Only issue LOCK TABLE commands when necessary

sfrost · sfrost · commit e1b120a8cbb0 · 2016-05-06T14:06:50.000-04:00
Reviewing the cases where we need to LOCK a given table during a dump,
it was pointed out by Tom that we really don't need to LOCK a table if
we are only looking to dump the ACL for it, or certain other
components.  After reviewing the queries run for all of the component
pieces, a list of components were determined to not require LOCK'ing
of the table.

This implements a check to avoid LOCK'ing those tables.

Initial complaint from Rushabh Lathia, discussed with Robert and Tom,
the patch is mine.
diff --git a/src/bin/pg_dump/pg_dump.c b/src/bin/pg_dump/pg_dump.c
@@ -5953,8 +5953,11 @@ getTables(Archive *fout, int *numTables)
 		 *
 		 * NOTE: it'd be kinda nice to lock other relations too, not only
 		 * plain tables, but the backend doesn't presently allow that.
+		 *
+		 * We only need to lock the table for certain components; see pg_dump.h
 		 */
-		if (tblinfo[i].dobj.dump && tblinfo[i].relkind == RELKIND_RELATION)
+		if (tblinfo[i].dobj.dump && tblinfo[i].relkind == RELKIND_RELATION &&
+				(tblinfo[i].dobj.dump & DUMP_COMPONENTS_REQUIRING_LOCK))
 		{
 			resetPQExpBuffer(query);
 			appendPQExpBuffer(query,
diff --git a/src/bin/pg_dump/pg_dump.h b/src/bin/pg_dump/pg_dump.h
@@ -93,6 +93,33 @@ typedef uint32 DumpComponents;	/* a bitmask of dump object components */
 #define DUMP_COMPONENT_USERMAP		(1 << 6)
 #define DUMP_COMPONENT_ALL			(0xFFFF)
 
+/*
+ * component types which require us to obtain a lock on the table
+ *
+ * Note that some components only require looking at the information
+ * in the pg_catalog tables and, for those components, we do not need
+ * to lock the table.  Be careful here though- some components use
+ * server-side functions which pull the latest information from
+ * SysCache and in those cases we *do* need to lock the table.
+ *
+ * We do not need locks for the COMMENT and SECLABEL components as
+ * those simply query their associated tables without using any
+ * server-side functions.  We do not need locks for the ACL component
+ * as we pull that information from pg_class without using any
+ * server-side functions that use SysCache.  The USERMAP component
+ * is only relevant for FOREIGN SERVERs and not tables, so no sense
+ * locking a table for that either (that can happen if we are going
+ * to dump "ALL" components for a table).
+ *
+ * We DO need locks for DEFINITION, due to various server-side
+ * functions that are used and POLICY due to pg_get_expr().  We set
+ * this up to grab the lock except in the cases we know to be safe.
+ */
+#define DUMP_COMPONENTS_REQUIRING_LOCK (\
+		DUMP_COMPONENT_DEFINITION |\
+		DUMP_COMPONENT_DATA |\
+		DUMP_COMPONENT_POLICY)
+
 typedef struct _dumpableObject
 {
 	DumpableObjectType objType;

Original file line number	Diff line number	Diff line change
`@@ -5953,8 +5953,11 @@ getTables(Archive fout, int numTables)`
`5953`	`5953`	`*`
`5954`	`5954`	`* NOTE: it'd be kinda nice to lock other relations too, not only`
`5955`	`5955`	`* plain tables, but the backend doesn't presently allow that.`
	`5956`	`+ *`
	`5957`	`+ * We only need to lock the table for certain components; see pg_dump.h`
`5956`	`5958`	`*/`
`5957`		`- if (tblinfo[i].dobj.dump && tblinfo[i].relkind == RELKIND_RELATION)`
	`5959`	`+ if (tblinfo[i].dobj.dump && tblinfo[i].relkind == RELKIND_RELATION &&`
	`5960`	`+ (tblinfo[i].dobj.dump & DUMP_COMPONENTS_REQUIRING_LOCK))`
`5958`	`5961`	`{`
`5959`	`5962`	`resetPQExpBuffer(query);`
`5960`	`5963`	`appendPQExpBuffer(query,`