[CF 5004] v5 - CREATE INDEX CONCURRENTLY for partitioned tables

This branch was automatically generated by a robot using patches from an
email thread registered at:

https://commitfest.postgresql.org/patch/5004

The branch will be overwritten each time a new patch version is posted to
the thread, and also periodically to check for bitrot caused by changes
on the master branch.

Patch(es): https://www.postgresql.org/message-id/428008d6-911d-4a14-986f-0fbb74db6fd8@gmail.com
Author(s): Justin Pryzby, Ilya Gladyshev

Author: Commitfest Bot

doc/src/sgml/ddl.sgml

@@ -4410,14 +4410,12 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
-     use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
+     indexed, but any future partitions will be as well. For more control over
+     locking of the partitions you can use <command>CREATE INDEX ON ONLY</command>
+     on the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
-     on each partition using <literal>CONCURRENTLY</literal> and
+     on each partition and
      <firstterm>attached</firstterm> to the partitioned index on the parent
      using <command>ALTER INDEX ... ATTACH PARTITION</command>.  Once indexes for
      all the partitions are attached to the parent index, the parent index will

doc/src/sgml/ref/create_index.sgml

@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 

src/backend/commands/indexcmds.c

@@ -98,6 +98,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -573,20 +578,17 @@ DefineIndex(Oid tableId,
 	amoptions_function amoptions;
 	bool		exclusion;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
 	bits16		flags;
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -724,20 +726,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1172,10 +1160,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1240,6 +1224,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1557,21 +1546,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1583,37 +1558,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1624,6 +1611,141 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Partition could have been dropped, since we looked it up. In
+			 * this case consider it done and go to the next one.
+			 */
+			tabrelid = IndexGetRelation(indrelid, true);
+			if (!tabrelid)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+			rel = try_table_open(tabrelid, ShareUpdateExclusiveLock);
+			if (!rel)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+			{
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			heaprelid = rel->rd_lockInfo.lockRelId;
+
+			/*
+			 * Close the table but retain the lock, that should be extended to
+			 * session level in DefineIndexConcurrentInternal.
+			 */
+			table_close(rel, NoLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Relation	indrel = try_index_open(indrelid, ShareUpdateExclusiveLock);
+
+			if (!indrel)
+				continue;
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+			index_close(indrel, ShareUpdateExclusiveLock);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1827,10 +1949,6 @@ DefineIndex(Oid tableId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }