LCOV - code coverage report
Current view: top level - src/backend/commands - cluster.c (source / functions) Hit Total Coverage
Test: PostgreSQL 18devel Lines: 410 453 90.5 %
Date: 2025-01-18 04:15:08 Functions: 13 13 100.0 %
Legend: Lines: hit not hit

          Line data    Source code
       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         236 : cluster(ParseState *pstate, ClusterStmt *stmt, bool isTopLevel)
     108             : {
     109             :     ListCell   *lc;
     110         236 :     ClusterParams params = {0};
     111         236 :     bool        verbose = false;
     112         236 :     Relation    rel = NULL;
     113         236 :     Oid         indexOid = InvalidOid;
     114             :     MemoryContext cluster_context;
     115             :     List       *rtcs;
     116             : 
     117             :     /* Parse option list */
     118         248 :     foreach(lc, stmt->params)
     119             :     {
     120          12 :         DefElem    *opt = (DefElem *) lfirst(lc);
     121             : 
     122          12 :         if (strcmp(opt->defname, "verbose") == 0)
     123          12 :             verbose = defGetBoolean(opt);
     124             :         else
     125           0 :             ereport(ERROR,
     126             :                     (errcode(ERRCODE_SYNTAX_ERROR),
     127             :                      errmsg("unrecognized CLUSTER option \"%s\"",
     128             :                             opt->defname),
     129             :                      parser_errposition(pstate, opt->location)));
     130             :     }
     131             : 
     132         236 :     params.options = (verbose ? CLUOPT_VERBOSE : 0);
     133             : 
     134         236 :     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         208 :         tableOid = RangeVarGetRelidExtended(stmt->relation,
     145             :                                             AccessExclusiveLock,
     146             :                                             0,
     147             :                                             RangeVarCallbackMaintainsTable,
     148             :                                             NULL);
     149         196 :         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         196 :         if (RELATION_IS_OTHER_TEMP(rel))
     156           0 :             ereport(ERROR,
     157             :                     (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
     158             :                      errmsg("cannot cluster temporary tables of other sessions")));
     159             : 
     160         196 :         if (stmt->indexname == NULL)
     161             :         {
     162             :             ListCell   *index;
     163             : 
     164             :             /* We need to find the index that has indisclustered set. */
     165          46 :             foreach(index, RelationGetIndexList(rel))
     166             :             {
     167          34 :                 indexOid = lfirst_oid(index);
     168          34 :                 if (get_index_isclustered(indexOid))
     169          22 :                     break;
     170          12 :                 indexOid = InvalidOid;
     171             :             }
     172             : 
     173          34 :             if (!OidIsValid(indexOid))
     174          12 :                 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         162 :             indexOid = get_relname_relid(stmt->indexname,
     186         162 :                                          rel->rd_rel->relnamespace);
     187         162 :             if (!OidIsValid(indexOid))
     188           0 :                 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         184 :         if (rel->rd_rel->relkind != RELKIND_PARTITIONED_TABLE)
     196             :         {
     197         158 :             cluster_rel(rel, indexOid, &params);
     198             :             /* cluster_rel closes the relation, but keeps lock */
     199             : 
     200         158 :             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          54 :     PreventInTransactionBlock(isTopLevel, "CLUSTER");
     211             : 
     212             :     /* Also, we need a memory context to hold our list of relations */
     213          54 :     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          54 :     params.options |= CLUOPT_RECHECK;
     230          54 :     if (rel != NULL)
     231             :     {
     232             :         Assert(rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE);
     233          26 :         check_index_is_clusterable(rel, indexOid, AccessShareLock);
     234          20 :         rtcs = get_tables_to_cluster_partitioned(cluster_context, indexOid);
     235             : 
     236             :         /* close relation, releasing lock on parent table */
     237          20 :         table_close(rel, AccessExclusiveLock);
     238             :     }
     239             :     else
     240             :     {
     241          28 :         rtcs = get_tables_to_cluster(cluster_context);
     242          28 :         params.options |= CLUOPT_RECHECK_ISCLUSTERED;
     243             :     }
     244             : 
     245             :     /* Do the job. */
     246          48 :     cluster_multiple_rels(rtcs, &params);
     247             : 
     248             :     /* Start a new transaction for the cleanup work. */
     249          48 :     StartTransactionCommand();
     250             : 
     251             :     /* Clean up working storage */
     252          48 :     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          48 : cluster_multiple_rels(List *rtcs, ClusterParams *params)
     264             : {
     265             :     ListCell   *lc;
     266             : 
     267             :     /* Commit to get out of starting transaction */
     268          48 :     PopActiveSnapshot();
     269          48 :     CommitTransactionCommand();
     270             : 
     271             :     /* Cluster the tables, each in a separate transaction */
     272          78 :     foreach(lc, rtcs)
     273             :     {
     274          30 :         RelToCluster *rtc = (RelToCluster *) lfirst(lc);
     275             :         Relation    rel;
     276             : 
     277             :         /* Start a new transaction for each relation. */
     278          30 :         StartTransactionCommand();
     279             : 
     280             :         /* functions in indexes may want a snapshot set */
     281          30 :         PushActiveSnapshot(GetTransactionSnapshot());
     282             : 
     283          30 :         rel = table_open(rtc->tableOid, AccessExclusiveLock);
     284             : 
     285             :         /* Process this table */
     286          30 :         cluster_rel(rel, rtc->indexOid, params);
     287             :         /* cluster_rel closes the relation, but keeps lock */
     288             : 
     289          30 :         PopActiveSnapshot();
     290          30 :         CommitTransactionCommand();
     291             :     }
     292          48 : }
     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         546 : cluster_rel(Relation OldHeap, Oid indexOid, ClusterParams *params)
     312             : {
     313         546 :     Oid         tableOid = RelationGetRelid(OldHeap);
     314             :     Oid         save_userid;
     315             :     int         save_sec_context;
     316             :     int         save_nestlevel;
     317         546 :     bool        verbose = ((params->options & CLUOPT_VERBOSE) != 0);
     318         546 :     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         546 :     CHECK_FOR_INTERRUPTS();
     325             : 
     326         546 :     pgstat_progress_start_command(PROGRESS_COMMAND_CLUSTER, tableOid);
     327         546 :     if (OidIsValid(indexOid))
     328         188 :         pgstat_progress_update_param(PROGRESS_CLUSTER_COMMAND,
     329             :                                      PROGRESS_CLUSTER_COMMAND_CLUSTER);
     330             :     else
     331         358 :         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         546 :     GetUserIdAndSecContext(&save_userid, &save_sec_context);
     340         546 :     SetUserIdAndSecContext(OldHeap->rd_rel->relowner,
     341             :                            save_sec_context | SECURITY_RESTRICTED_OPERATION);
     342         546 :     save_nestlevel = NewGUCNestLevel();
     343         546 :     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         546 :     if (recheck)
     354             :     {
     355             :         /* Check that the user still has privileges for the relation */
     356          30 :         if (!cluster_is_permitted_for_relation(tableOid, save_userid))
     357             :         {
     358           0 :             relation_close(OldHeap, AccessExclusiveLock);
     359           0 :             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          30 :         if (RELATION_IS_OTHER_TEMP(OldHeap))
     372             :         {
     373           0 :             relation_close(OldHeap, AccessExclusiveLock);
     374           0 :             goto out;
     375             :         }
     376             : 
     377          30 :         if (OidIsValid(indexOid))
     378             :         {
     379             :             /*
     380             :              * Check that the index still exists
     381             :              */
     382          30 :             if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(indexOid)))
     383             :             {
     384           0 :                 relation_close(OldHeap, AccessExclusiveLock);
     385           0 :                 goto out;
     386             :             }
     387             : 
     388             :             /*
     389             :              * Check that the index is still the one with indisclustered set,
     390             :              * if needed.
     391             :              */
     392          30 :             if ((params->options & CLUOPT_RECHECK_ISCLUSTERED) != 0 &&
     393           6 :                 !get_index_isclustered(indexOid))
     394             :             {
     395           0 :                 relation_close(OldHeap, AccessExclusiveLock);
     396           0 :                 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         546 :     if (OidIsValid(indexOid) && OldHeap->rd_rel->relisshared)
     408           0 :         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         546 :     if (RELATION_IS_OTHER_TEMP(OldHeap))
     417             :     {
     418           0 :         if (OidIsValid(indexOid))
     419           0 :             ereport(ERROR,
     420             :                     (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
     421             :                      errmsg("cannot cluster temporary tables of other sessions")));
     422             :         else
     423           0 :             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         546 :     CheckTableNotInUse(OldHeap, OidIsValid(indexOid) ? "CLUSTER" : "VACUUM");
     433             : 
     434             :     /* Check heap and index are valid to cluster on */
     435         546 :     if (OidIsValid(indexOid))
     436             :     {
     437             :         /* verify the index is good and lock it */
     438         188 :         check_index_is_clusterable(OldHeap, indexOid, AccessExclusiveLock);
     439             :         /* also open it */
     440         188 :         index = index_open(indexOid, NoLock);
     441             :     }
     442             :     else
     443         358 :         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         546 :     if (OldHeap->rd_rel->relkind == RELKIND_MATVIEW &&
     453           0 :         !RelationIsPopulated(OldHeap))
     454             :     {
     455           0 :         relation_close(OldHeap, AccessExclusiveLock);
     456           0 :         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         546 :     TransferPredicateLocksToHeapRelation(OldHeap);
     470             : 
     471             :     /* rebuild_relation does all the dirty work */
     472         546 :     rebuild_relation(OldHeap, index, verbose);
     473             :     /* rebuild_relation closes OldHeap, and index if valid */
     474             : 
     475         540 : out:
     476             :     /* Roll back any GUC changes executed by index functions */
     477         540 :     AtEOXact_GUC(false, save_nestlevel);
     478             : 
     479             :     /* Restore userid and security context */
     480         540 :     SetUserIdAndSecContext(save_userid, save_sec_context);
     481             : 
     482         540 :     pgstat_progress_end_command();
     483         540 : }
     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         278 : check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode)
     495             : {
     496             :     Relation    OldIndex;
     497             : 
     498         278 :     OldIndex = index_open(indexOid, lockmode);
     499             : 
     500             :     /*
     501             :      * Check that index is in fact an index on the given relation
     502             :      */
     503         278 :     if (OldIndex->rd_index == NULL ||
     504         278 :         OldIndex->rd_index->indrelid != RelationGetRelid(OldHeap))
     505           0 :         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         278 :     if (!OldIndex->rd_indam->amclusterable)
     513           0 :         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         278 :     if (!heap_attisnull(OldIndex->rd_indextuple, Anum_pg_index_indpred, NULL))
     525           0 :         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         278 :     if (!OldIndex->rd_index->indisvalid)
     539           6 :         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         272 :     index_close(OldIndex, NoLock);
     546         272 : }
     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         270 : 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         270 :     if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
     563          12 :         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         258 :     if (OidIsValid(indexOid))
     571             :     {
     572         246 :         if (get_index_isclustered(indexOid))
     573          40 :             return;
     574             :     }
     575             : 
     576             :     /*
     577             :      * Check each index of the relation and set/clear the bit as needed.
     578             :      */
     579         218 :     pg_index = table_open(IndexRelationId, RowExclusiveLock);
     580             : 
     581         642 :     foreach(index, RelationGetIndexList(rel))
     582             :     {
     583         424 :         Oid         thisIndexOid = lfirst_oid(index);
     584             : 
     585         424 :         indexTuple = SearchSysCacheCopy1(INDEXRELID,
     586             :                                          ObjectIdGetDatum(thisIndexOid));
     587         424 :         if (!HeapTupleIsValid(indexTuple))
     588           0 :             elog(ERROR, "cache lookup failed for index %u", thisIndexOid);
     589         424 :         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         424 :         if (indexForm->indisclustered)
     596             :         {
     597          30 :             indexForm->indisclustered = false;
     598          30 :             CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
     599             :         }
     600         394 :         else if (thisIndexOid == indexOid)
     601             :         {
     602             :             /* this was checked earlier, but let's be real sure */
     603         206 :             if (!indexForm->indisvalid)
     604           0 :                 elog(ERROR, "cannot cluster on invalid index %u", indexOid);
     605         206 :             indexForm->indisclustered = true;
     606         206 :             CatalogTupleUpdate(pg_index, &indexTuple->t_self, indexTuple);
     607             :         }
     608             : 
     609         424 :         InvokeObjectPostAlterHookArg(IndexRelationId, thisIndexOid, 0,
     610             :                                      InvalidOid, is_internal);
     611             : 
     612         424 :         heap_freetuple(indexTuple);
     613             :     }
     614             : 
     615         218 :     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         546 : rebuild_relation(Relation OldHeap, Relation index, bool verbose)
     630             : {
     631         546 :     Oid         tableOid = RelationGetRelid(OldHeap);
     632         546 :     Oid         accessMethod = OldHeap->rd_rel->relam;
     633         546 :     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         546 :     if (index)
     646             :         /* Mark the correct index as clustered */
     647         188 :         mark_index_clustered(OldHeap, RelationGetRelid(index), true);
     648             : 
     649             :     /* Remember info about rel before closing OldHeap */
     650         546 :     relpersistence = OldHeap->rd_rel->relpersistence;
     651         546 :     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         546 :     OIDNewHeap = make_new_heap(tableOid, tableSpace,
     660             :                                accessMethod,
     661             :                                relpersistence,
     662             :                                NoLock);
     663             :     Assert(CheckRelationOidLockedByMe(OIDNewHeap, AccessExclusiveLock, false));
     664         546 :     NewHeap = table_open(OIDNewHeap, NoLock);
     665             : 
     666             :     /* Copy the heap data into the new table in the desired order */
     667         546 :     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         546 :     table_close(OldHeap, NoLock);
     673         546 :     if (index)
     674         188 :         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         546 :     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         546 :     finish_heap_swap(tableOid, OIDNewHeap, is_system_catalog,
     688             :                      swap_toast_by_content, false, true,
     689             :                      frozenXid, cutoffMulti,
     690             :                      relpersistence);
     691         540 : }
     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        2036 : 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        2036 :     OldHeap = table_open(OIDOldHeap, lockmode);
     719        2036 :     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        2036 :     tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(OIDOldHeap));
     732        2036 :     if (!HeapTupleIsValid(tuple))
     733           0 :         elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap);
     734        2036 :     reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions,
     735             :                                  &isNull);
     736        2036 :     if (isNull)
     737        1998 :         reloptions = (Datum) 0;
     738             : 
     739        2036 :     if (relpersistence == RELPERSISTENCE_TEMP)
     740         152 :         namespaceid = LookupCreationNamespace("pg_temp");
     741             :     else
     742        1884 :         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        2036 :     snprintf(NewHeapName, sizeof(NewHeapName), "pg_temp_%u", OIDOldHeap);
     757             : 
     758        2036 :     OIDNewHeap = heap_create_with_catalog(NewHeapName,
     759             :                                           namespaceid,
     760             :                                           NewTableSpace,
     761             :                                           InvalidOid,
     762             :                                           InvalidOid,
     763             :                                           InvalidOid,
     764        2036 :                                           OldHeap->rd_rel->relowner,
     765             :                                           NewAccessMethod,
     766             :                                           OldHeapDesc,
     767             :                                           NIL,
     768             :                                           RELKIND_RELATION,
     769             :                                           relpersistence,
     770             :                                           false,
     771        2036 :                                           RelationIsMapped(OldHeap),
     772             :                                           ONCOMMIT_NOOP,
     773             :                                           reloptions,
     774             :                                           false,
     775             :                                           true,
     776             :                                           true,
     777             :                                           OIDOldHeap,
     778             :                                           NULL);
     779             :     Assert(OIDNewHeap != InvalidOid);
     780             : 
     781        2036 :     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        2036 :     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        2036 :     toastid = OldHeap->rd_rel->reltoastrelid;
     801        2036 :     if (OidIsValid(toastid))
     802             :     {
     803             :         /* keep the existing toast table's reloptions, if any */
     804         836 :         tuple = SearchSysCache1(RELOID, ObjectIdGetDatum(toastid));
     805         836 :         if (!HeapTupleIsValid(tuple))
     806           0 :             elog(ERROR, "cache lookup failed for relation %u", toastid);
     807         836 :         reloptions = SysCacheGetAttr(RELOID, tuple, Anum_pg_class_reloptions,
     808             :                                      &isNull);
     809         836 :         if (isNull)
     810         836 :             reloptions = (Datum) 0;
     811             : 
     812         836 :         NewHeapCreateToastTable(OIDNewHeap, reloptions, lockmode, toastid);
     813             : 
     814         836 :         ReleaseSysCache(tuple);
     815             :     }
     816             : 
     817        2036 :     table_close(OldHeap, NoLock);
     818             : 
     819        2036 :     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         546 : 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         546 :     double      num_tuples = 0,
     844         546 :                 tups_vacuumed = 0,
     845         546 :                 tups_recently_dead = 0;
     846             :     BlockNumber num_pages;
     847         546 :     int         elevel = verbose ? INFO : DEBUG2;
     848             :     PGRUsage    ru0;
     849             :     char       *nspname;
     850             : 
     851         546 :     pg_rusage_init(&ru0);
     852             : 
     853             :     /* Store a copy of the namespace name for logging purposes */
     854         546 :     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         546 :     oldTupDesc = RelationGetDescr(OldHeap);
     861         546 :     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         546 :     if (OldHeap->rd_rel->reltoastrelid)
     878         184 :         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         546 :     if (OldHeap->rd_rel->reltoastrelid && NewHeap->rd_rel->reltoastrelid)
     888             :     {
     889         184 :         *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         184 :         NewHeap->rd_toastoid = OldHeap->rd_rel->reltoastrelid;
     910             :     }
     911             :     else
     912         362 :         *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         546 :     memset(&params, 0, sizeof(VacuumParams));
     920         546 :     vacuum_get_cutoffs(OldHeap, &params, &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         546 :         TransactionId relfrozenxid = OldHeap->rd_rel->relfrozenxid;
     928             : 
     929        1092 :         if (TransactionIdIsValid(relfrozenxid) &&
     930         546 :             TransactionIdPrecedes(cutoffs.FreezeLimit, relfrozenxid))
     931          86 :             cutoffs.FreezeLimit = relfrozenxid;
     932             :     }
     933             : 
     934             :     /*
     935             :      * MultiXactCutoff, similarly, shouldn't go backwards either.
     936             :      */
     937             :     {
     938         546 :         MultiXactId relminmxid = OldHeap->rd_rel->relminmxid;
     939             : 
     940        1092 :         if (MultiXactIdIsValid(relminmxid) &&
     941         546 :             MultiXactIdPrecedes(cutoffs.MultiXactCutoff, relminmxid))
     942           0 :             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         546 :     if (OldIndex != NULL && OldIndex->rd_rel->relam == BTREE_AM_OID)
     953         188 :         use_sort = plan_cluster_use_sort(RelationGetRelid(OldHeap),
     954             :                                          RelationGetRelid(OldIndex));
     955             :     else
     956         358 :         use_sort = false;
     957             : 
     958             :     /* Log what we're doing */
     959         546 :     if (OldIndex != NULL && !use_sort)
     960          78 :         ereport(elevel,
     961             :                 (errmsg("clustering \"%s.%s\" using index scan on \"%s\"",
     962             :                         nspname,
     963             :                         RelationGetRelationName(OldHeap),
     964             :                         RelationGetRelationName(OldIndex))));
     965         468 :     else if (use_sort)
     966         110 :         ereport(elevel,
     967             :                 (errmsg("clustering \"%s.%s\" using sequential scan and sort",
     968             :                         nspname,
     969             :                         RelationGetRelationName(OldHeap))));
     970             :     else
     971         358 :         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         546 :     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         546 :     *pFreezeXid = cutoffs.FreezeLimit;
     990         546 :     *pCutoffMulti = cutoffs.MultiXactCutoff;
     991             : 
     992             :     /* Reset rd_toastoid just to be tidy --- it shouldn't be looked at again */
     993         546 :     NewHeap->rd_toastoid = InvalidOid;
     994             : 
     995         546 :     num_pages = RelationGetNumberOfBlocks(NewHeap);
     996             : 
     997             :     /* Log what we did */
     998         546 :     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         546 :     relRelation = table_open(RelationRelationId, RowExclusiveLock);
    1011             : 
    1012         546 :     reltup = SearchSysCacheCopy1(RELOID,
    1013             :                                  ObjectIdGetDatum(RelationGetRelid(NewHeap)));
    1014         546 :     if (!HeapTupleIsValid(reltup))
    1015           0 :         elog(ERROR, "cache lookup failed for relation %u",
    1016             :              RelationGetRelid(NewHeap));
    1017         546 :     relform = (Form_pg_class) GETSTRUCT(reltup);
    1018             : 
    1019         546 :     relform->relpages = num_pages;
    1020         546 :     relform->reltuples = num_tuples;
    1021             : 
    1022             :     /* Don't update the stats for pg_class.  See swap_relation_files. */
    1023         546 :     if (RelationGetRelid(OldHeap) != RelationRelationId)
    1024         508 :         CatalogTupleUpdate(relRelation, &reltup->t_self, reltup);
    1025             :     else
    1026          38 :         CacheInvalidateRelcacheByTuple(reltup);
    1027             : 
    1028             :     /* Clean up. */
    1029         546 :     heap_freetuple(reltup);
    1030         546 :     table_close(relRelation, RowExclusiveLock);
    1031             : 
    1032             :     /* Make the update visible */
    1033         546 :     CommandCounterIncrement();
    1034         546 : }
    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        2266 : 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        2266 :     relRelation = table_open(RelationRelationId, RowExclusiveLock);
    1084             : 
    1085        2266 :     reltup1 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r1));
    1086        2266 :     if (!HeapTupleIsValid(reltup1))
    1087           0 :         elog(ERROR, "cache lookup failed for relation %u", r1);
    1088        2266 :     relform1 = (Form_pg_class) GETSTRUCT(reltup1);
    1089             : 
    1090        2266 :     reltup2 = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(r2));
    1091        2266 :     if (!HeapTupleIsValid(reltup2))
    1092           0 :         elog(ERROR, "cache lookup failed for relation %u", r2);
    1093        2266 :     relform2 = (Form_pg_class) GETSTRUCT(reltup2);
    1094             : 
    1095        2266 :     relfilenumber1 = relform1->relfilenode;
    1096        2266 :     relfilenumber2 = relform2->relfilenode;
    1097        2266 :     relam1 = relform1->relam;
    1098        2266 :     relam2 = relform2->relam;
    1099             : 
    1100        2266 :     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        2098 :         swaptemp = relform1->relfilenode;
    1110        2098 :         relform1->relfilenode = relform2->relfilenode;
    1111        2098 :         relform2->relfilenode = swaptemp;
    1112             : 
    1113        2098 :         swaptemp = relform1->reltablespace;
    1114        2098 :         relform1->reltablespace = relform2->reltablespace;
    1115        2098 :         relform2->reltablespace = swaptemp;
    1116             : 
    1117        2098 :         swaptemp = relform1->relam;
    1118        2098 :         relform1->relam = relform2->relam;
    1119        2098 :         relform2->relam = swaptemp;
    1120             : 
    1121        2098 :         swptmpchr = relform1->relpersistence;
    1122        2098 :         relform1->relpersistence = relform2->relpersistence;
    1123        2098 :         relform2->relpersistence = swptmpchr;
    1124             : 
    1125             :         /* Also swap toast links, if we're swapping by links */
    1126        2098 :         if (!swap_toast_by_content)
    1127             :         {
    1128        1666 :             swaptemp = relform1->reltoastrelid;
    1129        1666 :             relform1->reltoastrelid = relform2->reltoastrelid;
    1130        1666 :             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         168 :         if (RelFileNumberIsValid(relfilenumber1) ||
    1140             :             RelFileNumberIsValid(relfilenumber2))
    1141           0 :             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         168 :         if (relform1->reltablespace != relform2->reltablespace)
    1152           0 :             elog(ERROR, "cannot change tablespace of mapped relation \"%s\"",
    1153             :                  NameStr(relform1->relname));
    1154         168 :         if (relform1->relpersistence != relform2->relpersistence)
    1155           0 :             elog(ERROR, "cannot change persistence of mapped relation \"%s\"",
    1156             :                  NameStr(relform1->relname));
    1157         168 :         if (relform1->relam != relform2->relam)
    1158           0 :             elog(ERROR, "cannot change access method of mapped relation \"%s\"",
    1159             :                  NameStr(relform1->relname));
    1160         168 :         if (!swap_toast_by_content &&
    1161          48 :             (relform1->reltoastrelid || relform2->reltoastrelid))
    1162           0 :             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         168 :         relfilenumber1 = RelationMapOidToFilenumber(r1, relform1->relisshared);
    1169         168 :         if (!RelFileNumberIsValid(relfilenumber1))
    1170           0 :             elog(ERROR, "could not find relation mapping for relation \"%s\", OID %u",
    1171             :                  NameStr(relform1->relname), r1);
    1172         168 :         relfilenumber2 = RelationMapOidToFilenumber(r2, relform2->relisshared);
    1173         168 :         if (!RelFileNumberIsValid(relfilenumber2))
    1174           0 :             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         168 :         RelationMapUpdateMap(r1, relfilenumber2, relform1->relisshared, false);
    1182         168 :         RelationMapUpdateMap(r2, relfilenumber1, relform2->relisshared, false);
    1183             : 
    1184             :         /* Pass OIDs of mapped r2 tables back to caller */
    1185         168 :         *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        2266 :         rel1 = relation_open(r1, NoLock);
    1198        2266 :         rel2 = relation_open(r2, NoLock);
    1199        2266 :         rel2->rd_createSubid = rel1->rd_createSubid;
    1200        2266 :         rel2->rd_newRelfilelocatorSubid = rel1->rd_newRelfilelocatorSubid;
    1201        2266 :         rel2->rd_firstRelfilelocatorSubid = rel1->rd_firstRelfilelocatorSubid;
    1202        2266 :         RelationAssumeNewRelfilelocator(rel1);
    1203        2266 :         relation_close(rel1, NoLock);
    1204        2266 :         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        2266 :     if (relform1->relkind != RELKIND_INDEX)
    1217             :     {
    1218             :         Assert(!TransactionIdIsValid(frozenXid) ||
    1219             :                TransactionIdIsNormal(frozenXid));
    1220        2082 :         relform1->relfrozenxid = frozenXid;
    1221        2082 :         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             : 
    1230        2266 :         swap_pages = relform1->relpages;
    1231        2266 :         relform1->relpages = relform2->relpages;
    1232        2266 :         relform2->relpages = swap_pages;
    1233             : 
    1234        2266 :         swap_tuples = relform1->reltuples;
    1235        2266 :         relform1->reltuples = relform2->reltuples;
    1236        2266 :         relform2->reltuples = swap_tuples;
    1237             : 
    1238        2266 :         swap_allvisible = relform1->relallvisible;
    1239        2266 :         relform1->relallvisible = relform2->relallvisible;
    1240        2266 :         relform2->relallvisible = swap_allvisible;
    1241             :     }
    1242             : 
    1243             :     /*
    1244             :      * Update the tuples in pg_class --- unless the target relation of the
    1245             :      * swap is pg_class itself.  In that case, there is zero point in making
    1246             :      * changes because we'd be updating the old data that we're about to throw
    1247             :      * away.  Because the real work being done here for a mapped relation is
    1248             :      * just to change the relation map settings, it's all right to not update
    1249             :      * the pg_class rows in this case. The most important changes will instead
    1250             :      * performed later, in finish_heap_swap() itself.
    1251             :      */
    1252        2266 :     if (!target_is_pg_class)
    1253             :     {
    1254             :         CatalogIndexState indstate;
    1255             : 
    1256        2228 :         indstate = CatalogOpenIndexes(relRelation);
    1257        2228 :         CatalogTupleUpdateWithInfo(relRelation, &reltup1->t_self, reltup1,
    1258             :                                    indstate);
    1259        2228 :         CatalogTupleUpdateWithInfo(relRelation, &reltup2->t_self, reltup2,
    1260             :                                    indstate);
    1261        2228 :         CatalogCloseIndexes(indstate);
    1262             :     }
    1263             :     else
    1264             :     {
    1265             :         /* no update ... but we do still need relcache inval */
    1266          38 :         CacheInvalidateRelcacheByTuple(reltup1);
    1267          38 :         CacheInvalidateRelcacheByTuple(reltup2);
    1268             :     }
    1269             : 
    1270             :     /*
    1271             :      * Now that pg_class has been updated with its relevant information for
    1272             :      * the swap, update the dependency of the relations to point to their new
    1273             :      * table AM, if it has changed.
    1274             :      */
    1275        2266 :     if (relam1 != relam2)
    1276             :     {
    1277          36 :         if (changeDependencyFor(RelationRelationId,
    1278             :                                 r1,
    1279             :                                 AccessMethodRelationId,
    1280             :                                 relam1,
    1281             :                                 relam2) != 1)
    1282           0 :             elog(ERROR, "could not change access method dependency for relation \"%s.%s\"",
    1283             :                  get_namespace_name(get_rel_namespace(r1)),
    1284             :                  get_rel_name(r1));
    1285          36 :         if (changeDependencyFor(RelationRelationId,
    1286             :                                 r2,
    1287             :                                 AccessMethodRelationId,
    1288             :                                 relam2,
    1289             :                                 relam1) != 1)
    1290           0 :             elog(ERROR, "could not change access method dependency for relation \"%s.%s\"",
    1291             :                  get_namespace_name(get_rel_namespace(r2)),
    1292             :                  get_rel_name(r2));
    1293             :     }
    1294             : 
    1295             :     /*
    1296             :      * Post alter hook for modified relations. The change to r2 is always
    1297             :      * internal, but r1 depends on the invocation context.
    1298             :      */
    1299        2266 :     InvokeObjectPostAlterHookArg(RelationRelationId, r1, 0,
    1300             :                                  InvalidOid, is_internal);
    1301        2266 :     InvokeObjectPostAlterHookArg(RelationRelationId, r2, 0,
    1302             :                                  InvalidOid, true);
    1303             : 
    1304             :     /*
    1305             :      * If we have toast tables associated with the relations being swapped,
    1306             :      * deal with them too.
    1307             :      */
    1308        2266 :     if (relform1->reltoastrelid || relform2->reltoastrelid)
    1309             :     {
    1310         794 :         if (swap_toast_by_content)
    1311             :         {
    1312         184 :             if (relform1->reltoastrelid && relform2->reltoastrelid)
    1313             :             {
    1314             :                 /* Recursively swap the contents of the toast tables */
    1315         184 :                 swap_relation_files(relform1->reltoastrelid,
    1316             :                                     relform2->reltoastrelid,
    1317             :                                     target_is_pg_class,
    1318             :                                     swap_toast_by_content,
    1319             :                                     is_internal,
    1320             :                                     frozenXid,
    1321             :                                     cutoffMulti,
    1322             :                                     mapped_tables);
    1323             :             }
    1324             :             else
    1325             :             {
    1326             :                 /* caller messed up */
    1327           0 :                 elog(ERROR, "cannot swap toast files by content when there's only one");
    1328             :             }
    1329             :         }
    1330             :         else
    1331             :         {
    1332             :             /*
    1333             :              * We swapped the ownership links, so we need to change dependency
    1334             :              * data to match.
    1335             :              *
    1336             :              * NOTE: it is possible that only one table has a toast table.
    1337             :              *
    1338             :              * NOTE: at present, a TOAST table's only dependency is the one on
    1339             :              * its owning table.  If more are ever created, we'd need to use
    1340             :              * something more selective than deleteDependencyRecordsFor() to
    1341             :              * get rid of just the link we want.
    1342             :              */
    1343             :             ObjectAddress baseobject,
    1344             :                         toastobject;
    1345             :             long        count;
    1346             : 
    1347             :             /*
    1348             :              * We disallow this case for system catalogs, to avoid the
    1349             :              * possibility that the catalog we're rebuilding is one of the
    1350             :              * ones the dependency changes would change.  It's too late to be
    1351             :              * making any data changes to the target catalog.
    1352             :              */
    1353         610 :             if (IsSystemClass(r1, relform1))
    1354           0 :                 elog(ERROR, "cannot swap toast files by links for system catalogs");
    1355             : 
    1356             :             /* Delete old dependencies */
    1357         610 :             if (relform1->reltoastrelid)
    1358             :             {
    1359         578 :                 count = deleteDependencyRecordsFor(RelationRelationId,
    1360             :                                                    relform1->reltoastrelid,
    1361             :                                                    false);
    1362         578 :                 if (count != 1)
    1363           0 :                     elog(ERROR, "expected one dependency record for TOAST table, found %ld",
    1364             :                          count);
    1365             :             }
    1366         610 :             if (relform2->reltoastrelid)
    1367             :             {
    1368         610 :                 count = deleteDependencyRecordsFor(RelationRelationId,
    1369             :                                                    relform2->reltoastrelid,
    1370             :                                                    false);
    1371         610 :                 if (count != 1)
    1372           0 :                     elog(ERROR, "expected one dependency record for TOAST table, found %ld",
    1373             :                          count);
    1374             :             }
    1375             : 
    1376             :             /* Register new dependencies */
    1377         610 :             baseobject.classId = RelationRelationId;
    1378         610 :             baseobject.objectSubId = 0;
    1379         610 :             toastobject.classId = RelationRelationId;
    1380         610 :             toastobject.objectSubId = 0;
    1381             : 
    1382         610 :             if (relform1->reltoastrelid)
    1383             :             {
    1384         578 :                 baseobject.objectId = r1;
    1385         578 :                 toastobject.objectId = relform1->reltoastrelid;
    1386         578 :                 recordDependencyOn(&toastobject, &baseobject,
    1387             :                                    DEPENDENCY_INTERNAL);
    1388             :             }
    1389             : 
    1390         610 :             if (relform2->reltoastrelid)
    1391             :             {
    1392         610 :                 baseobject.objectId = r2;
    1393         610 :                 toastobject.objectId = relform2->reltoastrelid;
    1394         610 :                 recordDependencyOn(&toastobject, &baseobject,
    1395             :                                    DEPENDENCY_INTERNAL);
    1396             :             }
    1397             :         }
    1398             :     }
    1399             : 
    1400             :     /*
    1401             :      * If we're swapping two toast tables by content, do the same for their
    1402             :      * valid index. The swap can actually be safely done only if the relations
    1403             :      * have indexes.
    1404             :      */
    1405        2266 :     if (swap_toast_by_content &&
    1406         552 :         relform1->relkind == RELKIND_TOASTVALUE &&
    1407         184 :         relform2->relkind == RELKIND_TOASTVALUE)
    1408             :     {
    1409             :         Oid         toastIndex1,
    1410             :                     toastIndex2;
    1411             : 
    1412             :         /* Get valid index for each relation */
    1413         184 :         toastIndex1 = toast_get_valid_index(r1,
    1414             :                                             AccessExclusiveLock);
    1415         184 :         toastIndex2 = toast_get_valid_index(r2,
    1416             :                                             AccessExclusiveLock);
    1417             : 
    1418         184 :         swap_relation_files(toastIndex1,
    1419             :                             toastIndex2,
    1420             :                             target_is_pg_class,
    1421             :                             swap_toast_by_content,
    1422             :                             is_internal,
    1423             :                             InvalidTransactionId,
    1424             :                             InvalidMultiXactId,
    1425             :                             mapped_tables);
    1426             :     }
    1427             : 
    1428             :     /* Clean up. */
    1429        2266 :     heap_freetuple(reltup1);
    1430        2266 :     heap_freetuple(reltup2);
    1431             : 
    1432        2266 :     table_close(relRelation, RowExclusiveLock);
    1433        2266 : }
    1434             : 
    1435             : /*
    1436             :  * Remove the transient table that was built by make_new_heap, and finish
    1437             :  * cleaning up (including rebuilding all indexes on the old heap).
    1438             :  */
    1439             : void
    1440        1898 : finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
    1441             :                  bool is_system_catalog,
    1442             :                  bool swap_toast_by_content,
    1443             :                  bool check_constraints,
    1444             :                  bool is_internal,
    1445             :                  TransactionId frozenXid,
    1446             :                  MultiXactId cutoffMulti,
    1447             :                  char newrelpersistence)
    1448             : {
    1449             :     ObjectAddress object;
    1450             :     Oid         mapped_tables[4];
    1451             :     int         reindex_flags;
    1452        1898 :     ReindexParams reindex_params = {0};
    1453             :     int         i;
    1454             : 
    1455             :     /* Report that we are now swapping relation files */
    1456        1898 :     pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE,
    1457             :                                  PROGRESS_CLUSTER_PHASE_SWAP_REL_FILES);
    1458             : 
    1459             :     /* Zero out possible results from swapped_relation_files */
    1460        1898 :     memset(mapped_tables, 0, sizeof(mapped_tables));
    1461             : 
    1462             :     /*
    1463             :      * Swap the contents of the heap relations (including any toast tables).
    1464             :      * Also set old heap's relfrozenxid to frozenXid.
    1465             :      */
    1466        1898 :     swap_relation_files(OIDOldHeap, OIDNewHeap,
    1467             :                         (OIDOldHeap == RelationRelationId),
    1468             :                         swap_toast_by_content, is_internal,
    1469             :                         frozenXid, cutoffMulti, mapped_tables);
    1470             : 
    1471             :     /*
    1472             :      * If it's a system catalog, queue a sinval message to flush all catcaches
    1473             :      * on the catalog when we reach CommandCounterIncrement.
    1474             :      */
    1475        1898 :     if (is_system_catalog)
    1476         216 :         CacheInvalidateCatalog(OIDOldHeap);
    1477             : 
    1478             :     /*
    1479             :      * Rebuild each index on the relation (but not the toast table, which is
    1480             :      * all-new at this point).  It is important to do this before the DROP
    1481             :      * step because if we are processing a system catalog that will be used
    1482             :      * during DROP, we want to have its indexes available.  There is no
    1483             :      * advantage to the other order anyway because this is all transactional,
    1484             :      * so no chance to reclaim disk space before commit.  We do not need a
    1485             :      * final CommandCounterIncrement() because reindex_relation does it.
    1486             :      *
    1487             :      * Note: because index_build is called via reindex_relation, it will never
    1488             :      * set indcheckxmin true for the indexes.  This is OK even though in some
    1489             :      * sense we are building new indexes rather than rebuilding existing ones,
    1490             :      * because the new heap won't contain any HOT chains at all, let alone
    1491             :      * broken ones, so it can't be necessary to set indcheckxmin.
    1492             :      */
    1493        1898 :     reindex_flags = REINDEX_REL_SUPPRESS_INDEX_USE;
    1494        1898 :     if (check_constraints)
    1495        1352 :         reindex_flags |= REINDEX_REL_CHECK_CONSTRAINTS;
    1496             : 
    1497             :     /*
    1498             :      * Ensure that the indexes have the same persistence as the parent
    1499             :      * relation.
    1500             :      */
    1501        1898 :     if (newrelpersistence == RELPERSISTENCE_UNLOGGED)
    1502          38 :         reindex_flags |= REINDEX_REL_FORCE_INDEXES_UNLOGGED;
    1503        1860 :     else if (newrelpersistence == RELPERSISTENCE_PERMANENT)
    1504        1780 :         reindex_flags |= REINDEX_REL_FORCE_INDEXES_PERMANENT;
    1505             : 
    1506             :     /* Report that we are now reindexing relations */
    1507        1898 :     pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE,
    1508             :                                  PROGRESS_CLUSTER_PHASE_REBUILD_INDEX);
    1509             : 
    1510        1898 :     reindex_relation(NULL, OIDOldHeap, reindex_flags, &reindex_params);
    1511             : 
    1512             :     /* Report that we are now doing clean up */
    1513        1880 :     pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE,
    1514             :                                  PROGRESS_CLUSTER_PHASE_FINAL_CLEANUP);
    1515             : 
    1516             :     /*
    1517             :      * If the relation being rebuilt is pg_class, swap_relation_files()
    1518             :      * couldn't update pg_class's own pg_class entry (check comments in
    1519             :      * swap_relation_files()), thus relfrozenxid was not updated. That's
    1520             :      * annoying because a potential reason for doing a VACUUM FULL is a
    1521             :      * imminent or actual anti-wraparound shutdown.  So, now that we can
    1522             :      * access the new relation using its indices, update relfrozenxid.
    1523             :      * pg_class doesn't have a toast relation, so we don't need to update the
    1524             :      * corresponding toast relation. Not that there's little point moving all
    1525             :      * relfrozenxid updates here since swap_relation_files() needs to write to
    1526             :      * pg_class for non-mapped relations anyway.
    1527             :      */
    1528        1880 :     if (OIDOldHeap == RelationRelationId)
    1529             :     {
    1530             :         Relation    relRelation;
    1531             :         HeapTuple   reltup;
    1532             :         Form_pg_class relform;
    1533             : 
    1534          38 :         relRelation = table_open(RelationRelationId, RowExclusiveLock);
    1535             : 
    1536          38 :         reltup = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(OIDOldHeap));
    1537          38 :         if (!HeapTupleIsValid(reltup))
    1538           0 :             elog(ERROR, "cache lookup failed for relation %u", OIDOldHeap);
    1539          38 :         relform = (Form_pg_class) GETSTRUCT(reltup);
    1540             : 
    1541          38 :         relform->relfrozenxid = frozenXid;
    1542          38 :         relform->relminmxid = cutoffMulti;
    1543             : 
    1544          38 :         CatalogTupleUpdate(relRelation, &reltup->t_self, reltup);
    1545             : 
    1546          38 :         table_close(relRelation, RowExclusiveLock);
    1547             :     }
    1548             : 
    1549             :     /* Destroy new heap with old filenumber */
    1550        1880 :     object.classId = RelationRelationId;
    1551        1880 :     object.objectId = OIDNewHeap;
    1552        1880 :     object.objectSubId = 0;
    1553             : 
    1554             :     /*
    1555             :      * The new relation is local to our transaction and we know nothing
    1556             :      * depends on it, so DROP_RESTRICT should be OK.
    1557             :      */
    1558        1880 :     performDeletion(&object, DROP_RESTRICT, PERFORM_DELETION_INTERNAL);
    1559             : 
    1560             :     /* performDeletion does CommandCounterIncrement at end */
    1561             : 
    1562             :     /*
    1563             :      * Now we must remove any relation mapping entries that we set up for the
    1564             :      * transient table, as well as its toast table and toast index if any. If
    1565             :      * we fail to do this before commit, the relmapper will complain about new
    1566             :      * permanent map entries being added post-bootstrap.
    1567             :      */
    1568        2048 :     for (i = 0; OidIsValid(mapped_tables[i]); i++)
    1569         168 :         RelationMapRemoveMapping(mapped_tables[i]);
    1570             : 
    1571             :     /*
    1572             :      * At this point, everything is kosher except that, if we did toast swap
    1573             :      * by links, the toast table's name corresponds to the transient table.
    1574             :      * The name is irrelevant to the backend because it's referenced by OID,
    1575             :      * but users looking at the catalogs could be confused.  Rename it to
    1576             :      * prevent this problem.
    1577             :      *
    1578             :      * Note no lock required on the relation, because we already hold an
    1579             :      * exclusive lock on it.
    1580             :      */
    1581        1880 :     if (!swap_toast_by_content)
    1582             :     {
    1583             :         Relation    newrel;
    1584             : 
    1585        1696 :         newrel = table_open(OIDOldHeap, NoLock);
    1586        1696 :         if (OidIsValid(newrel->rd_rel->reltoastrelid))
    1587             :         {
    1588             :             Oid         toastidx;
    1589             :             char        NewToastName[NAMEDATALEN];
    1590             : 
    1591             :             /* Get the associated valid index to be renamed */
    1592         578 :             toastidx = toast_get_valid_index(newrel->rd_rel->reltoastrelid,
    1593             :                                              NoLock);
    1594             : 
    1595             :             /* rename the toast table ... */
    1596         578 :             snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u",
    1597             :                      OIDOldHeap);
    1598         578 :             RenameRelationInternal(newrel->rd_rel->reltoastrelid,
    1599             :                                    NewToastName, true, false);
    1600             : 
    1601             :             /* ... and its valid index too. */
    1602         578 :             snprintf(NewToastName, NAMEDATALEN, "pg_toast_%u_index",
    1603             :                      OIDOldHeap);
    1604             : 
    1605         578 :             RenameRelationInternal(toastidx,
    1606             :                                    NewToastName, true, true);
    1607             : 
    1608             :             /*
    1609             :              * Reset the relrewrite for the toast. The command-counter
    1610             :              * increment is required here as we are about to update the tuple
    1611             :              * that is updated as part of RenameRelationInternal.
    1612             :              */
    1613         578 :             CommandCounterIncrement();
    1614         578 :             ResetRelRewrite(newrel->rd_rel->reltoastrelid);
    1615             :         }
    1616        1696 :         relation_close(newrel, NoLock);
    1617             :     }
    1618             : 
    1619             :     /* if it's not a catalog table, clear any missing attribute settings */
    1620        1880 :     if (!is_system_catalog)
    1621             :     {
    1622             :         Relation    newrel;
    1623             : 
    1624        1664 :         newrel = table_open(OIDOldHeap, NoLock);
    1625        1664 :         RelationClearMissing(newrel);
    1626        1664 :         relation_close(newrel, NoLock);
    1627             :     }
    1628        1880 : }
    1629             : 
    1630             : 
    1631             : /*
    1632             :  * Get a list of tables that the current user has privileges on and
    1633             :  * have indisclustered set.  Return the list in a List * of RelToCluster
    1634             :  * (stored in the specified memory context), each one giving the tableOid
    1635             :  * and the indexOid on which the table is already clustered.
    1636             :  */
    1637             : static List *
    1638          28 : get_tables_to_cluster(MemoryContext cluster_context)
    1639             : {
    1640             :     Relation    indRelation;
    1641             :     TableScanDesc scan;
    1642             :     ScanKeyData entry;
    1643             :     HeapTuple   indexTuple;
    1644             :     Form_pg_index index;
    1645             :     MemoryContext old_context;
    1646          28 :     List       *rtcs = NIL;
    1647             : 
    1648             :     /*
    1649             :      * Get all indexes that have indisclustered set and that the current user
    1650             :      * has the appropriate privileges for.
    1651             :      */
    1652          28 :     indRelation = table_open(IndexRelationId, AccessShareLock);
    1653          28 :     ScanKeyInit(&entry,
    1654             :                 Anum_pg_index_indisclustered,
    1655             :                 BTEqualStrategyNumber, F_BOOLEQ,
    1656             :                 BoolGetDatum(true));
    1657          28 :     scan = table_beginscan_catalog(indRelation, 1, &entry);
    1658          46 :     while ((indexTuple = heap_getnext(scan, ForwardScanDirection)) != NULL)
    1659             :     {
    1660             :         RelToCluster *rtc;
    1661             : 
    1662          18 :         index = (Form_pg_index) GETSTRUCT(indexTuple);
    1663             : 
    1664          18 :         if (!cluster_is_permitted_for_relation(index->indrelid, GetUserId()))
    1665          12 :             continue;
    1666             : 
    1667             :         /* Use a permanent memory context for the result list */
    1668           6 :         old_context = MemoryContextSwitchTo(cluster_context);
    1669             : 
    1670           6 :         rtc = (RelToCluster *) palloc(sizeof(RelToCluster));
    1671           6 :         rtc->tableOid = index->indrelid;
    1672           6 :         rtc->indexOid = index->indexrelid;
    1673           6 :         rtcs = lappend(rtcs, rtc);
    1674             : 
    1675           6 :         MemoryContextSwitchTo(old_context);
    1676             :     }
    1677          28 :     table_endscan(scan);
    1678             : 
    1679          28 :     relation_close(indRelation, AccessShareLock);
    1680             : 
    1681          28 :     return rtcs;
    1682             : }
    1683             : 
    1684             : /*
    1685             :  * Given an index on a partitioned table, return a list of RelToCluster for
    1686             :  * all the children leaves tables/indexes.
    1687             :  *
    1688             :  * Like expand_vacuum_rel, but here caller must hold AccessExclusiveLock
    1689             :  * on the table containing the index.
    1690             :  */
    1691             : static List *
    1692          20 : get_tables_to_cluster_partitioned(MemoryContext cluster_context, Oid indexOid)
    1693             : {
    1694             :     List       *inhoids;
    1695             :     ListCell   *lc;
    1696          20 :     List       *rtcs = NIL;
    1697             :     MemoryContext old_context;
    1698             : 
    1699             :     /* Do not lock the children until they're processed */
    1700          20 :     inhoids = find_all_inheritors(indexOid, NoLock, NULL);
    1701             : 
    1702         104 :     foreach(lc, inhoids)
    1703             :     {
    1704          84 :         Oid         indexrelid = lfirst_oid(lc);
    1705          84 :         Oid         relid = IndexGetRelation(indexrelid, false);
    1706             :         RelToCluster *rtc;
    1707             : 
    1708             :         /* consider only leaf indexes */
    1709          84 :         if (get_rel_relkind(indexrelid) != RELKIND_INDEX)
    1710          38 :             continue;
    1711             : 
    1712             :         /*
    1713             :          * It's possible that the user does not have privileges to CLUSTER the
    1714             :          * leaf partition despite having such privileges on the partitioned
    1715             :          * table.  We skip any partitions which the user is not permitted to
    1716             :          * CLUSTER.
    1717             :          */
    1718          46 :         if (!cluster_is_permitted_for_relation(relid, GetUserId()))
    1719          22 :             continue;
    1720             : 
    1721             :         /* Use a permanent memory context for the result list */
    1722          24 :         old_context = MemoryContextSwitchTo(cluster_context);
    1723             : 
    1724          24 :         rtc = (RelToCluster *) palloc(sizeof(RelToCluster));
    1725          24 :         rtc->tableOid = relid;
    1726          24 :         rtc->indexOid = indexrelid;
    1727          24 :         rtcs = lappend(rtcs, rtc);
    1728             : 
    1729          24 :         MemoryContextSwitchTo(old_context);
    1730             :     }
    1731             : 
    1732          20 :     return rtcs;
    1733             : }
    1734             : 
    1735             : /*
    1736             :  * Return whether userid has privileges to CLUSTER relid.  If not, this
    1737             :  * function emits a WARNING.
    1738             :  */
    1739             : static bool
    1740          94 : cluster_is_permitted_for_relation(Oid relid, Oid userid)
    1741             : {
    1742          94 :     if (pg_class_aclcheck(relid, userid, ACL_MAINTAIN) == ACLCHECK_OK)
    1743          60 :         return true;
    1744             : 
    1745          34 :     ereport(WARNING,
    1746             :             (errmsg("permission denied to cluster \"%s\", skipping it",
    1747             :                     get_rel_name(relid))));
    1748          34 :     return false;
    1749             : }

Generated by: LCOV version 1.14