Merge pull request #3164 from citusdata/propagate_activate

Propagate isactive to metadata nodes.

src/backend/distributed/utils/node_metadata.c

@@ -273,29 +273,51 @@ master_disable_node(PG_FUNCTION_ARGS)
 	WorkerNode *workerNode = ModifiableWorkerNode(nodeName, nodePort);
 	bool isActive = false;
 	bool onlyConsiderActivePlacements = false;
+	MemoryContext savedContext = CurrentMemoryContext;
 
-	if (WorkerNodeIsPrimary(workerNode))
+	PG_TRY();
 	{
-		/*
-		 * Delete reference table placements so they are not taken into account
-		 * for the check if there are placements after this
-		 */
-		DeleteAllReferenceTablePlacementsFromNodeGroup(workerNode->groupId);
-
-		if (NodeGroupHasShardPlacements(workerNode->groupId,
-										onlyConsiderActivePlacements))
+		if (WorkerNodeIsPrimary(workerNode))
 		{
-			ereport(NOTICE, (errmsg(
-								 "Node %s:%d has active shard placements. Some queries "
-								 "may fail after this operation. Use "
-								 "SELECT master_activate_node('%s', %d) to activate this "
-								 "node back.",
-								 workerNode->workerName, nodePort, workerNode->workerName,
-								 nodePort)));
-		}
-	}
+			/*
+			 * Delete reference table placements so they are not taken into account
+			 * for the check if there are placements after this.
+			 */
+			DeleteAllReferenceTablePlacementsFromNodeGroup(workerNode->groupId);
 
-	SetNodeState(nodeName, nodePort, isActive);
+			if (NodeGroupHasShardPlacements(workerNode->groupId,
+											onlyConsiderActivePlacements))
+			{
+				ereport(NOTICE, (errmsg(
+									 "Node %s:%d has active shard placements. Some queries "
+									 "may fail after this operation. Use "
+									 "SELECT master_activate_node('%s', %d) to activate this "
+									 "node back.",
+									 workerNode->workerName, nodePort,
+									 workerNode->workerName,
+									 nodePort)));
+			}
+		}
+
+		SetNodeState(nodeName, nodePort, isActive);
+	}
+	PG_CATCH();
+	{
+		ErrorData *edata = NULL;
+
+		/* CopyErrorData() requires (CurrentMemoryContext != ErrorContext) */
+		MemoryContextSwitchTo(savedContext);
+		edata = CopyErrorData();
+
+		ereport(ERROR, (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+						errmsg("Disabling %s:%d failed", workerNode->workerName,
+							   nodePort),
+						errdetail("%s", edata->message),
+						errhint(
+							"If you are using MX, try stop_metadata_sync_to_node(hostname, port) "
+							"for nodes that are down before disabling them.")));
+	}
+	PG_END_TRY();
 
 	PG_RETURN_VOID();
 }
@@ -350,7 +372,7 @@ SetUpDistributedTableDependencies(WorkerNode *newWorkerNode)
 										  newWorkerNode->workerPort);
 
 		/*
-		 * Let the maintanince deamon do the hard work of syncing the metadata.
+		 * Let the maintenance daemon do the hard work of syncing the metadata.
 		 * We prefer this because otherwise node activation might fail within
 		 * transaction blocks.
 		 */
@@ -1129,8 +1151,8 @@ SetWorkerColumn(WorkerNode *workerNode, int columnIndex, Datum value)
 	{
 		case Anum_pg_dist_node_isactive:
 		{
-			metadataSyncCommand = ShouldHaveShardsUpdateCommand(workerNode->nodeId,
-																DatumGetBool(value));
+			metadataSyncCommand = NodeStateUpdateCommand(workerNode->nodeId,
+														 DatumGetBool(value));
 			break;
 		}
 

src/backend/distributed/utils/reference_table_utils.c

@@ -406,8 +406,7 @@ CreateReferenceTableColocationId()
 /*
  * DeleteAllReferenceTablePlacementsFromNodeGroup function iterates over list of reference
  * tables and deletes all reference table placements from pg_dist_placement table
- * for given group. However, it does not modify replication factor of the colocation
- * group of reference tables. It is caller's responsibility to do that if it is necessary.
+ * for given group.
  */
 void
 DeleteAllReferenceTablePlacementsFromNodeGroup(int32 groupId)

src/test/regress/expected/multi_mx_node_metadata.out

@@ -210,7 +210,7 @@ SELECT nodeid, hasmetadata, metadatasynced FROM pg_dist_node;
 --------------------------------------------------------------------------
 -- Test updating a node when another node is in readonly-mode
 --------------------------------------------------------------------------
-SELECT FROM master_add_node('localhost', :worker_2_port) AS nodeid_2 \gset
+SELECT master_add_node('localhost', :worker_2_port) AS nodeid_2 \gset
 NOTICE:  Replicating reference table "ref_table" to the node localhost:57638
 SELECT 1 FROM start_metadata_sync_to_node('localhost', :worker_2_port);
  ?column? 
@@ -403,8 +403,166 @@ SELECT verify_metadata('localhost', :worker_1_port),
  t               | t
 (1 row)
 
+--------------------------------------------------------------------------
+-- Test that changes in isactive is propagated to the metadata nodes
+--------------------------------------------------------------------------
+-- Don't drop the reference table so it has shards on the nodes being disabled
+DROP TABLE dist_table_1, dist_table_2;
+SELECT 1 FROM master_disable_node('localhost', :worker_2_port);
+ ?column? 
+----------
+        1
+(1 row)
+
+SELECT verify_metadata('localhost', :worker_1_port);
+ verify_metadata 
+-----------------
+ t
+(1 row)
+
+SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
+NOTICE:  Replicating reference table "ref_table" to the node localhost:57638
+ ?column? 
+----------
+        1
+(1 row)
+
+SELECT verify_metadata('localhost', :worker_1_port);
+ verify_metadata 
+-----------------
+ t
+(1 row)
+
+------------------------------------------------------------------------------------
+-- Test master_disable_node() when the node that is being disabled is actually down
+------------------------------------------------------------------------------------
+SELECT master_update_node(:nodeid_2, 'localhost', 1);
+ master_update_node 
+--------------------
+ 
+(1 row)
+
+SELECT wait_until_metadata_sync();
+ wait_until_metadata_sync 
+--------------------------
+ 
+(1 row)
+
+-- set metadatasynced so we try porpagating metadata changes
+UPDATE pg_dist_node SET metadatasynced = TRUE WHERE nodeid IN (:nodeid_1, :nodeid_2);
+-- should error out
+SELECT 1 FROM master_disable_node('localhost', 1);
+ERROR:  Disabling localhost:1 failed
+DETAIL:  connection error: localhost:1
+HINT:  If you are using MX, try stop_metadata_sync_to_node(hostname, port) for nodes that are down before disabling them.
+-- try again after stopping metadata sync
+SELECT stop_metadata_sync_to_node('localhost', 1);
+ stop_metadata_sync_to_node 
+----------------------------
+ 
+(1 row)
+
+SELECT 1 FROM master_disable_node('localhost', 1);
+ ?column? 
+----------
+        1
+(1 row)
+
+SELECT verify_metadata('localhost', :worker_1_port);
+ verify_metadata 
+-----------------
+ t
+(1 row)
+
+SELECT master_update_node(:nodeid_2, 'localhost', :worker_2_port);
+ master_update_node 
+--------------------
+ 
+(1 row)
+
+SELECT wait_until_metadata_sync();
+ wait_until_metadata_sync 
+--------------------------
+ 
+(1 row)
+
+SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
+NOTICE:  Replicating reference table "ref_table" to the node localhost:57638
+ ?column? 
+----------
+        1
+(1 row)
+
+SELECT verify_metadata('localhost', :worker_1_port);
+ verify_metadata 
+-----------------
+ t
+(1 row)
+
+------------------------------------------------------------------------------------
+-- Test master_disable_node() when the other node is down
+------------------------------------------------------------------------------------
+-- node 1 is down.
+SELECT master_update_node(:nodeid_1, 'localhost', 1);
+ master_update_node 
+--------------------
+ 
+(1 row)
+
+SELECT wait_until_metadata_sync();
+ wait_until_metadata_sync 
+--------------------------
+ 
+(1 row)
+
+-- set metadatasynced so we try porpagating metadata changes
+UPDATE pg_dist_node SET metadatasynced = TRUE WHERE nodeid IN (:nodeid_1, :nodeid_2);
+-- should error out
+SELECT 1 FROM master_disable_node('localhost', :worker_2_port);
+ERROR:  Disabling localhost:57638 failed
+DETAIL:  connection error: localhost:1
+HINT:  If you are using MX, try stop_metadata_sync_to_node(hostname, port) for nodes that are down before disabling them.
+-- try again after stopping metadata sync
+SELECT stop_metadata_sync_to_node('localhost', 1);
+ stop_metadata_sync_to_node 
+----------------------------
+ 
+(1 row)
+
+SELECT 1 FROM master_disable_node('localhost', :worker_2_port);
+ ?column? 
+----------
+        1
+(1 row)
+
+-- bring up node 1
+SELECT master_update_node(:nodeid_1, 'localhost', :worker_1_port);
+ master_update_node 
+--------------------
+ 
+(1 row)
+
+SELECT wait_until_metadata_sync();
+ wait_until_metadata_sync 
+--------------------------
+ 
+(1 row)
+
+SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
+NOTICE:  Replicating reference table "ref_table" to the node localhost:57638
+ ?column? 
+----------
+        1
+(1 row)
+
+SELECT verify_metadata('localhost', :worker_1_port);
+ verify_metadata 
+-----------------
+ t
+(1 row)
+
 -- cleanup
-DROP TABLE dist_table_1, ref_table, dist_table_2;
+DROP TABLE ref_table;
 TRUNCATE pg_dist_colocation;
 SELECT count(*) FROM (SELECT master_remove_node(nodename, nodeport) FROM pg_dist_node) t;
  count 

src/test/regress/multi_mx_schedule

@@ -14,7 +14,7 @@
 # Tests around schema changes, these are run first, so there's no preexisting objects.
 # ---
 test: multi_extension
-test: multi_mx_master_update_node
+test: multi_mx_node_metadata
 test: multi_cluster_management
 test: multi_test_helpers
 

src/test/regress/sql/multi_mx_node_metadata.sql

@@ -121,7 +121,7 @@ SELECT nodeid, hasmetadata, metadatasynced FROM pg_dist_node;
 -- Test updating a node when another node is in readonly-mode
 --------------------------------------------------------------------------
 
-SELECT FROM master_add_node('localhost', :worker_2_port) AS nodeid_2 \gset
+SELECT master_add_node('localhost', :worker_2_port) AS nodeid_2 \gset
 SELECT 1 FROM start_metadata_sync_to_node('localhost', :worker_2_port);
 
 -- Create a table with shards on both nodes
@@ -197,8 +197,70 @@ SELECT nodeid, hasmetadata, metadatasynced FROM pg_dist_node ORDER BY nodeid;
 SELECT verify_metadata('localhost', :worker_1_port),
        verify_metadata('localhost', :worker_2_port);
 
+--------------------------------------------------------------------------
+-- Test that changes in isactive is propagated to the metadata nodes
+--------------------------------------------------------------------------
+-- Don't drop the reference table so it has shards on the nodes being disabled
+DROP TABLE dist_table_1, dist_table_2;
+
+SELECT 1 FROM master_disable_node('localhost', :worker_2_port);
+SELECT verify_metadata('localhost', :worker_1_port);
+
+SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
+SELECT verify_metadata('localhost', :worker_1_port);
+
+------------------------------------------------------------------------------------
+-- Test master_disable_node() when the node that is being disabled is actually down
+------------------------------------------------------------------------------------
+SELECT master_update_node(:nodeid_2, 'localhost', 1);
+SELECT wait_until_metadata_sync();
+
+-- set metadatasynced so we try porpagating metadata changes
+UPDATE pg_dist_node SET metadatasynced = TRUE WHERE nodeid IN (:nodeid_1, :nodeid_2);
+
+-- should error out
+SELECT 1 FROM master_disable_node('localhost', 1);
+
+-- try again after stopping metadata sync
+SELECT stop_metadata_sync_to_node('localhost', 1);
+SELECT 1 FROM master_disable_node('localhost', 1);
+
+SELECT verify_metadata('localhost', :worker_1_port);
+
+SELECT master_update_node(:nodeid_2, 'localhost', :worker_2_port);
+SELECT wait_until_metadata_sync();
+
+SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
+SELECT verify_metadata('localhost', :worker_1_port);
+
+
+------------------------------------------------------------------------------------
+-- Test master_disable_node() when the other node is down
+------------------------------------------------------------------------------------
+-- node 1 is down.
+SELECT master_update_node(:nodeid_1, 'localhost', 1);
+SELECT wait_until_metadata_sync();
+
+-- set metadatasynced so we try porpagating metadata changes
+UPDATE pg_dist_node SET metadatasynced = TRUE WHERE nodeid IN (:nodeid_1, :nodeid_2);
+
+-- should error out
+SELECT 1 FROM master_disable_node('localhost', :worker_2_port);
+
+-- try again after stopping metadata sync
+SELECT stop_metadata_sync_to_node('localhost', 1);
+SELECT 1 FROM master_disable_node('localhost', :worker_2_port);
+
+-- bring up node 1
+SELECT master_update_node(:nodeid_1, 'localhost', :worker_1_port);
+SELECT wait_until_metadata_sync();
+
+SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
+
+SELECT verify_metadata('localhost', :worker_1_port);
+
 -- cleanup
-DROP TABLE dist_table_1, ref_table, dist_table_2;
+DROP TABLE ref_table;
 TRUNCATE pg_dist_colocation;
 SELECT count(*) FROM (SELECT master_remove_node(nodename, nodeport) FROM pg_dist_node) t;
 ALTER SEQUENCE pg_catalog.pg_dist_groupid_seq RESTART :last_group_id;