Merge pull request #5597 from citusdata/move_placement_deletions

Move placement deletion from disable node to activate node
2022-01-07 10:03:35 +01:00 · 2022-01-07 10:03:35 +01:00 · b6cf5a969b
parent 9edfbe7718 9f2d9e1487
commit b6cf5a969b
10 changed files with 98 additions and 53 deletions
--- a/src/backend/distributed/metadata/node_metadata.c
+++ b/src/backend/distributed/metadata/node_metadata.c
@ -496,16 +496,6 @@ citus_disable_node(PG_FUNCTION_ARGS)
 								 workerNode->workerName,
 								 nodePort)));
 		}
-
-		/*
-		 * Delete replicated table placements from the coordinator's metadata,
-		 * but not remotely. That is because one more more of the remote
-		 * nodes might be down. Instead, we let the background worker
-		 * to sync the metadata when possible.
-		 */
-		bool forceRemoteDelete = false;
-		DeleteAllReplicatedTablePlacementsFromNodeGroup(workerNode->groupId,
-														forceRemoteDelete);
 	}

 	TransactionModifiedNodeMetadata = true;
@ -515,6 +505,12 @@ citus_disable_node(PG_FUNCTION_ARGS)
 	 * active nodes get the metadata updates. We defer this operation to the
 	 * background worker to make it possible disabling nodes when multiple nodes
 	 * are down.
+	 *
+	 * Note that the active placements reside on the active nodes. Hence, when
+	 * Citus finds active placements, it filters out the placements that are on
+	 * the disabled nodes. That's why, we don't have to change/sync placement
+	 * metadata at this point. Instead, we defer that to citus_activate_node()
+	 * where we expect all nodes up and running.
 	 */
 	if (UnsetMetadataSyncedForAll())
 	{
@ -868,6 +864,22 @@ ActivateNode(char *nodeName, int nodePort)
 		ereport(ERROR, (errmsg("node at \"%s:%u\" does not exist", nodeName, nodePort)));
 	}

+	/*
+	 * Delete existing reference and replicated table placements on the
+	 * given groupId if the group has been disabled earlier (e.g., isActive
+	 * set to false).
+	 *
+	 * Sync the metadata changes to all existing metadata nodes irrespective
+	 * of the current nodes' metadata sync state. We expect all nodes up
+	 * and running when another node is activated.
+	 */
+	if (!workerNode->isActive && NodeIsPrimary(workerNode))
+	{
+		bool localOnly = false;
+		DeleteAllReplicatedTablePlacementsFromNodeGroup(workerNode->groupId,
+														localOnly);
+	}
+
 	workerNode =
 		SetWorkerColumnLocalOnly(workerNode, Anum_pg_dist_node_isactive,
 								 BoolGetDatum(isActive));
@ -1353,9 +1365,9 @@ RemoveNodeFromCluster(char *nodeName, int32 nodePort)
 		 * Delete reference table placements so they are not taken into account
 		 * for the check if there are placements after this.
 		 */
-		bool forceRemoteDelete = true;
+		bool localOnly = false;
 		DeleteAllReplicatedTablePlacementsFromNodeGroup(workerNode->groupId,
-														forceRemoteDelete);
+														localOnly);

 		/*
 		 * Secondary nodes are read-only, never 2PC is used.
--- a/src/backend/distributed/utils/reference_table_utils.c
+++ b/src/backend/distributed/utils/reference_table_utils.c
@ -432,7 +432,7 @@ CreateReferenceTableColocationId()
 * all placements from pg_dist_placement table for given group.
 */
 void
-DeleteAllReplicatedTablePlacementsFromNodeGroup(int32 groupId, bool forceRemoteDelete)
+DeleteAllReplicatedTablePlacementsFromNodeGroup(int32 groupId, bool localOnly)
 {
 	List *referenceTableList = CitusTableTypeIdList(REFERENCE_TABLE);
 	List *replicatedMetadataSyncedDistributedTableList =
@ -469,7 +469,7 @@ DeleteAllReplicatedTablePlacementsFromNodeGroup(int32 groupId, bool forceRemoteD

 			DeleteShardPlacementRow(placement->placementId);

-			if (forceRemoteDelete)
+			if (!localOnly)
 			{
 				resetStringInfo(deletePlacementCommand);
 				appendStringInfo(deletePlacementCommand,
--- a/src/include/distributed/reference_table_utils.h
+++ b/src/include/distributed/reference_table_utils.h
@ -21,8 +21,8 @@
 extern void EnsureReferenceTablesExistOnAllNodes(void);
 extern void EnsureReferenceTablesExistOnAllNodesExtended(char transferMode);
 extern uint32 CreateReferenceTableColocationId(void);
-extern void DeleteAllReplicatedTablePlacementsFromNodeGroup(int32 groupId, bool
-															forceRemoteDelete);
+extern void DeleteAllReplicatedTablePlacementsFromNodeGroup(int32 groupId,
+															bool localOnly);
 extern int CompareOids(const void *leftElement, const void *rightElement);
 extern int ReferenceTableReplicationFactor(void);
 extern void ReplicateAllReferenceTablesToNode(char *nodeName, int nodePort);
--- a/src/test/regress/expected/failure_add_disable_node.out
+++ b/src/test/regress/expected/failure_add_disable_node.out
@ -74,8 +74,8 @@ ORDER BY 1, 2;
 (1 row)

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;
 shardid | shardstate
 ---------------------------------------------------------------------
@ -91,8 +91,8 @@ ORDER BY 1, 2;
 (1 row)

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;
 shardid | shardstate
 ---------------------------------------------------------------------
@ -120,8 +120,8 @@ ORDER BY 1, 2;
 (1 row)

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;
 shardid | shardstate
 ---------------------------------------------------------------------
@ -157,8 +157,8 @@ ORDER BY 1, 2;
 (1 row)

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;
 shardid | shardstate
 ---------------------------------------------------------------------
@ -181,8 +181,8 @@ SELECT master_remove_node('localhost', :worker_2_proxy_port);
 (1 row)

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;
 shardid | shardstate
 ---------------------------------------------------------------------
@ -212,8 +212,8 @@ ORDER BY 1, 2;
 (2 rows)

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;
 shardid | shardstate
 ---------------------------------------------------------------------
--- a/src/test/regress/expected/multi_cluster_management.out
+++ b/src/test/regress/expected/multi_cluster_management.out
@ -510,6 +510,7 @@ ERROR:  there is a shard placement in node group 6 but there are no nodes in tha
 SELECT * INTO old_placements FROM pg_dist_placement WHERE groupid = :worker_2_group;
 DELETE FROM pg_dist_placement WHERE groupid = :worker_2_group;
 SELECT master_add_node('localhost', :worker_2_port) AS new_node \gset
+WARNING:  could not find any shard placements for shardId 1220001
 WARNING:  citus.enable_object_propagation is off, not creating distributed objects on worker
 DETAIL:  distributed objects are only kept in sync when citus.enable_object_propagation is set to on. Newly activated nodes will not get these objects created
 WARNING:  could not find any shard placements for shardId 1220001
--- a/src/test/regress/expected/multi_mx_node_metadata.out
+++ b/src/test/regress/expected/multi_mx_node_metadata.out
@ -641,19 +641,32 @@ SELECT verify_metadata('localhost', :worker_1_port),
 ---------------------------------------------------------------------
 -- Don't drop the reference table so it has shards on the nodes being disabled
 DROP TABLE dist_table_1, dist_table_2;
-SELECT pg_catalog.citus_disable_node_and_wait('localhost', :worker_2_port);
+SELECT pg_catalog.citus_disable_node('localhost', :worker_2_port);
 NOTICE:  Node localhost:xxxxx has active shard placements. Some queries may fail after this operation. Use SELECT citus_activate_node('localhost', 57638) to activate this node back.
- citus_disable_node_and_wait
+ citus_disable_node
 ---------------------------------------------------------------------

 (1 row)

-SELECT verify_metadata('localhost', :worker_1_port);
- verify_metadata
+SELECT wait_until_metadata_sync(30000);
+ wait_until_metadata_sync
 ---------------------------------------------------------------------
- t
+
 (1 row)

+-- show that node metadata is the same
+-- note that we cannot use verify_metadata here
+-- because there are several shards/placements
+-- in the metadata that are manually modified on the coordinator
+-- not on the worker, and pg_catalog.citus_disable_node does
+-- not sync the metadata
+SELECT result FROM run_command_on_workers($$SELECT jsonb_agg(row_to_json(row(pg_dist_node.*))) FROM pg_dist_node$$) WHERE nodeport=:worker_1_port
+EXCEPT
+SELECT jsonb_agg(row_to_json(row(pg_dist_node.*)))::text FROM pg_dist_node;
+ result
+---------------------------------------------------------------------
+(0 rows)
+
 SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
 ?column?
 ---------------------------------------------------------------------
--- a/src/test/regress/expected/replicated_table_disable_node.out
+++ b/src/test/regress/expected/replicated_table_disable_node.out
@ -32,16 +32,20 @@ SELECT public.wait_until_metadata_sync();

 (1 row)

-- the placement should be removed both from the coordinator
+-- the active placement should NOT be removed from the coordinator
 -- and from the workers
-SELECT count(*) FROM pg_dist_placement WHERE shardid IN (101500, 101501, 101502);
+SELECT count(*) FROM pg_dist_placement p JOIN pg_dist_node n USING(groupid)
+	WHERE n.isactive AND n.noderole = 'primary'
+		  AND p.shardid IN (101500, 101501, 101502);
 count
 ---------------------------------------------------------------------
     3
 (1 row)

 \c - - - :worker_1_port
-SELECT count(*) FROM pg_dist_placement WHERE shardid IN (101500, 101501, 101502);
+SELECT count(*) FROM pg_dist_placement p JOIN pg_dist_node n USING(groupid)
+	WHERE n.isactive AND n.noderole = 'primary'
+		  AND p.shardid IN (101500, 101501, 101502);
 count
 ---------------------------------------------------------------------
     3
--- a/src/test/regress/sql/failure_add_disable_node.sql
+++ b/src/test/regress/sql/failure_add_disable_node.sql
@ -37,8 +37,8 @@ SELECT * FROM master_get_active_worker_nodes()
 ORDER BY 1, 2;

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;

 -- verify node is not activated
@ -46,8 +46,8 @@ SELECT * FROM master_get_active_worker_nodes()
 ORDER BY 1, 2;

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;

 -- fail create schema command
@ -60,8 +60,8 @@ SELECT * FROM master_get_active_worker_nodes()
 ORDER BY 1, 2;

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;

 BEGIN;
@ -81,8 +81,8 @@ SELECT * FROM master_get_active_worker_nodes()
 ORDER BY 1, 2;

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;

 -- test master_add_inactive_node
@ -93,8 +93,8 @@ SELECT master_add_inactive_node('localhost', :worker_2_proxy_port);
 SELECT master_remove_node('localhost', :worker_2_proxy_port);

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;

 -- reset cluster to original state
@ -106,8 +106,8 @@ SELECT * FROM master_get_active_worker_nodes()
 ORDER BY 1, 2;

 SELECT shardid, shardstate
-FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid)
-WHERE s.logicalrelid = 'user_table'::regclass
+FROM pg_dist_placement p JOIN pg_dist_shard s USING (shardid) JOIN pg_dist_node n USING(groupid)
+WHERE s.logicalrelid = 'user_table'::regclass AND n.isactive
 ORDER BY placementid;

 SELECT citus.mitmproxy('conn.allow()');
--- a/src/test/regress/sql/multi_mx_node_metadata.sql
+++ b/src/test/regress/sql/multi_mx_node_metadata.sql
@ -286,8 +286,18 @@ SELECT verify_metadata('localhost', :worker_1_port),
 -- Don't drop the reference table so it has shards on the nodes being disabled
 DROP TABLE dist_table_1, dist_table_2;

-SELECT pg_catalog.citus_disable_node_and_wait('localhost', :worker_2_port);
-SELECT verify_metadata('localhost', :worker_1_port);
+SELECT pg_catalog.citus_disable_node('localhost', :worker_2_port);
+SELECT wait_until_metadata_sync(30000);
+
+-- show that node metadata is the same
+-- note that we cannot use verify_metadata here
+-- because there are several shards/placements
+-- in the metadata that are manually modified on the coordinator
+-- not on the worker, and pg_catalog.citus_disable_node does
+-- not sync the metadata
+SELECT result FROM run_command_on_workers($$SELECT jsonb_agg(row_to_json(row(pg_dist_node.*))) FROM pg_dist_node$$) WHERE nodeport=:worker_1_port
+EXCEPT
+SELECT jsonb_agg(row_to_json(row(pg_dist_node.*)))::text FROM pg_dist_node;

 SELECT 1 FROM master_activate_node('localhost', :worker_2_port);
 SELECT verify_metadata('localhost', :worker_1_port);
--- a/src/test/regress/sql/replicated_table_disable_node.sql
+++ b/src/test/regress/sql/replicated_table_disable_node.sql
@ -18,12 +18,17 @@ INSERT INTO ref SELECT i,i FROM generate_series(0,10)i;
 SELECT citus_disable_node('localhost', :worker_2_port, true);
 SELECT public.wait_until_metadata_sync();

-- the placement should be removed both from the coordinator
+-- the active placement should NOT be removed from the coordinator
 -- and from the workers
-SELECT count(*) FROM pg_dist_placement WHERE shardid IN (101500, 101501, 101502);
+SELECT count(*) FROM pg_dist_placement p JOIN pg_dist_node n USING(groupid)
+	WHERE n.isactive AND n.noderole = 'primary'
+		  AND p.shardid IN (101500, 101501, 101502);

 \c - - - :worker_1_port
-SELECT count(*) FROM pg_dist_placement WHERE shardid IN (101500, 101501, 101502);
+SELECT count(*) FROM pg_dist_placement p JOIN pg_dist_node n USING(groupid)
+	WHERE n.isactive AND n.noderole = 'primary'
+		  AND p.shardid IN (101500, 101501, 101502);
+
 SET search_path TO disable_node_with_replicated_tables;

 -- should be able to ingest data from both the worker and the coordinator