928
#define ACL_MAINTAIN
Definition: parsenodes.h:90

References ACL_MAINTAIN, ACLCHECK_OK, ereport, errmsg(), get_rel_name(), pg_class_aclcheck(), and WARNING.

Referenced by cluster_rel(), get_tables_to_cluster(), and get_tables_to_cluster_partitioned().

◆ cluster_multiple_rels()

static void cluster_multiple_rels ( List rtcs,
ClusterParams params 
)
static

Definition at line 266 of file cluster.c.

267 {
268  ListCell *lc;
269 
270  /* Commit to get out of starting transaction */
273 
274  /* Cluster the tables, each in a separate transaction */
275  foreach(lc, rtcs)
276  {
277  RelToCluster *rtc = (RelToCluster *) lfirst(lc);
278 
279  /* Start a new transaction for each relation. */
281 
282  /* functions in indexes may want a snapshot set */
284 
285  /* Do the job. */
286  cluster_rel(rtc->tableOid, rtc->indexOid, params);
287 
290  }
291 }
Snapshot GetTransactionSnapshot(void)
Definition: snapmgr.c:216
void PushActiveSnapshot(Snapshot snapshot)
Definition: snapmgr.c:648
void PopActiveSnapshot(void)
Definition: snapmgr.c:743
Oid indexOid
Definition: cluster.c:68
Oid tableOid
Definition: cluster.c:67
void CommitTransactionCommand(void)
Definition: xact.c:3093

References cluster_rel(), CommitTransactionCommand(), GetTransactionSnapshot(), RelToCluster::indexOid, lfirst, PopActiveSnapshot(), PushActiveSnapshot(), StartTransactionCommand(), and RelToCluster::tableOid.

Referenced by cluster().

◆ cluster_rel()

void cluster_rel ( Oid  tableOid,
Oid  indexOid,
ClusterParams params 
)

Definition at line 311 of file cluster.c.

312 {
313  Relation 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 
320  /* Check for user-requested abort. */
322 
324  if (OidIsValid(indexOid))
327  else
330 
331  /*
332  * We grab exclusive access to the target rel and index for the duration
333  * of the transaction. (This is redundant for the single-transaction
334  * case, since cluster() already did it.) The index lock is taken inside
335  * check_index_is_clusterable.
336  */
337  OldHeap = try_relation_open(tableOid, AccessExclusiveLock);
338 
339  /* If the table has gone away, we can skip processing it */
340  if (!OldHeap)
341  {
343  return;
344  }
345 
346  /*
347  * Switch to the table owner's userid, so that any index functions are run
348  * as that user. Also lock down security-restricted operations and
349  * arrange to make GUC variable changes local to this command.
350  */
351  GetUserIdAndSecContext(&save_userid, &save_sec_context);
352  SetUserIdAndSecContext(OldHeap->rd_rel->relowner,
353  save_sec_context | SECURITY_RESTRICTED_OPERATION);
354  save_nestlevel = NewGUCNestLevel();
356 
357  /*
358  * Since we may open a new transaction for each relation, we have to check
359  * that the relation still is what we think it is.
360  *
361  * If this is a single-transaction CLUSTER, we can skip these tests. We
362  * *must* skip the one on indisclustered since it would reject an attempt
363  * to cluster a not-previously-clustered index.
364  */
365  if (recheck)
366  {
367  /* Check that the user still has privileges for the relation */
368  if (!cluster_is_permitted_for_relation(tableOid, save_userid))
369  {
371  goto out;
372  }
373 
374  /*
375  * Silently skip a temp table for a remote session. Only doing this
376  * check in the "recheck" case is appropriate (which currently means
377  * somebody is executing a database-wide CLUSTER or on a partitioned
378  * table), because there is another check in cluster() which will stop
379  * any attempt to cluster remote temp tables by name. There is
380  * another check in cluster_rel which is redundant, but we leave it
381  * for extra safety.
382  */
383  if (RELATION_IS_OTHER_TEMP(OldHeap))
384  {
386  goto out;
387  }
388 
389  if (OidIsValid(indexOid))
390  {
391  /*
392  * Check that the index still exists
393  */
394  if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(indexOid)))
395  {
397  goto out;
398  }
399 
400  /*
401  * Check that the index is still the one with indisclustered set,
402  * if needed.
403  */
404  if ((params->options & CLUOPT_RECHECK_ISCLUSTERED) != 0 &&
405  !get_index_isclustered(indexOid))
406  {
408  goto out;
409  }
410  }
411  }
412 
413  /*
414  * We allow VACUUM FULL, but not CLUSTER, on shared catalogs. CLUSTER
415  * would work in most respects, but the index would only get marked as
416  * indisclustered in the current database, leading to unexpected behavior
417  * if CLUSTER were later invoked in another database.
418  */
419  if (OidIsValid(indexOid) && OldHeap->rd_rel->relisshared)
420  ereport(ERROR,
421  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
422  errmsg("cannot cluster a shared catalog")));
423 
424  /*
425  * Don't process temp tables of other backends ... their local buffer
426  * manager is not going to cope.
427  */
428  if (RELATION_IS_OTHER_TEMP(OldHeap))
429  {
430  if (OidIsValid(indexOid))
431  ereport(ERROR,
432  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
433  errmsg("cannot cluster temporary tables of other sessions")));
434  else
435  ereport(ERROR,
436  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
437  errmsg("cannot vacuum temporary tables of other sessions")));
438  }
439 
440  /*
441  * Also check for active uses of the relation in the current transaction,
442  * including open scans and pending AFTER trigger events.
443  */
444  CheckTableNotInUse(OldHeap, OidIsValid(indexOid) ? "CLUSTER" : "VACUUM");
445 
446  /* Check heap and index are valid to cluster on */
447  if (OidIsValid(indexOid))
449 
450  /*
451  * Quietly ignore the request if this is a materialized view which has not
452  * been populated from its query. No harm is done because there is no data
453  * to deal with, and we don't want to throw an error if this is part of a
454  * multi-relation request -- for example, CLUSTER was run on the entire
455  * database.
456  */
457  if (OldHeap->rd_rel->relkind == RELKIND_MATVIEW &&
458  !RelationIsPopulated(OldHeap))
459  {
461  goto out;
462  }
463 
464  Assert(OldHeap->rd_rel->relkind == RELKIND_RELATION ||
465  OldHeap->rd_rel->relkind == RELKIND_MATVIEW ||
466  OldHeap->rd_rel->relkind == RELKIND_TOASTVALUE);
467 
468  /*
469  * All predicate locks on the tuples or pages are about to be made
470  * invalid, because we move tuples around. Promote them to relation
471  * locks. Predicate locks on indexes will be promoted when they are
472  * reindexed.
473  */
475 
476  /* rebuild_relation does all the dirty work */
477  rebuild_relation(OldHeap, indexOid, verbose);
478 
479  /* NB: rebuild_relation does table_close() on OldHeap */
480 
481 out:
482  /* Roll back any GUC changes executed by index functions */
483  AtEOXact_GUC(false, save_nestlevel);
484 
485  /* Restore userid and security context */
486  SetUserIdAndSecContext(save_userid, save_sec_context);
487 
489 }
void pgstat_progress_start_command(ProgressCommandType cmdtype, Oid relid)
void pgstat_progress_update_param(int index, int64 val)
void pgstat_progress_end_command(void)
@ PROGRESS_COMMAND_CLUSTER
static void rebuild_relation(Relation OldHeap, Oid indexOid, bool verbose)
Definition: cluster.c:633
static bool cluster_is_permitted_for_relation(Oid relid, Oid userid)
Definition: cluster.c:1738
int NewGUCNestLevel(void)
Definition: guc.c:2237
void RestrictSearchPath(void)
Definition: guc.c:2248
void AtEOXact_GUC(bool isCommit, int nestLevel)
Definition: guc.c:2264
#define SECURITY_RESTRICTED_OPERATION
Definition: miscadmin.h:315
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:122
void GetUserIdAndSecContext(Oid *userid, int *sec_context)
Definition: miscinit.c:635
void SetUserIdAndSecContext(Oid userid, int sec_context)
Definition: miscinit.c:642
static Datum ObjectIdGetDatum(Oid X)
Definition: postgres.h:252
void TransferPredicateLocksToHeapRelation(Relation relation)
Definition: predicate.c:3108
#define PROGRESS_CLUSTER_COMMAND_VACUUM_FULL
Definition: progress.h:77
#define PROGRESS_CLUSTER_COMMAND_CLUSTER
Definition: progress.h:76
#define PROGRESS_CLUSTER_COMMAND
Definition: progress.h:57
#define RelationIsPopulated(relation)
Definition: rel.h:677
void relation_close(Relation relation, LOCKMODE lockmode)
Definition: relation.c:205
Relation try_relation_open(Oid relationId, LOCKMODE lockmode)
Definition: relation.c:88
#define SearchSysCacheExists1(cacheId, key1)
Definition: syscache.h:95
void CheckTableNotInUse(Relation rel, const char *stmt)
Definition: tablecmds.c:4250

References AccessExclusiveLock, Assert, AtEOXact_GUC(), CHECK_FOR_INTERRUPTS, check_index_is_clusterable(), CheckTableNotInUse(), CLUOPT_RECHECK, CLUOPT_RECHECK_ISCLUSTERED, CLUOPT_VERBOSE, cluster_is_permitted_for_relation(), ereport, errcode(), errmsg(), ERROR, get_index_isclustered(), GetUserIdAndSecContext(), NewGUCNestLevel(), ObjectIdGetDatum(), OidIsValid, ClusterParams::options, pgstat_progress_end_command(), pgstat_progress_start_command(), pgstat_progress_update_param(), PROGRESS_CLUSTER_COMMAND, PROGRESS_CLUSTER_COMMAND_CLUSTER, PROGRESS_CLUSTER_COMMAND_VACUUM_FULL, PROGRESS_COMMAND_CLUSTER, RelationData::rd_rel, rebuild_relation(), relation_close(), RELATION_IS_OTHER_TEMP, RelationIsPopulated, RestrictSearchPath(), SearchSysCacheExists1, SECURITY_RESTRICTED_OPERATION, SetUserIdAndSecContext(), TransferPredicateLocksToHeapRelation(), try_relation_open(), and verbose.

Referenced by cluster(), cluster_multiple_rels(), and vacuum_rel().

◆ copy_table_data()

static void copy_table_data ( Oid  OIDNewHeap,
Oid  OIDOldHeap,
Oid  OIDOldIndex,
bool  verbose,
bool pSwapToastByContent,
TransactionId pFreezeXid,
MultiXactId pCutoffMulti 
)
static

Definition at line 814 of file cluster.c.

817 {
818  Relation NewHeap,
819  OldHeap,
820  OldIndex;
821  Relation relRelation;
822  HeapTuple reltup;
823  Form_pg_class relform;
826  VacuumParams params;
827  struct VacuumCutoffs cutoffs;
828  bool use_sort;
829  double num_tuples = 0,
830  tups_vacuumed = 0,
831  tups_recently_dead = 0;
832  BlockNumber num_pages;
833  int elevel = verbose ? INFO : DEBUG2;
834  PGRUsage ru0;
835  char *nspname;
836 
837  pg_rusage_init(&ru0);
838 
839  /*
840  * Open the relations we need.
841  */
842  NewHeap = table_open(OIDNewHeap, AccessExclusiveLock);
843  OldHeap = table_open(OIDOldHeap, AccessExclusiveLock);
844  if (OidIsValid(OIDOldIndex))
845  OldIndex = index_open(OIDOldIndex, AccessExclusiveLock);
846  else
847  OldIndex = NULL;
848 
849  /* Store a copy of the namespace name for logging purposes */
850  nspname = get_namespace_name(RelationGetNamespace(OldHeap));
851 
852  /*
853  * Their tuple descriptors should be exactly alike, but here we only need
854  * assume that they have the same number of columns.
855  */
856  oldTupDesc = RelationGetDescr(OldHeap);
857  newTupDesc = RelationGetDescr(NewHeap);
858  Assert(newTupDesc->natts == oldTupDesc->natts);
859 
860  /*
861  * If the OldHeap has a toast table, get lock on the toast table to keep
862  * it from being vacuumed. This is needed because autovacuum processes
863  * toast tables independently of their main tables, with no lock on the
864  * latter. If an autovacuum were to start on the toast table after we
865  * compute our OldestXmin below, it would use a later OldestXmin, and then
866  * possibly remove as DEAD toast tuples belonging to main tuples we think
867  * are only RECENTLY_DEAD. Then we'd fail while trying to copy those
868  * tuples.
869  *
870  * We don't need to open the toast relation here, just lock it. The lock
871  * will be held till end of transaction.
872  */
873  if (OldHeap->rd_rel->reltoastrelid)
874  LockRelationOid(OldHeap->rd_rel->reltoastrelid, AccessExclusiveLock);
875 
876  /*
877  * If both tables have TOAST tables, perform toast swap by content. It is
878  * possible that the old table has a toast table but the new one doesn't,
879  * if toastable columns have been dropped. In that case we have to do
880  * swap by links. This is okay because swap by content is only essential
881  * for system catalogs, and we don't support schema changes for them.
882  */
883  if (OldHeap->rd_rel->reltoastrelid && NewHeap->rd_rel->reltoastrelid)
884  {
885  *pSwapToastByContent = true;
886 
887  /*
888  * When doing swap by content, any toast pointers written into NewHeap
889  * must use the old toast table's OID, because that's where the toast
890  * data will eventually be found. Set this up by setting rd_toastoid.
891  * This also tells toast_save_datum() to preserve the toast value
892  * OIDs, which we want so as not to invalidate toast pointers in
893  * system catalog caches, and to avoid making multiple copies of a
894  * single toast value.
895  *
896  * Note that we must hold NewHeap open until we are done writing data,
897  * since the relcache will not guarantee to remember this setting once
898  * the relation is closed. Also, this technique depends on the fact
899  * that no one will try to read from the NewHeap until after we've
900  * finished writing it and swapping the rels --- otherwise they could
901  * follow the toast pointers to the wrong place. (It would actually
902  * work for values copied over from the old toast table, but not for
903  * any values that we toast which were previously not toasted.)
904  */
905  NewHeap->rd_toastoid = OldHeap->rd_rel->reltoastrelid;
906  }
907  else
908  *pSwapToastByContent = false;
909 
910  /*
911  * Compute xids used to freeze and weed out dead tuples and multixacts.
912  * Since we're going to rewrite the whole table anyway, there's no reason
913  * not to be aggressive about this.
914  */
915  memset(&params, 0, sizeof(VacuumParams));
916  vacuum_get_cutoffs(OldHeap, &params, &cutoffs);
917 
918  /*
919  * FreezeXid will become the table's new relfrozenxid, and that mustn't go
920  * backwards, so take the max.
921  */
922  {
923  TransactionId relfrozenxid = OldHeap->rd_rel->relfrozenxid;
924 
926  TransactionIdPrecedes(cutoffs.FreezeLimit, relfrozenxid))
927  cutoffs.FreezeLimit = relfrozenxid;
928  }
929 
930  /*
931  * MultiXactCutoff, similarly, shouldn't go backwards either.
932  */
933  {
934  MultiXactId relminmxid = OldHeap->rd_rel->relminmxid;
935 
937  MultiXactIdPrecedes(cutoffs.MultiXactCutoff, relminmxid))
938  cutoffs.MultiXactCutoff = relminmxid;
939  }
940 
941  /*
942  * Decide whether to use an indexscan or seqscan-and-optional-sort to scan
943  * the OldHeap. We know how to use a sort to duplicate the ordering of a
944  * btree index, and will use seqscan-and-sort for that case if the planner
945  * tells us it's cheaper. Otherwise, always indexscan if an index is
946  * provided, else plain seqscan.
947  */
948  if (OldIndex != NULL && OldIndex->rd_rel->relam == BTREE_AM_OID)
949  use_sort = plan_cluster_use_sort(OIDOldHeap, OIDOldIndex);
950  else
951  use_sort = false;
952 
953  /* Log what we're doing */
954  if (OldIndex != NULL && !use_sort)
955  ereport(elevel,
956  (errmsg("clustering \"%s.%s\" using index scan on \"%s\"",
957  nspname,
958  RelationGetRelationName(OldHeap),
959  RelationGetRelationName(OldIndex))));
960  else if (use_sort)
961  ereport(elevel,
962  (errmsg("clustering \"%s.%s\" using sequential scan and sort",
963  nspname,
964  RelationGetRelationName(OldHeap))));
965  else
966  ereport(elevel,
967  (errmsg("vacuuming \"%s.%s\"",
968  nspname,
969  RelationGetRelationName(OldHeap))));
970 
971  /*
972  * Hand off the actual copying to AM specific function, the generic code
973  * cannot know how to deal with visibility across AMs. Note that this
974  * routine is allowed to set FreezeXid / MultiXactCutoff to different
975  * values (e.g. because the AM doesn't use freezing).
976  */
977  table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
978  cutoffs.OldestXmin, &cutoffs.FreezeLimit,
979  &cutoffs.MultiXactCutoff,
980  &num_tuples, &tups_vacuumed,
981  &tups_recently_dead);
982 
983  /* return selected values to caller, get set as relfrozenxid/minmxid */
984  *pFreezeXid = cutoffs.FreezeLimit;
985  *pCutoffMulti = cutoffs.MultiXactCutoff;
986 
987  /* Reset rd_toastoid just to be tidy --- it shouldn't be looked at again */
988  NewHeap->rd_toastoid = InvalidOid;
989 
990  num_pages = RelationGetNumberOfBlocks(NewHeap);
991 
992  /* Log what we did */
993  ereport(elevel,
994  (errmsg("\"%s.%s\": found %.0f removable, %.0f nonremovable row versions in %u pages",
995  nspname,
996  RelationGetRelationName(OldHeap),
997  tups_vacuumed, num_tuples,
998  RelationGetNumberOfBlocks(OldHeap)),
999  errdetail("%.0f dead row versions cannot be removed yet.\n"
1000  "%s.",
1001  tups_recently_dead,
1002  pg_rusage_show(&ru0))));
1003 
1004  if (OldIndex != NULL)
1005  index_close(OldIndex, NoLock);
1006  table_close(OldHeap, NoLock);
1007  table_close(NewHeap, NoLock);
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, ObjectIdGetDatum(OIDNewHeap));
1013  if (!HeapTupleIsValid(reltup))
1014  elog(ERROR, "cache lookup failed for relation %u", OIDNewHeap);
1015  relform = (Form_pg_class) GETSTRUCT(reltup);
1016 
1017  relform->relpages = num_pages;
1018  relform->reltuples = num_tuples;
1019 
1020  /* Don't update the stats for pg_class. See swap_relation_files. */
1021  if (OIDOldHeap != RelationRelationId)
1022  CatalogTupleUpdate(relRelation, &reltup->t_self, reltup);
1023  else
1025 
1026  /* Clean up. */
1027  heap_freetuple(reltup);
1028  table_close(relRelation, RowExclusiveLock);
1029 
1030  /* Make the update visible */
1032 }
uint32 BlockNumber
Definition: block.h:31
#define RelationGetNumberOfBlocks(reln)
Definition: bufmgr.h:277
#define PG_USED_FOR_ASSERTS_ONLY
Definition: c.h:182
TransactionId MultiXactId
Definition: c.h:662
uint32 TransactionId
Definition: c.h:652
int errdetail(const char *fmt,...)
Definition: elog.c:1203
#define DEBUG2
Definition: elog.h:29
#define elog(elevel,...)
Definition: elog.h:224
#define INFO
Definition: elog.h:34
void heap_freetuple(HeapTuple htup)
Definition: heaptuple.c:1434
#define HeapTupleIsValid(tuple)
Definition: htup.h:78
#define GETSTRUCT(TUP)
Definition: htup_details.h:653
void CatalogTupleUpdate(Relation heapRel, ItemPointer otid, HeapTuple tup)
Definition: indexing.c:313
void CacheInvalidateRelcacheByTuple(HeapTuple classTuple)
Definition: inval.c:1396
void LockRelationOid(Oid relid, LOCKMODE lockmode)
Definition: lmgr.c:108
#define RowExclusiveLock
Definition: lockdefs.h:38
char * get_namespace_name(Oid nspid)
Definition: lsyscache.c:3366
bool MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
Definition: multixact.c:3260
#define MultiXactIdIsValid(multi)
Definition: multixact.h:28
FormData_pg_class * Form_pg_class
Definition: pg_class.h:153
const char * pg_rusage_show(const PGRUsage *ru0)
Definition: pg_rusage.c:40
void pg_rusage_init(PGRUsage *ru0)
Definition: pg_rusage.c:27
bool plan_cluster_use_sort(Oid tableOid, Oid indexOid)
Definition: planner.c:6571
#define RelationGetDescr(relation)
Definition: rel.h:531
#define RelationGetNamespace(relation)
Definition: rel.h:546
ItemPointerData t_self
Definition: htup.h:65
Oid rd_toastoid
Definition: rel.h:251
TransactionId relfrozenxid
Definition: vacuum.h:251
MultiXactId relminmxid
Definition: vacuum.h:252
#define SearchSysCacheCopy1(cacheId, key1)
Definition: syscache.h:86
static void table_relation_copy_for_cluster(Relation OldTable, Relation NewTable, Relation OldIndex, bool use_sort, TransactionId OldestXmin, TransactionId *xid_cutoff, MultiXactId *multi_cutoff, double *num_tuples, double *tups_vacuumed, double *tups_recently_dead)
Definition: tableam.h:1688
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:280
#define TransactionIdIsValid(xid)
Definition: transam.h:41
bool vacuum_get_cutoffs(Relation rel, const VacuumParams *params, struct VacuumCutoffs *cutoffs)
Definition: vacuum.c:1072
void CommandCounterIncrement(void)
Definition: xact.c:1097

References AccessExclusiveLock, Assert, CacheInvalidateRelcacheByTuple(), CatalogTupleUpdate(), CommandCounterIncrement(), DEBUG2, elog, ereport, errdetail(), errmsg(), ERROR, VacuumCutoffs::FreezeLimit, get_namespace_name(), GETSTRUCT, heap_freetuple(), HeapTupleIsValid, index_close(), index_open(), INFO, InvalidOid, LockRelationOid(), VacuumCutoffs::MultiXactCutoff, MultiXactIdIsValid, MultiXactIdPrecedes(), NoLock, ObjectIdGetDatum(), OidIsValid, VacuumCutoffs::OldestXmin, pg_rusage_init(), pg_rusage_show(), PG_USED_FOR_ASSERTS_ONLY, plan_cluster_use_sort(), RelationData::rd_rel, RelationData::rd_toastoid, RelationGetDescr, RelationGetNamespace, RelationGetNumberOfBlocks, RelationGetRelationName, VacuumCutoffs::relfrozenxid, VacuumCutoffs::relminmxid, RowExclusiveLock, SearchSysCacheCopy1, HeapTupleData::t_self, table_close(), table_open(), table_relation_copy_for_cluster(), TransactionIdIsValid, TransactionIdPrecedes(), vacuum_get_cutoffs(), and verbose.

Referenced by rebuild_relation().

◆ finish_heap_swap()

void finish_heap_swap ( Oid  OIDOldHeap,
Oid  OIDNewHeap,
bool  is_system_catalog,
bool  swap_toast_by_content,
bool  check_constraints,
bool  is_internal,
TransactionId  frozenXid,
MultiXactId  cutoffMulti,
char  newrelpersistence 
)

Definition at line 1438 of file cluster.c.

1446 {
1447  ObjectAddress object;
1448  Oid mapped_tables[4];
1449  int reindex_flags;
1450  ReindexParams reindex_params = {0};
1451  int i;
1452 
1453  /* Report that we are now swapping relation files */
1456 
1457  /* Zero out possible results from swapped_relation_files */
1458  memset(mapped_tables, 0, sizeof(mapped_tables));
1459 
1460  /*
1461  * Swap the contents of the heap relations (including any toast tables).
1462  * Also set old heap's relfrozenxid to frozenXid.
1463  */
1464  swap_relation_files(OIDOldHeap, OIDNewHeap,
1465  (OIDOldHeap == RelationRelationId),
1466  swap_toast_by_content, is_internal,
1467  frozenXid, cutoffMulti, mapped_tables);
1468 
1469  /*
1470  * If it's a system catalog, queue a sinval message to flush all catcaches
1471  * on the catalog when we reach CommandCounterIncrement.
1472  */
1473  if (is_system_catalog)
1474  CacheInvalidateCatalog(OIDOldHeap);
1475 
1476  /*
1477  * Rebuild each index on the relation (but not the toast table, which is
1478  * all-new at this point). It is important to do this before the DROP
1479  * step because if we are processing a system catalog that will be used
1480  * during DROP, we want to have its indexes available. There is no
1481  * advantage to the other order anyway because this is all transactional,
1482  * so no chance to reclaim disk space before commit. We do not need a
1483  * final CommandCounterIncrement() because reindex_relation does it.
1484  *
1485  * Note: because index_build is called via reindex_relation, it will never
1486  * set indcheckxmin true for the indexes. This is OK even though in some
1487  * sense we are building new indexes rather than rebuilding existing ones,
1488  * because the new heap won't contain any HOT chains at all, let alone
1489  * broken ones, so it can't be necessary to set indcheckxmin.
1490  */
1491  reindex_flags = REINDEX_REL_SUPPRESS_INDEX_USE;
1492  if (check_constraints)
1493  reindex_flags |= REINDEX_REL_CHECK_CONSTRAINTS;
1494 
1495  /*
1496  * Ensure that the indexes have the same persistence as the parent
1497  * relation.
1498  */
1499  if (newrelpersistence == RELPERSISTENCE_UNLOGGED)
1500  reindex_flags |= REINDEX_REL_FORCE_INDEXES_UNLOGGED;
1501  else if (newrelpersistence == RELPERSISTENCE_PERMANENT)
1502  reindex_flags |= REINDEX_REL_FORCE_INDEXES_PERMANENT;
1503 
1504  /* Report that we are now reindexing relations */
1507 
1508  reindex_relation(NULL, OIDOldHeap, reindex_flags, &reindex_params);
1509 
1510  /* Report that we are now doing clean up */
1513 
1514  /*
1515  * If the relation being rebuilt is pg_class, swap_relation_files()
1516  * couldn't update pg_class's own pg_class entry (check comments in
1517  * swap_relation_files()), thus relfrozenxid was not updated. That's
1518  * annoying because a potential reason for doing a VACUUM FULL is a
1519  * imminent or actual anti-wraparound shutdown. So, now that we can
1520  * access the new relation using its indices, update relfrozenxid.
1521  * pg_class doesn't have a toast relation, so we don't need to update the
1522  * corresponding toast relation. Not that there's little point moving all
1523  * relfrozenxid updates here since swap_relation_files() needs to write to
1524  * pg_class for non-mapped relations anyway.
1525  */
1526  if (OIDOldHeap == RelationRelationId)
1527  {
1528  Relation relRelation;
1529  HeapTuple reltup;
1530  Form_pg_class relform;
1531 
1532  relRelation = table_open(RelationRelationId, RowExclusiveLock);
1533 
1534  reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(OIDOldHeap));
1535  if (!HeapTupleIsValid(reltup))
1536  elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap);
1537  relform = (Form_pg_class) GETSTRUCT(reltup);
1538 
1539  relform->relfrozenxid = frozenXid;
1540  relform->relminmxid = cutoffMulti;
1541 
1542  CatalogTupleUpdate(relRelation, &reltup->t_self, reltup);
1543 
1544  table_close(relRelation, RowExclusiveLock);
1545  }
1546 
1547  /* Destroy new heap with old filenumber */
1548  object.classId = RelationRelationId;
1549  object.objectId = OIDNewHeap;
1550  object.objectSubId = 0;
1551 
1552  /*
1553  * The new relation is local to our transaction and we know nothing
1554  * depends on it, so DROP_RESTRICT should be OK.
1555  */
1557 
1558  /* performDeletion does CommandCounterIncrement at end */
1559 
1560  /*
1561  * Now we must remove any relation mapping entries that we set up for the
1562  * transient table, as well as its toast table and toast index if any. If
1563  * we fail to do this before commit, the relmapper will complain about new
1564  * permanent map entries being added post-bootstrap.
1565  */
1566  for (i = 0; OidIsValid(mapped_tables[i]); i++)
1567  RelationMapRemoveMapping(mapped_tables[i]);
1568 
1569  /*
1570  * At this point, everything is kosher except that, if we did toast swap
1571  * by links, the toast table's name corresponds to the transient table.
1572  * The name is irrelevant to the backend because it's referenced by OID,
1573  * but users looking at the catalogs could be confused. Rename it to
1574  * prevent this problem.
1575  *
1576  * Note no lock required on the relation, because we already hold an
1577  * exclusive lock on it.
1578  */
1579  if (!swap_toast_by_content)
1580  {
1581  Relation newrel;
1582 
1583  newrel = table_open(OIDOldHeap, NoLock);
1584  if (OidIsValid(newrel->rd_rel->reltoastrelid))
1585  {
1586  Oid toastidx;
1587  char NewToastName[NAMEDATALEN];
1588 
1589  /* Get the associated valid index to be renamed */
1590  toastidx = toast_get_valid_index(newrel->rd_rel->reltoastrelid,
1591  NoLock);
1592 
1593  /* rename the toast table ... */
1594  snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u",
1595  OIDOldHeap);
1596  RenameRelationInternal(newrel->rd_rel->reltoastrelid,
1597  NewToastName, true, false);
1598 
1599  /* ... and its valid index too. */
1600  snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u_index",
1601  OIDOldHeap);
1602 
1603  RenameRelationInternal(toastidx,
1604  NewToastName, true, true);
1605 
1606  /*
1607  * Reset the relrewrite for the toast. The command-counter
1608  * increment is required here as we are about to update the tuple
1609  * that is updated as part of RenameRelationInternal.
1610  */
1612  ResetRelRewrite(newrel->rd_rel->reltoastrelid);
1613  }
1614  relation_close(newrel, NoLock);
1615  }
1616 
1617  /* if it's not a catalog table, clear any missing attribute settings */
1618  if (!is_system_catalog)
1619  {
1620  Relation newrel;
1621 
1622  newrel = table_open(OIDOldHeap, NoLock);
1623  RelationClearMissing(newrel);
1624  relation_close(newrel, NoLock);
1625  }
1626 }
static void swap_relation_files(Oid r1, Oid r2, bool target_is_pg_class, bool swap_toast_by_content, bool is_internal, TransactionId frozenXid, MultiXactId cutoffMulti, Oid *mapped_tables)
Definition: cluster.c:1061
void performDeletion(const ObjectAddress *object, DropBehavior behavior, int flags)
Definition: dependency.c:273
#define PERFORM_DELETION_INTERNAL
Definition: dependency.h:92
void RelationClearMissing(Relation rel)
Definition: heap.c:1937
bool reindex_relation(const ReindexStmt *stmt, Oid relid, int flags, const ReindexParams *params)
Definition: index.c:3888
#define REINDEX_REL_FORCE_INDEXES_UNLOGGED
Definition: index.h:161
#define REINDEX_REL_SUPPRESS_INDEX_USE
Definition: index.h:159
#define REINDEX_REL_FORCE_INDEXES_PERMANENT
Definition: index.h:162
#define REINDEX_REL_CHECK_CONSTRAINTS
Definition: index.h:160
void CacheInvalidateCatalog(Oid catalogId)
Definition: inval.c:1336
int i
Definition: isn.c:73
@ DROP_RESTRICT
Definition: parsenodes.h:2334
#define NAMEDATALEN
#define snprintf
Definition: port.h:238
#define PROGRESS_CLUSTER_PHASE
Definition: progress.h:58
#define PROGRESS_CLUSTER_PHASE_REBUILD_INDEX
Definition: progress.h:72
#define PROGRESS_CLUSTER_PHASE_FINAL_CLEANUP
Definition: progress.h:73
#define PROGRESS_CLUSTER_PHASE_SWAP_REL_FILES
Definition: progress.h:71
void RelationMapRemoveMapping(Oid relationId)
Definition: relmapper.c:438
void ResetRelRewrite(Oid myrelid)
Definition: tablecmds.c:4197
void RenameRelationInternal(Oid myrelid, const char *newrelname, bool is_internal, bool is_index)
Definition: tablecmds.c:4107
Oid toast_get_valid_index(Oid toastoid, LOCKMODE lock)

References CacheInvalidateCatalog(), CatalogTupleUpdate(), CommandCounterIncrement(), DROP_RESTRICT, elog, ERROR, GETSTRUCT, HeapTupleIsValid, i, NAMEDATALEN, NoLock, ObjectIdGetDatum(), OidIsValid, PERFORM_DELETION_INTERNAL, performDeletion(), pgstat_progress_update_param(), PROGRESS_CLUSTER_PHASE, PROGRESS_CLUSTER_PHASE_FINAL_CLEANUP, PROGRESS_CLUSTER_PHASE_REBUILD_INDEX, PROGRESS_CLUSTER_PHASE_SWAP_REL_FILES, RelationData::rd_rel, REINDEX_REL_CHECK_CONSTRAINTS, REINDEX_REL_FORCE_INDEXES_PERMANENT, REINDEX_REL_FORCE_INDEXES_UNLOGGED, REINDEX_REL_SUPPRESS_INDEX_USE, reindex_relation(), relation_close(), RelationClearMissing(), RelationMapRemoveMapping(), RenameRelationInternal(), ResetRelRewrite(), RowExclusiveLock, SearchSysCacheCopy1, snprintf, swap_relation_files(), HeapTupleData::t_self, table_close(), table_open(), and toast_get_valid_index().

Referenced by ATRewriteTables(), rebuild_relation(), and refresh_by_heap_swap().

◆ get_tables_to_cluster()

static List * get_tables_to_cluster ( MemoryContext  cluster_context)
static

Definition at line 1636 of file cluster.c.

1637 {
1638  Relation indRelation;
1639  TableScanDesc scan;
1640  ScanKeyData entry;
1641  HeapTuple indexTuple;
1643  MemoryContext old_context;
1644  List *rtcs = NIL;
1645 
1646  /*
1647  * Get all indexes that have indisclustered set and that the current user
1648  * has the appropriate privileges for.
1649  */
1650  indRelation = table_open(IndexRelationId, AccessShareLock);
1651  ScanKeyInit(&entry,
1652  Anum_pg_index_indisclustered,
1653  BTEqualStrategyNumber, F_BOOLEQ,
1654  BoolGetDatum(true));
1655  scan = table_beginscan_catalog(indRelation, 1, &entry);
1656  while ((indexTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
1657  {
1658  RelToCluster *rtc;
1659 
1660  index = (Form_pg_index) GETSTRUCT(indexTuple);
1661 
1663  continue;
1664 
1665  /* Use a permanent memory context for the result list */
1666  old_context = MemoryContextSwitchTo(cluster_context);
1667 
1668  rtc = (RelToCluster *) palloc(sizeof(RelToCluster));
1669  rtc->tableOid = index->indrelid;
1670  rtc->indexOid = index->indexrelid;
1671  rtcs = lappend(rtcs, rtc);
1672 
1673  MemoryContextSwitchTo(old_context);
1674  }
1675  table_endscan(scan);
1676 
1677  relation_close(indRelation, AccessShareLock);
1678 
1679  return rtcs;
1680 }
HeapTuple heap_getnext(TableScanDesc sscan, ScanDirection direction)
Definition: heapam.c:1251
List * lappend(List *list, void *datum)
Definition: list.c:339
void * palloc(Size size)
Definition: mcxt.c:1316
Oid GetUserId(void)
Definition: miscinit.c:514
FormData_pg_index * Form_pg_index
Definition: pg_index.h:70
#define NIL
Definition: pg_list.h:68
static Datum BoolGetDatum(bool X)
Definition: postgres.h:102
MemoryContextSwitchTo(old_ctx)
void ScanKeyInit(ScanKey entry, AttrNumber attributeNumber, StrategyNumber strategy, RegProcedure procedure, Datum argument)
Definition: scankey.c:76
@ ForwardScanDirection
Definition: sdir.h:28
#define BTEqualStrategyNumber
Definition: stratnum.h:31
TableScanDesc table_beginscan_catalog(Relation relation, int nkeys, struct ScanKeyData *key)
Definition: tableam.c:112
static void table_endscan(TableScanDesc scan)
Definition: tableam.h:1029

References AccessShareLock, BoolGetDatum(), BTEqualStrategyNumber, cluster_is_permitted_for_relation(), ForwardScanDirection, GETSTRUCT, GetUserId(), heap_getnext(), RelToCluster::indexOid, lappend(), MemoryContextSwitchTo(), NIL, palloc(), relation_close(), ScanKeyInit(), table_beginscan_catalog(), table_endscan(), table_open(), and RelToCluster::tableOid.

Referenced by cluster().

◆ get_tables_to_cluster_partitioned()

static List * get_tables_to_cluster_partitioned ( MemoryContext  cluster_context,
Oid  indexOid 
)
static

Definition at line 1690 of file cluster.c.

1691 {
1692  List *inhoids;
1693  ListCell *lc;
1694  List *rtcs = NIL;
1695  MemoryContext old_context;
1696 
1697  /* Do not lock the children until they're processed */
1698  inhoids = find_all_inheritors(indexOid, NoLock, NULL);
1699 
1700  foreach(lc, inhoids)
1701  {
1702  Oid indexrelid = lfirst_oid(lc);
1703  Oid relid = IndexGetRelation(indexrelid, false);
1704  RelToCluster *rtc;
1705 
1706  /* consider only leaf indexes */
1707  if (get_rel_relkind(indexrelid) != RELKIND_INDEX)
1708  continue;
1709 
1710  /*
1711  * It's possible that the user does not have privileges to CLUSTER the
1712  * leaf partition despite having such privileges on the partitioned
1713  * table. We skip any partitions which the user is not permitted to
1714  * CLUSTER.
1715  */
1717  continue;
1718 
1719  /* Use a permanent memory context for the result list */
1720  old_context = MemoryContextSwitchTo(cluster_context);
1721 
1722  rtc = (RelToCluster *) palloc(sizeof(RelToCluster));
1723  rtc->tableOid = relid;
1724  rtc->indexOid = indexrelid;
1725  rtcs = lappend(rtcs, rtc);
1726 
1727  MemoryContextSwitchTo(old_context);
1728  }
1729 
1730  return rtcs;
1731 }
Oid IndexGetRelation(Oid indexId, bool missing_ok)
Definition: index.c:3523
char get_rel_relkind(Oid relid)
Definition: lsyscache.c:2003
List * find_all_inheritors(Oid parentrelId, LOCKMODE lockmode, List **numparents)
Definition: pg_inherits.c:255

References cluster_is_permitted_for_relation(), find_all_inheritors(), get_rel_relkind(), GetUserId(), IndexGetRelation(), RelToCluster::indexOid, lappend(), lfirst_oid, MemoryContextSwitchTo(), NIL, NoLock, palloc(), and RelToCluster::tableOid.

Referenced by cluster().

◆ make_new_heap()

Oid make_new_heap ( Oid  OIDOldHeap,
Oid  NewTableSpace,
Oid  NewAccessMethod,
char  relpersistence,
LOCKMODE  lockmode 
)

Definition at line 688 of file cluster.c.

690 {
691  TupleDesc OldHeapDesc;
692  char NewHeapName[NAMEDATALEN];
693  Oid OIDNewHeap;
694  Oid toastid;
695  Relation OldHeap;
696  HeapTuple tuple;
697  Datum reloptions;
698  bool isNull;
699  Oid namespaceid;
700 
701  OldHeap = table_open(OIDOldHeap, lockmode);
702  OldHeapDesc = RelationGetDescr(OldHeap);
703 
704  /*
705  * Note that the NewHeap will not receive any of the defaults or
706  * constraints associated with the OldHeap; we don't need 'em, and there's
707  * no reason to spend cycles inserting them into the catalogs only to
708  * delete them.
709  */
710 
711  /*
712  * But we do want to use reloptions of the old heap for new heap.
713  */
714  tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(OIDOldHeap));
715  if (!HeapTupleIsValid(tuple))
716  elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap);
717  reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions,
718  &isNull);
719  if (isNull)
720  reloptions = (Datum) 0;
721 
722  if (relpersistence == RELPERSISTENCE_TEMP)
723  namespaceid = LookupCreationNamespace("pg_temp");
724  else
725  namespaceid = RelationGetNamespace(OldHeap);
726 
727  /*
728  * Create the new heap, using a temporary name in the same namespace as
729  * the existing table. NOTE: there is some risk of collision with user
730  * relnames. Working around this seems more trouble than it's worth; in
731  * particular, we can't create the new heap in a different namespace from
732  * the old, or we will have problems with the TEMP status of temp tables.
733  *
734  * Note: the new heap is not a shared relation, even if we are rebuilding
735  * a shared rel. However, we do make the new heap mapped if the source is
736  * mapped. This simplifies swap_relation_files, and is absolutely
737  * necessary for rebuilding pg_class, for reasons explained there.
738  */
739  snprintf(NewHeapName, sizeof(NewHeapName), "pg_temp_%u", OIDOldHeap);
740 
741  OIDNewHeap = heap_create_with_catalog(NewHeapName,
742  namespaceid,
743  NewTableSpace,
744  InvalidOid,
745  InvalidOid,
746  InvalidOid,
747  OldHeap->rd_rel->relowner,
748  NewAccessMethod,
749  OldHeapDesc,
750  NIL,
751  RELKIND_RELATION,
752  relpersistence,
753  false,
754  RelationIsMapped(OldHeap),
756  reloptions,
757  false,
758  true,
759  true,
760  OIDOldHeap,
761  NULL);
762  Assert(OIDNewHeap != InvalidOid);
763 
764  ReleaseSysCache(tuple);
765 
766  /*
767  * Advance command counter so that the newly-created relation's catalog
768  * tuples will be visible to table_open.
769  */
771 
772  /*
773  * If necessary, create a TOAST table for the new relation.
774  *
775  * If the relation doesn't have a TOAST table already, we can't need one
776  * for the new relation. The other way around is possible though: if some
777  * wide columns have been dropped, NewHeapCreateToastTable can decide that
778  * no TOAST table is needed for the new table.
779  *
780  * Note that NewHeapCreateToastTable ends with CommandCounterIncrement, so
781  * that the TOAST table will be visible for insertion.
782  */
783  toastid = OldHeap->rd_rel->reltoastrelid;
784  if (OidIsValid(toastid))
785  {
786  /* keep the existing toast table's reloptions, if any */
787  tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(toastid));
788  if (!HeapTupleIsValid(tuple))
789  elog(ERROR, "cache lookup failed for relation %u", toastid);
790  reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions,
791  &isNull);
792  if (isNull)
793  reloptions = (Datum) 0;
794 
795  NewHeapCreateToastTable(OIDNewHeap, reloptions, lockmode, toastid);
796 
797  ReleaseSysCache(tuple);
798  }
799 
800  table_close(OldHeap, NoLock);
801 
802  return OIDNewHeap;
803 }
Oid heap_create_with_catalog(const char *relname, Oid relnamespace, Oid reltablespace, Oid relid, Oid reltypeid, Oid reloftypeid, Oid ownerid, Oid accessmtd, TupleDesc tupdesc, List *cooked_constraints, char relkind, char relpersistence, bool shared_relation, bool mapped_relation, OnCommitAction oncommit, Datum reloptions, bool use_user_acl, bool allow_system_table_mods, bool is_internal, Oid relrewrite, ObjectAddress *typaddress)
Definition: heap.c:1104
Oid LookupCreationNamespace(const char *nspname)
Definition: namespace.c:3413
uintptr_t Datum
Definition: postgres.h:64
@ ONCOMMIT_NOOP
Definition: primnodes.h:57
#define RelationIsMapped(relation)
Definition: rel.h:554
void ReleaseSysCache(HeapTuple tuple)
Definition: syscache.c:266
HeapTuple SearchSysCache1(int cacheId, Datum key1)
Definition: syscache.c:218
Datum SysCacheGetAttr(int cacheId, HeapTuple tup, AttrNumber attributeNumber, bool *isNull)
Definition: syscache.c:479
void NewHeapCreateToastTable(Oid relOid, Datum reloptions, LOCKMODE lockmode, Oid OIDOldToast)
Definition: toasting.c:63

References Assert, CommandCounterIncrement(), elog, ERROR, heap_create_with_catalog(), HeapTupleIsValid, InvalidOid, LookupCreationNamespace(), NAMEDATALEN, NewHeapCreateToastTable(), NIL, NoLock, ObjectIdGetDatum(), OidIsValid, ONCOMMIT_NOOP, RelationData::rd_rel, RelationGetDescr, RelationGetNamespace, RelationIsMapped, ReleaseSysCache(), SearchSysCache1(), snprintf, SysCacheGetAttr(), table_close(), and table_open().

Referenced by ATRewriteTables(), ExecRefreshMatView(), and rebuild_relation().

◆ mark_index_clustered()

void mark_index_clustered ( Relation  rel,
Oid  indexOid,
bool  is_internal 
)

Definition at line 560 of file cluster.c.

561 {
562  HeapTuple indexTuple;
563  Form_pg_index indexForm;
564  Relation pg_index;
565  ListCell *index;
566 
567  /* Disallow applying to a partitioned table */
568  if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
569  ereport(ERROR,
570  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
571  errmsg("cannot mark index clustered in partitioned table")));
572 
573  /*
574  * If the index is already marked clustered, no need to do anything.
575  */
576  if (OidIsValid(indexOid))
577  {
578  if (get_index_isclustered(indexOid))
579  return;
580  }
581 
582  /*
583  * Check each index of the relation and set/clear the bit as needed.
584  */
585  pg_index = table_open(IndexRelationId, RowExclusiveLock);
586 
587  foreach(index, RelationGetIndexList(rel))
588  {
589  Oid thisIndexOid = lfirst_oid(index);
590 
591  indexTuple = SearchSysCacheCopy1(INDEXRELID,
592  ObjectIdGetDatum(thisIndexOid));
593  if (!HeapTupleIsValid(indexTuple))
594  elog(ERROR, "cache lookup failed for index %u", thisIndexOid);
595  indexForm = (Form_pg_index) GETSTRUCT(indexTuple);
596 
597  /*
598  * Unset the bit if set. We know it's wrong because we checked this
599  * earlier.
600  */
601  if (indexForm->indisclustered)
602  {
603  indexForm->indisclustered = false;
604  CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
605  }
606  else if (thisIndexOid == indexOid)
607  {
608  /* this was checked earlier, but let's be real sure */
609  if (!indexForm->indisvalid)
610  elog(ERROR, "cannot cluster on invalid index %u", indexOid);
611  indexForm->indisclustered = true;
612  CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
613  }
614 
615  InvokeObjectPostAlterHookArg(IndexRelationId, thisIndexOid, 0,
616  InvalidOid, is_internal);
617 
618  heap_freetuple(indexTuple);
619  }
620 
621  table_close(pg_index, RowExclusiveLock);
622 }
#define InvokeObjectPostAlterHookArg(classId, objectId, subId, auxiliaryId, is_internal)
Definition: objectaccess.h:200

References CatalogTupleUpdate(), elog, ereport, errcode(), errmsg(), ERROR, get_index_isclustered(), GETSTRUCT, heap_freetuple(), HeapTupleIsValid, InvalidOid, InvokeObjectPostAlterHookArg, lfirst_oid, ObjectIdGetDatum(), OidIsValid, RelationData::rd_rel, RelationGetIndexList(), RowExclusiveLock, SearchSysCacheCopy1, HeapTupleData::t_self, table_close(), and table_open().

Referenced by ATExecClusterOn(), ATExecDropCluster(), and rebuild_relation().

◆ rebuild_relation()

static void rebuild_relation ( Relation  OldHeap,
Oid  indexOid,
bool  verbose 
)
static

Definition at line 633 of file cluster.c.

634 {
635  Oid tableOid = RelationGetRelid(OldHeap);
636  Oid accessMethod = OldHeap->rd_rel->relam;
637  Oid tableSpace = OldHeap->rd_rel->reltablespace;
638  Oid OIDNewHeap;
639  char relpersistence;
640  bool is_system_catalog;
641  bool swap_toast_by_content;
642  TransactionId frozenXid;
643  MultiXactId cutoffMulti;
644 
645  if (OidIsValid(indexOid))
646  /* Mark the correct index as clustered */
647  mark_index_clustered(OldHeap, indexOid, true);
648 
649  /* Remember info about rel before closing OldHeap */
650  relpersistence = OldHeap->rd_rel->relpersistence;
651  is_system_catalog = IsSystemRelation(OldHeap);
652 
653  /* Close relcache entry, but keep lock until transaction commit */
654  table_close(OldHeap, NoLock);
655 
656  /* Create the transient table that will receive the re-ordered data */
657  OIDNewHeap = make_new_heap(tableOid, tableSpace,
658  accessMethod,
659  relpersistence,
661 
662  /* Copy the heap data into the new table in the desired order */
663  copy_table_data(OIDNewHeap, tableOid, indexOid, verbose,
664  &swap_toast_by_content, &frozenXid, &cutoffMulti);
665 
666  /*
667  * Swap the physical files of the target and transient tables, then
668  * rebuild the target's indexes and throw away the transient table.
669  */
670  finish_heap_swap(tableOid, OIDNewHeap, is_system_catalog,
671  swap_toast_by_content, false, true,
672  frozenXid, cutoffMulti,
673  relpersistence);
674 }
bool IsSystemRelation(Relation relation)
Definition: catalog.c:73
void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap, bool is_system_catalog, bool swap_toast_by_content, bool check_constraints, bool is_internal, TransactionId frozenXid, MultiXactId cutoffMulti, char newrelpersistence)
Definition: cluster.c:1438
static void copy_table_data(Oid OIDNewHeap, Oid OIDOldHeap, Oid OIDOldIndex, bool verbose, bool *pSwapToastByContent, TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
Definition: cluster.c:814
Oid make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod, char relpersistence, LOCKMODE lockmode)
Definition: cluster.c:688
void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal)
Definition: cluster.c:560

References AccessExclusiveLock, copy_table_data(), finish_heap_swap(), IsSystemRelation(), make_new_heap(), mark_index_clustered(), NoLock, OidIsValid, RelationData::rd_rel, RelationGetRelid, table_close(), and verbose.

Referenced by cluster_rel().

◆ swap_relation_files()

static void swap_relation_files ( Oid  r1,
Oid  r2,
bool  target_is_pg_class,
bool  swap_toast_by_content,
bool  is_internal,
TransactionId  frozenXid,
MultiXactId  cutoffMulti,
Oid mapped_tables 
)
static

Definition at line 1061 of file cluster.c.

1067 {
1068  Relation relRelation;
1069  HeapTuple reltup1,
1070  reltup2;
1071  Form_pg_class relform1,
1072  relform2;
1073  RelFileNumber relfilenumber1,
1074  relfilenumber2;
1075  RelFileNumber swaptemp;
1076  char swptmpchr;
1077  Oid relam1,
1078  relam2;
1079 
1080  /* We need writable copies of both pg_class tuples. */
1081  relRelation = table_open(RelationRelationId, RowExclusiveLock);
1082 
1083  reltup1 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r1));
1084  if (!HeapTupleIsValid(reltup1))
1085  elog(ERROR, "cache lookup failed for relation %u", r1);
1086  relform1 = (Form_pg_class) GETSTRUCT(reltup1);
1087 
1088  reltup2 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r2));
1089  if (!HeapTupleIsValid(reltup2))
1090  elog(ERROR, "cache lookup failed for relation %u", r2);
1091  relform2 = (Form_pg_class) GETSTRUCT(reltup2);
1092 
1093  relfilenumber1 = relform1->relfilenode;
1094  relfilenumber2 = relform2->relfilenode;
1095  relam1 = relform1->relam;
1096  relam2 = relform2->relam;
1097 
1098  if (RelFileNumberIsValid(relfilenumber1) &&
1099  RelFileNumberIsValid(relfilenumber2))
1100  {
1101  /*
1102  * Normal non-mapped relations: swap relfilenumbers, reltablespaces,
1103  * relpersistence
1104  */
1105  Assert(!target_is_pg_class);
1106 
1107  swaptemp = relform1->relfilenode;
1108  relform1->relfilenode = relform2->relfilenode;
1109  relform2->relfilenode = swaptemp;
1110 
1111  swaptemp = relform1->reltablespace;
1112  relform1->reltablespace = relform2->reltablespace;
1113  relform2->reltablespace = swaptemp;
1114 
1115  swaptemp = relform1->relam;
1116  relform1->relam = relform2->relam;
1117  relform2->relam = swaptemp;
1118 
1119  swptmpchr = relform1->relpersistence;
1120  relform1->relpersistence = relform2->relpersistence;
1121  relform2->relpersistence = swptmpchr;
1122 
1123  /* Also swap toast links, if we're swapping by links */
1124  if (!swap_toast_by_content)
1125  {
1126  swaptemp = relform1->reltoastrelid;
1127  relform1->reltoastrelid = relform2->reltoastrelid;
1128  relform2->reltoastrelid = swaptemp;
1129  }
1130  }
1131  else
1132  {
1133  /*
1134  * Mapped-relation case. Here we have to swap the relation mappings
1135  * instead of modifying the pg_class columns. Both must be mapped.
1136  */
1137  if (RelFileNumberIsValid(relfilenumber1) ||
1138  RelFileNumberIsValid(relfilenumber2))
1139  elog(ERROR, "cannot swap mapped relation \"%s\" with non-mapped relation",
1140  NameStr(relform1->relname));
1141 
1142  /*
1143  * We can't change the tablespace nor persistence of a mapped rel, and
1144  * we can't handle toast link swapping for one either, because we must
1145  * not apply any critical changes to its pg_class row. These cases
1146  * should be prevented by upstream permissions tests, so these checks
1147  * are non-user-facing emergency backstop.
1148  */
1149  if (relform1->reltablespace != relform2->reltablespace)
1150  elog(ERROR, "cannot change tablespace of mapped relation \"%s\"",
1151  NameStr(relform1->relname));
1152  if (relform1->relpersistence != relform2->relpersistence)
1153  elog(ERROR, "cannot change persistence of mapped relation \"%s\"",
1154  NameStr(relform1->relname));
1155  if (relform1->relam != relform2->relam)
1156  elog(ERROR, "cannot change access method of mapped relation \"%s\"",
1157  NameStr(relform1->relname));
1158  if (!swap_toast_by_content &&
1159  (relform1->reltoastrelid || relform2->reltoastrelid))
1160  elog(ERROR, "cannot swap toast by links for mapped relation \"%s\"",
1161  NameStr(relform1->relname));
1162 
1163  /*
1164  * Fetch the mappings --- shouldn't fail, but be paranoid
1165  */
1166  relfilenumber1 = RelationMapOidToFilenumber(r1, relform1->relisshared);
1167  if (!RelFileNumberIsValid(relfilenumber1))
1168  elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
1169  NameStr(relform1->relname), r1);
1170  relfilenumber2 = RelationMapOidToFilenumber(r2, relform2->relisshared);
1171  if (!RelFileNumberIsValid(relfilenumber2))
1172  elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
1173  NameStr(relform2->relname), r2);
1174 
1175  /*
1176  * Send replacement mappings to relmapper. Note these won't actually
1177  * take effect until CommandCounterIncrement.
1178  */
1179  RelationMapUpdateMap(r1, relfilenumber2, relform1->relisshared, false);
1180  RelationMapUpdateMap(r2, relfilenumber1, relform2->relisshared, false);
1181 
1182  /* Pass OIDs of mapped r2 tables back to caller */
1183  *mapped_tables++ = r2;
1184  }
1185 
1186  /*
1187  * Recognize that rel1's relfilenumber (swapped from rel2) is new in this
1188  * subtransaction. The rel2 storage (swapped from rel1) may or may not be
1189  * new.
1190  */
1191  {
1192  Relation rel1,
1193  rel2;
1194 
1195  rel1 = relation_open(r1, NoLock);
1196  rel2 = relation_open(r2, NoLock);
1197  rel2->rd_createSubid = rel1->rd_createSubid;
1201  relation_close(rel1, NoLock);
1202  relation_close(rel2, NoLock);
1203  }
1204 
1205  /*
1206  * In the case of a shared catalog, these next few steps will only affect
1207  * our own database's pg_class row; but that's okay, because they are all
1208  * noncritical updates. That's also an important fact for the case of a
1209  * mapped catalog, because it's possible that we'll commit the map change
1210  * and then fail to commit the pg_class update.
1211  */
1212 
1213  /* set rel1's frozen Xid and minimum MultiXid */
1214  if (relform1->relkind != RELKIND_INDEX)
1215  {
1216  Assert(!TransactionIdIsValid(frozenXid) ||
1217  TransactionIdIsNormal(frozenXid));
1218  relform1->relfrozenxid = frozenXid;
1219  relform1->relminmxid = cutoffMulti;
1220  }
1221 
1222  /* swap size statistics too, since new rel has freshly-updated stats */
1223  {
1224  int32 swap_pages;
1225  float4 swap_tuples;
1226  int32 swap_allvisible;
1227 
1228  swap_pages = relform1->relpages;
1229  relform1->relpages = relform2->relpages;
1230  relform2->relpages = swap_pages;
1231 
1232  swap_tuples = relform1->reltuples;
1233  relform1->reltuples = relform2->reltuples;
1234  relform2->reltuples = swap_tuples;
1235 
1236  swap_allvisible = relform1->relallvisible;
1237  relform1->relallvisible = relform2->relallvisible;
1238  relform2->relallvisible = swap_allvisible;
1239  }
1240 
1241  /*
1242  * Update the tuples in pg_class --- unless the target relation of the
1243  * swap is pg_class itself. In that case, there is zero point in making
1244  * changes because we'd be updating the old data that we're about to throw
1245  * away. Because the real work being done here for a mapped relation is
1246  * just to change the relation map settings, it's all right to not update
1247  * the pg_class rows in this case. The most important changes will instead
1248  * performed later, in finish_heap_swap() itself.
1249  */
1250  if (!target_is_pg_class)
1251  {
1252  CatalogIndexState indstate;
1253 
1254  indstate = CatalogOpenIndexes(relRelation);
1255  CatalogTupleUpdateWithInfo(relRelation, &reltup1->t_self, reltup1,
1256  indstate);
1257  CatalogTupleUpdateWithInfo(relRelation, &reltup2->t_self, reltup2,
1258  indstate);
1259  CatalogCloseIndexes(indstate);
1260  }
1261  else
1262  {
1263  /* no update ... but we do still need relcache inval */
1266  }
1267 
1268  /*
1269  * Now that pg_class has been updated with its relevant information for
1270  * the swap, update the dependency of the relations to point to their new
1271  * table AM, if it has changed.
1272  */
1273  if (relam1 != relam2)
1274  {
1275  if (changeDependencyFor(RelationRelationId,
1276  r1,
1277  AccessMethodRelationId,
1278  relam1,
1279  relam2) != 1)
1280  elog(ERROR, "could not change access method dependency for relation \"%s.%s\"",
1282  get_rel_name(r1));
1283  if (changeDependencyFor(RelationRelationId,
1284  r2,
1285  AccessMethodRelationId,
1286  relam2,
1287  relam1) != 1)
1288  elog(ERROR, "could not change access method dependency for relation \"%s.%s\"",
1290  get_rel_name(r2));
1291  }
1292 
1293  /*
1294  * Post alter hook for modified relations. The change to r2 is always
1295  * internal, but r1 depends on the invocation context.
1296  */
1297  InvokeObjectPostAlterHookArg(RelationRelationId, r1, 0,
1298  InvalidOid, is_internal);
1299  InvokeObjectPostAlterHookArg(RelationRelationId, r2, 0,
1300  InvalidOid, true);
1301 
1302  /*
1303  * If we have toast tables associated with the relations being swapped,
1304  * deal with them too.
1305  */
1306  if (relform1->reltoastrelid || relform2->reltoastrelid)
1307  {
1308  if (swap_toast_by_content)
1309  {
1310  if (relform1->reltoastrelid && relform2->reltoastrelid)
1311  {
1312  /* Recursively swap the contents of the toast tables */
1313  swap_relation_files(relform1->reltoastrelid,
1314  relform2->reltoastrelid,
1315  target_is_pg_class,
1316  swap_toast_by_content,
1317  is_internal,
1318  frozenXid,
1319  cutoffMulti,
1320  mapped_tables);
1321  }
1322  else
1323  {
1324  /* caller messed up */
1325  elog(ERROR, "cannot swap toast files by content when there's only one");
1326  }
1327  }
1328  else
1329  {
1330  /*
1331  * We swapped the ownership links, so we need to change dependency
1332  * data to match.
1333  *
1334  * NOTE: it is possible that only one table has a toast table.
1335  *
1336  * NOTE: at present, a TOAST table's only dependency is the one on
1337  * its owning table. If more are ever created, we'd need to use
1338  * something more selective than deleteDependencyRecordsFor() to
1339  * get rid of just the link we want.
1340  */
1341  ObjectAddress baseobject,
1342  toastobject;
1343  long count;
1344 
1345  /*
1346  * We disallow this case for system catalogs, to avoid the
1347  * possibility that the catalog we're rebuilding is one of the
1348  * ones the dependency changes would change. It's too late to be
1349  * making any data changes to the target catalog.
1350  */
1351  if (IsSystemClass(r1, relform1))
1352  elog(ERROR, "cannot swap toast files by links for system catalogs");
1353 
1354  /* Delete old dependencies */
1355  if (relform1->reltoastrelid)
1356  {
1357  count = deleteDependencyRecordsFor(RelationRelationId,
1358  relform1->reltoastrelid,
1359  false);
1360  if (count != 1)
1361  elog(ERROR, "expected one dependency record for TOAST table, found %ld",
1362  count);
1363  }
1364  if (relform2->reltoastrelid)
1365  {
1366  count = deleteDependencyRecordsFor(RelationRelationId,
1367  relform2->reltoastrelid,
1368  false);
1369  if (count != 1)
1370  elog(ERROR, "expected one dependency record for TOAST table, found %ld",
1371  count);
1372  }
1373 
1374  /* Register new dependencies */
1375  baseobject.classId = RelationRelationId;
1376  baseobject.objectSubId = 0;
1377  toastobject.classId = RelationRelationId;
1378  toastobject.objectSubId = 0;
1379 
1380  if (relform1->reltoastrelid)
1381  {
1382  baseobject.objectId = r1;
1383  toastobject.objectId = relform1->reltoastrelid;
1384  recordDependencyOn(&toastobject, &baseobject,
1386  }
1387 
1388  if (relform2->reltoastrelid)
1389  {
1390  baseobject.objectId = r2;
1391  toastobject.objectId = relform2->reltoastrelid;
1392  recordDependencyOn(&toastobject, &baseobject,
1394  }
1395  }
1396  }
1397 
1398  /*
1399  * If we're swapping two toast tables by content, do the same for their
1400  * valid index. The swap can actually be safely done only if the relations
1401  * have indexes.
1402  */
1403  if (swap_toast_by_content &&
1404  relform1->relkind == RELKIND_TOASTVALUE &&
1405  relform2->relkind == RELKIND_TOASTVALUE)
1406  {
1407  Oid toastIndex1,
1408  toastIndex2;
1409 
1410  /* Get valid index for each relation */
1411  toastIndex1 = toast_get_valid_index(r1,
1413  toastIndex2 = toast_get_valid_index(r2,
1415 
1416  swap_relation_files(toastIndex1,
1417  toastIndex2,
1418  target_is_pg_class,
1419  swap_toast_by_content,
1420  is_internal,
1423  mapped_tables);
1424  }
1425 
1426  /* Clean up. */
1427  heap_freetuple(reltup1);
1428  heap_freetuple(reltup2);
1429 
1430  table_close(relRelation, RowExclusiveLock);
1431 }
#define NameStr(name)
Definition: c.h:746
signed int int32
Definition: c.h:494
float float4
Definition: c.h:629
bool IsSystemClass(Oid relid, Form_pg_class reltuple)
Definition: catalog.c:85
@ DEPENDENCY_INTERNAL
Definition: dependency.h:35
void CatalogCloseIndexes(CatalogIndexState indstate)
Definition: indexing.c:61
CatalogIndexState CatalogOpenIndexes(Relation heapRel)
Definition: indexing.c:43
void CatalogTupleUpdateWithInfo(Relation heapRel, ItemPointer otid, HeapTuple tup, CatalogIndexState indstate)
Definition: indexing.c:337
Oid get_rel_namespace(Oid relid)
Definition: lsyscache.c:1952
#define InvalidMultiXactId
Definition: multixact.h:24
void recordDependencyOn(const ObjectAddress *depender, const ObjectAddress *referenced, DependencyType behavior)
Definition: pg_depend.c:46
long changeDependencyFor(Oid classId, Oid objectId, Oid refClassId, Oid oldRefObjectId, Oid newRefObjectId)
Definition: pg_depend.c:458
long deleteDependencyRecordsFor(Oid classId, Oid objectId, bool skipExtensionDeps)
Definition: pg_depend.c:302
void RelationAssumeNewRelfilelocator(Relation relation)
Definition: relcache.c:3925
RelFileNumber RelationMapOidToFilenumber(Oid relationId, bool shared)
Definition: relmapper.c:165
void RelationMapUpdateMap(Oid relationId, RelFileNumber fileNumber, bool shared, bool immediate)
Definition: relmapper.c:325
Oid RelFileNumber
Definition: relpath.h:25
#define RelFileNumberIsValid(relnumber)
Definition: relpath.h:27
Relation relation_open(Oid relationId, LOCKMODE lockmode)
Definition: relation.c:47
SubTransactionId rd_firstRelfilelocatorSubid
Definition: rel.h:106
SubTransactionId rd_newRelfilelocatorSubid
Definition: rel.h:104
SubTransactionId rd_createSubid
Definition: rel.h:103
#define InvalidTransactionId
Definition: transam.h:31
#define TransactionIdIsNormal(xid)
Definition: transam.h:42

References AccessExclusiveLock, Assert, CacheInvalidateRelcacheByTuple(), CatalogCloseIndexes(), CatalogOpenIndexes(), CatalogTupleUpdateWithInfo(), changeDependencyFor(), ObjectAddress::classId, deleteDependencyRecordsFor(), DEPENDENCY_INTERNAL, elog, ERROR, get_namespace_name(), get_rel_name(), get_rel_namespace(), GETSTRUCT, heap_freetuple(), HeapTupleIsValid, InvalidMultiXactId, InvalidOid, InvalidTransactionId, InvokeObjectPostAlterHookArg, IsSystemClass(), NameStr, NoLock, ObjectAddress::objectId, ObjectIdGetDatum(), ObjectAddress::objectSubId, RelationData::rd_createSubid, RelationData::rd_firstRelfilelocatorSubid, RelationData::rd_newRelfilelocatorSubid, recordDependencyOn(), relation_close(), relation_open(), RelationAssumeNewRelfilelocator(), RelationMapOidToFilenumber(), RelationMapUpdateMap(), RelFileNumberIsValid, RowExclusiveLock, SearchSysCacheCopy1, HeapTupleData::t_self, table_close(), table_open(), toast_get_valid_index(), TransactionIdIsNormal, and TransactionIdIsValid.

Referenced by finish_heap_swap().