Allow specifying an access method for partitioned tables

alvherre · alvherre · commit 374c7a229042 · 2024-03-25T16:30:36.000+01:00
It's now possible to specify a table access method via CREATE TABLE ... USING for a partitioned table, as well change it with ALTER TABLE ... SET ACCESS METHOD. Specifying an AM for a partitioned table lets the value be used for all future partitions created under it, closely mirroring the behavior of the TABLESPACE option for partitioned tables. Existing partitions are not modified. For a partitioned table with no AM specified, any new partitions are created with the default_table_access_method. Also add ALTER TABLE ... SET ACCESS METHOD DEFAULT, which reverts to the original state of using the default for new partitions. The relcache of partitioned tables is not changed: rd_tableam is not set, even if a partitioned table has a relam set. Author: Justin Pryzby <pryzby@telsasoft.com> Author: Soumyadeep Chakraborty <soumyadeep2007@gmail.com> Author: Michaël Paquier <michael@paquier.xyz> Reviewed-by: The authors themselves Discussion: https://postgr.es/m/CAE-ML+9zM4wJCGCBGv01k96qQ3gFv4WFcFy=zqPHKeaEFwwv6A@mail.gmail.com Discussion: https://postgr.es/m/20210308010707.GA29832%40telsasoft.com
diff --git a/doc/src/sgml/catalogs.sgml b/doc/src/sgml/catalogs.sgml
@@ -1988,9 +1988,13 @@ SCRAM-SHA-256$<replaceable>&lt;iteration count&gt;</replaceable>:<replaceable>&l
        (references <link linkend="catalog-pg-am"><structname>pg_am</structname></link>.<structfield>oid</structfield>)
       </para>
       <para>
-       If this is a table or an index, the access method used (heap,
-       B-tree, hash, etc.); otherwise zero (zero occurs for sequences,
-       as well as relations without storage, such as views)
+       The access method used to access this table or index.
+       Not meaningful if the relation is a sequence or
+       has no on-disk file,
+       except for partitioned tables, where, if set, it takes
+       precedence over <varname>default_table_access_method</varname>
+       when determining the access method to use for partitions created
+       when one is not specified in the creation command.
       </para></entry>
      </row>
 
diff --git a/doc/src/sgml/ref/alter_table.sgml b/doc/src/sgml/ref/alter_table.sgml
@@ -732,10 +732,20 @@ WITH ( MODULUS <replaceable class="parameter">numeric_literal</replaceable>, REM
     <term><literal>SET ACCESS METHOD</literal></term>
     <listitem>
      <para>
-      This form changes the access method of the table by rewriting it. See
-      <xref linkend="tableam"/> for more information. Writing
-      <literal>DEFAULT</literal> changes the access method of the table
-      to <xref linkend="guc-default-table-access-method"/>.
+      This form changes the access method of the table by rewriting it
+      using the indicated access method; specifying
+      <literal>DEFAULT</literal> selects the access method set as the
+      <xref linkend="guc-default-table-access-method"/> configuration
+      parameter.
+      See <xref linkend="tableam"/> for more information.
+     </para>
+     <para>
+      When applied to a partitioned table, there is no data to rewrite,
+      but partitions created afterwards will default to the given access
+      method unless overridden by a <literal>USING</literal> clause.
+      Specifying <varname>DEFAULT</varname> removes a previous value,
+      causing future partitions to default to
+      <varname>default_table_access_method</varname>.
      </para>
     </listitem>
    </varlistentry>
diff --git a/doc/src/sgml/ref/create_table.sgml b/doc/src/sgml/ref/create_table.sgml
@@ -1365,6 +1365,10 @@ WITH ( MODULUS <replaceable class="parameter">numeric_literal</replaceable>, REM
       method is chosen for the new table. See <xref
       linkend="guc-default-table-access-method"/> for more information.
      </para>
+     <para>
+      When creating a partition, the table access method is the access method
+      of its partitioned table, if set.
+     </para>
     </listitem>
    </varlistentry>
 
diff --git a/src/backend/commands/tablecmds.c b/src/backend/commands/tablecmds.c
@@ -184,7 +184,9 @@ typedef struct AlteredTableInfo
 	List	   *afterStmts;		/* List of utility command parsetrees */
 	bool		verify_new_notnull; /* T if we should recheck NOT NULL */
 	int			rewrite;		/* Reason for forced rewrite, if any */
-	Oid			newAccessMethod;	/* new access method; 0 means no change */
+	bool		chgAccessMethod;	/* T if SET ACCESS METHOD is used */
+	Oid			newAccessMethod;	/* new access method; 0 means no change,
+									 * if above is true */
 	Oid			newTableSpace;	/* new tablespace; 0 means no change */
 	bool		chgPersistence; /* T if SET LOGGED/UNLOGGED is used */
 	char		newrelpersistence;	/* if above is true */
@@ -595,6 +597,7 @@ static ObjectAddress ATExecClusterOn(Relation rel, const char *indexName,
 									 LOCKMODE lockmode);
 static void ATExecDropCluster(Relation rel, LOCKMODE lockmode);
 static void ATPrepSetAccessMethod(AlteredTableInfo *tab, Relation rel, const char *amname);
+static void ATExecSetAccessMethodNoStorage(Relation rel, Oid newAccessMethod);
 static bool ATPrepChangePersistence(Relation rel, bool toLogged);
 static void ATPrepSetTableSpace(AlteredTableInfo *tab, Relation rel,
 								const char *tablespacename, LOCKMODE lockmode);
@@ -709,7 +712,6 @@ DefineRelation(CreateStmt *stmt, char relkind, Oid ownerId,
 	Oid			ofTypeId;
 	ObjectAddress address;
 	LOCKMODE	parentLockmode;
-	const char *accessMethod = NULL;
 	Oid			accessMethodId = InvalidOid;
 
 	/*
@@ -954,24 +956,22 @@ DefineRelation(CreateStmt *stmt, char relkind, Oid ownerId,
 	}
 
 	/*
-	 * If the statement hasn't specified an access method, but we're defining
-	 * a type of relation that needs one, use the default.
+	 * Select access method to use: an explicitly indicated one, or (in the
+	 * case of a partitioned table) the parent's, if it has one.
 	 */
 	if (stmt->accessMethod != NULL)
+		accessMethodId = get_table_am_oid(stmt->accessMethod, false);
+	else if (stmt->partbound)
 	{
-		accessMethod = stmt->accessMethod;
-
-		if (partitioned)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("specifying a table access method is not supported on a partitioned table")));
+		Assert(list_length(inheritOids) == 1);
+		accessMethodId = get_rel_relam(linitial_oid(inheritOids));
 	}
-	else if (RELKIND_HAS_TABLE_AM(relkind))
-		accessMethod = default_table_access_method;
+	else
+		accessMethodId = InvalidOid;
 
-	/* look up the access method, verify it is for a table */
-	if (accessMethod != NULL)
-		accessMethodId = get_table_am_oid(accessMethod, false);
+	/* still nothing? use the default */
+	if (RELKIND_HAS_TABLE_AM(relkind) && !OidIsValid(accessMethodId))
+		accessMethodId = get_table_am_oid(default_table_access_method, false);
 
 	/*
 	 * Create the relation.  Inherited defaults and constraints are passed in
@@ -5047,14 +5047,8 @@ ATPrepCmd(List **wqueue, Relation rel, AlterTableCmd *cmd,
 		case AT_SetAccessMethod:	/* SET ACCESS METHOD */
 			ATSimplePermissions(cmd->subtype, rel, ATT_TABLE | ATT_MATVIEW);
 
-			/* partitioned tables don't have an access method */
-			if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
-				ereport(ERROR,
-						(errcode(ERRCODE_WRONG_OBJECT_TYPE),
-						 errmsg("cannot change access method of a partitioned table")));
-
 			/* check if another access method change was already requested */
-			if (OidIsValid(tab->newAccessMethod))
+			if (tab->chgAccessMethod)
 				ereport(ERROR,
 						(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
 						 errmsg("cannot have multiple SET ACCESS METHOD subcommands")));
@@ -5408,7 +5402,14 @@ ATExecCmd(List **wqueue, AlteredTableInfo *tab,
 			/* nothing to do here, oid columns don't exist anymore */
 			break;
 		case AT_SetAccessMethod:	/* SET ACCESS METHOD */
-			/* handled specially in Phase 3 */
+
+			/*
+			 * Only do this for partitioned tables, for which this is just a
+			 * catalog change.  Tables with storage are handled by Phase 3.
+			 */
+			if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE &&
+				tab->chgAccessMethod)
+				ATExecSetAccessMethodNoStorage(rel, tab->newAccessMethod);
 			break;
 		case AT_SetTableSpace:	/* SET TABLESPACE */
 
@@ -5814,7 +5815,7 @@ ATRewriteTables(AlterTableStmt *parsetree, List **wqueue, LOCKMODE lockmode,
 			 * Select destination access method (same as original unless user
 			 * requested a change)
 			 */
-			if (OidIsValid(tab->newAccessMethod))
+			if (tab->chgAccessMethod)
 				NewAccessMethod = tab->newAccessMethod;
 			else
 				NewAccessMethod = OldHeap->rd_rel->relam;
@@ -6402,6 +6403,7 @@ ATGetQueueEntry(List **wqueue, Relation rel)
 	tab->relkind = rel->rd_rel->relkind;
 	tab->oldDesc = CreateTupleDescCopyConstr(RelationGetDescr(rel));
 	tab->newAccessMethod = InvalidOid;
+	tab->chgAccessMethod = false;
 	tab->newTableSpace = InvalidOid;
 	tab->newrelpersistence = RELPERSISTENCE_PERMANENT;
 	tab->chgPersistence = false;
@@ -15343,25 +15345,128 @@ ATExecDropCluster(Relation rel, LOCKMODE lockmode)
 /*
  * Preparation phase for SET ACCESS METHOD
  *
- * Check that access method exists.  If it is the same as the table's current
- * access method, it is a no-op.  Otherwise, a table rewrite is necessary.
- * If amname is NULL, select default_table_access_method as access method.
+ * Check that the access method exists and determine whether a change is
+ * actually needed.
  */
 static void
 ATPrepSetAccessMethod(AlteredTableInfo *tab, Relation rel, const char *amname)
 {
 	Oid			amoid;
 
-	/* Check that the table access method exists */
-	amoid = get_table_am_oid(amname ? amname : default_table_access_method,
-							 false);
+	/*
+	 * Look up the access method name and check that it differs from the
+	 * table's current AM.  If DEFAULT was specified for a partitioned table
+	 * (amname is NULL), set it to InvalidOid to reset the catalogued AM.
+	 */
+	if (amname != NULL)
+		amoid = get_table_am_oid(amname, false);
+	else if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
+		amoid = InvalidOid;
+	else
+		amoid = get_table_am_oid(default_table_access_method, false);
 
+	/* if it's a match, phase 3 doesn't need to do anything */
 	if (rel->rd_rel->relam == amoid)
 		return;
 
 	/* Save info for Phase 3 to do the real work */
 	tab->rewrite |= AT_REWRITE_ACCESS_METHOD;
 	tab->newAccessMethod = amoid;
+	tab->chgAccessMethod = true;
+}
+
+/*
+ * Special handling of ALTER TABLE SET ACCESS METHOD for relations with no
+ * storage that have an interest in preserving AM.
+ *
+ * Since these have no storage, setting the access method is a catalog only
+ * operation.
+ */
+static void
+ATExecSetAccessMethodNoStorage(Relation rel, Oid newAccessMethodId)
+{
+	Relation	pg_class;
+	Oid			oldAccessMethodId;
+	HeapTuple	tuple;
+	Form_pg_class rd_rel;
+	Oid			reloid = RelationGetRelid(rel);
+
+	/*
+	 * Shouldn't be called on relations having storage; these are processed in
+	 * phase 3.
+	 */
+	Assert(!RELKIND_HAS_STORAGE(rel->rd_rel->relkind));
+
+	/* Get a modifiable copy of the relation's pg_class row. */
+	pg_class = table_open(RelationRelationId, RowExclusiveLock);
+
+	tuple = SearchSysCacheCopy1(RELOID, ObjectIdGetDatum(reloid));
+	if (!HeapTupleIsValid(tuple))
+		elog(ERROR, "cache lookup failed for relation %u", reloid);
+	rd_rel = (Form_pg_class) GETSTRUCT(tuple);
+
+	/* Update the pg_class row. */
+	oldAccessMethodId = rd_rel->relam;
+	rd_rel->relam = newAccessMethodId;
+
+	/* Leave if no update required */
+	if (rd_rel->relam == oldAccessMethodId)
+	{
+		heap_freetuple(tuple);
+		table_close(pg_class, RowExclusiveLock);
+		return;
+	}
+
+	CatalogTupleUpdate(pg_class, &tuple->t_self, tuple);
+
+	/*
+	 * Update the dependency on the new access method.  No dependency is added
+	 * if the new access method is InvalidOid (default case).  Be very careful
+	 * that this has to compare the previous value stored in pg_class with the
+	 * new one.
+	 */
+	if (!OidIsValid(oldAccessMethodId) && OidIsValid(rd_rel->relam))
+	{
+		ObjectAddress relobj,
+					referenced;
+
+		/*
+		 * New access method is defined and there was no dependency
+		 * previously, so record a new one.
+		 */
+		ObjectAddressSet(relobj, RelationRelationId, reloid);
+		ObjectAddressSet(referenced, AccessMethodRelationId, rd_rel->relam);
+		recordDependencyOn(&relobj, &referenced, DEPENDENCY_NORMAL);
+	}
+	else if (OidIsValid(oldAccessMethodId) &&
+			 !OidIsValid(rd_rel->relam))
+	{
+		/*
+		 * There was an access method defined, and no new one, so just remove
+		 * the existing dependency.
+		 */
+		deleteDependencyRecordsForClass(RelationRelationId, reloid,
+										AccessMethodRelationId,
+										DEPENDENCY_NORMAL);
+	}
+	else
+	{
+		Assert(OidIsValid(oldAccessMethodId) &&
+			   OidIsValid(rd_rel->relam));
+
+		/* Both are valid, so update the dependency */
+		changeDependencyFor(RelationRelationId, reloid,
+							AccessMethodRelationId,
+							oldAccessMethodId, rd_rel->relam);
+	}
+
+	/* make the relam and dependency changes visible */
+	CommandCounterIncrement();
+
+	InvokeObjectPostAlterHook(RelationRelationId, RelationGetRelid(rel), 0);
+
+	heap_freetuple(tuple);
+	table_close(pg_class, RowExclusiveLock);
 }
 
 /*
diff --git a/src/backend/utils/cache/lsyscache.c b/src/backend/utils/cache/lsyscache.c
@@ -2069,6 +2069,28 @@ get_rel_persistence(Oid relid)
 	return result;
 }
 
+/*
+ * get_rel_relam
+ *
+ *		Returns the relam associated with a given relation.
+ */
+Oid
+get_rel_relam(Oid relid)
+{
+	HeapTuple	tp;
+	Form_pg_class reltup;
+	Oid			result;
+
+	tp = SearchSysCache1(RELOID, ObjectIdGetDatum(relid));
+	if (!HeapTupleIsValid(tp))
+		elog(ERROR, "cache lookup failed for relation %u", relid);
+	reltup = (Form_pg_class) GETSTRUCT(tp);
+	result = reltup->relam;
+	ReleaseSysCache(tp);
+
+	return result;
+}
+
 
 /*				---------- TRANSFORM CACHE ----------						 */
 
diff --git a/src/backend/utils/cache/relcache.c b/src/backend/utils/cache/relcache.c
@@ -1208,6 +1208,13 @@ RelationBuildDesc(Oid targetRelId, bool insertIt)
 	else if (RELKIND_HAS_TABLE_AM(relation->rd_rel->relkind) ||
 			 relation->rd_rel->relkind == RELKIND_SEQUENCE)
 		RelationInitTableAccessMethod(relation);
+	else if (relation->rd_rel->relkind == RELKIND_PARTITIONED_TABLE)
+	{
+		/*
+		 * Do nothing: access methods are a setting that partitions can
+		 * inherit.
+		 */
+	}
 	else
 		Assert(relation->rd_rel->relam == InvalidOid);
 
diff --git a/src/bin/pg_dump/pg_dump.c b/src/bin/pg_dump/pg_dump.c
@@ -16656,7 +16656,8 @@ dumpTableSchema(Archive *fout, const TableInfo *tbinfo)
 		if (RELKIND_HAS_TABLESPACE(tbinfo->relkind))
 			tablespace = tbinfo->reltablespace;
 
-		if (RELKIND_HAS_TABLE_AM(tbinfo->relkind))
+		if (RELKIND_HAS_TABLE_AM(tbinfo->relkind) ||
+			tbinfo->relkind == RELKIND_PARTITIONED_TABLE)
 			tableam = tbinfo->amname;
 
 		ArchiveEntry(fout, tbinfo->dobj.catId, tbinfo->dobj.dumpId,
diff --git a/src/bin/pg_dump/t/002_pg_dump.pl b/src/bin/pg_dump/t/002_pg_dump.pl
@@ -4587,6 +4587,41 @@
 			no_table_access_method => 1,
 			only_dump_measurement => 1,
 		},
+	},
+
+	# CREATE TABLE with partitioned table and various AMs.  One
+	# partition uses the same default as the parent, and a second
+	# uses its own AM.
+	'CREATE TABLE regress_pg_dump_table_part' => {
+		create_order => 19,
+		create_sql => '
+			CREATE TABLE dump_test.regress_pg_dump_table_am_parent (id int) PARTITION BY LIST (id);
+			ALTER TABLE dump_test.regress_pg_dump_table_am_parent SET ACCESS METHOD regress_table_am;
+			CREATE TABLE dump_test.regress_pg_dump_table_am_child_1
+			  PARTITION OF dump_test.regress_pg_dump_table_am_parent FOR VALUES IN (1) USING heap;
+			CREATE TABLE dump_test.regress_pg_dump_table_am_child_2
+			  PARTITION OF dump_test.regress_pg_dump_table_am_parent FOR VALUES IN (2);',
+		regexp => qr/^
+			\QSET default_table_access_method = regress_table_am;\E
+			(\n(?!SET[^;]+;)[^\n]*)*
+			\n\QCREATE TABLE dump_test.regress_pg_dump_table_am_parent (\E
+			(.*\n)*
+			\QSET default_table_access_method = heap;\E
+			(\n(?!SET[^;]+;)[^\n]*)*
+			\n\QCREATE TABLE dump_test.regress_pg_dump_table_am_child_1 (\E
+			(.*\n)*
+			\QSET default_table_access_method = regress_table_am;\E
+			(\n(?!SET[^;]+;)[^\n]*)*
+			\n\QCREATE TABLE dump_test.regress_pg_dump_table_am_child_2 (\E
+			(.*\n)*/xm,
+		like => {
+			%full_runs, %dump_test_schema_runs, section_pre_data => 1,
+		},
+		unlike => {
+			exclude_dump_test_schema => 1,
+			no_table_access_method => 1,
+			only_dump_measurement => 1,
+		},
 	});
 
 #########################################
diff --git a/src/include/catalog/pg_class.h b/src/include/catalog/pg_class.h
diff --git a/src/include/utils/lsyscache.h b/src/include/utils/lsyscache.h
diff --git a/src/test/regress/expected/create_am.out b/src/test/regress/expected/create_am.out
diff --git a/src/test/regress/sql/create_am.sql b/src/test/regress/sql/create_am.sql
