From cd6a55c7cb9b81ea63eae18925f03a7f18c46ebc Mon Sep 17 00:00:00 2001
From: Ian Barwick <barwick@gmail.com>
Date: Thu, 4 Apr 2019 16:03:55 +0900
Subject: [PATCH] repmgrd: improve primary visibility consensus check

Exclude sibling nodes which report they're following a different
node. This shouldn't happen, but could.
---
 dbutils.c          | 15 ++++++++++++---
 dbutils.h          |  1 +
 repmgr.c           |  4 ----
 repmgrd-physical.c | 28 +++++++++++++++++++---------
 4 files changed, 32 insertions(+), 16 deletions(-)

diff --git a/dbutils.c b/dbutils.c
index 21601ecd..c26bce5c 100644
--- a/dbutils.c
+++ b/dbutils.c
@@ -4993,7 +4993,8 @@ get_replication_info(PGconn *conn, t_server_type node_type, ReplInfo *replicatio
 						 "        END AS replication_lag_time, "
 						 "        last_wal_receive_lsn >= last_wal_replay_lsn AS receiving_streamed_wal, "
 						 "        wal_replay_paused, "
-						 "        upstream_last_seen "
+						 "        upstream_last_seen, "
+						 "        upstream_node_id "
 						 "   FROM ( "
 						 " SELECT CURRENT_TIMESTAMP AS ts, "
 						 "        pg_catalog.pg_is_in_recovery() AS in_recovery, "
@@ -5033,10 +5034,12 @@ get_replication_info(PGconn *conn, t_server_type node_type, ReplInfo *replicatio
 							 "        END AS wal_replay_paused, ");
 	}
 
+	/* Add information about upstream node from shared memory */
 	if (node_type == WITNESS)
 	{
 		appendPQExpBufferStr(&query,
-							 "        repmgr.get_upstream_last_seen() AS upstream_last_seen");
+							 "        repmgr.get_upstream_last_seen() AS upstream_last_seen, "
+							 "        repmgr.get_upstream_node_id() AS upstream_node_id ");
 	}
 	else
 	{
@@ -5044,7 +5047,12 @@ get_replication_info(PGconn *conn, t_server_type node_type, ReplInfo *replicatio
 							 "        CASE WHEN pg_catalog.pg_is_in_recovery() IS FALSE "
 							 "          THEN -1 "
 							 "          ELSE repmgr.get_upstream_last_seen() "
-							 "        END AS upstream_last_seen ");
+							 "        END AS upstream_last_seen, ");
+		appendPQExpBufferStr(&query,
+							 "        CASE WHEN pg_catalog.pg_is_in_recovery() IS FALSE "
+							 "          THEN -1 "
+							 "          ELSE repmgr.get_upstream_node_id() "
+							 "        END AS upstream_node_id ");
 	}
 
 	appendPQExpBufferStr(&query,
@@ -5075,6 +5083,7 @@ get_replication_info(PGconn *conn, t_server_type node_type, ReplInfo *replicatio
 		replication_info->receiving_streamed_wal = atobool(PQgetvalue(res, 0, 6));
 		replication_info->wal_replay_paused = atobool(PQgetvalue(res, 0, 7));
 		replication_info->upstream_last_seen = atoi(PQgetvalue(res, 0, 8));
+		replication_info->upstream_node_id = atoi(PQgetvalue(res, 0, 9));
 	}
 
 	termPQExpBuffer(&query);
diff --git a/dbutils.h b/dbutils.h
index 29bdf4b4..bddb8c21 100644
--- a/dbutils.h
+++ b/dbutils.h
@@ -310,6 +310,7 @@ typedef struct
 	bool		receiving_streamed_wal;
 	bool		wal_replay_paused;
 	int			upstream_last_seen;
+	int			upstream_node_id;
 } ReplInfo;
 
 typedef struct
diff --git a/repmgr.c b/repmgr.c
index 666b2719..ccb3b21a 100644
--- a/repmgr.c
+++ b/repmgr.c
@@ -436,10 +436,6 @@ get_upstream_node_id(PG_FUNCTION_ARGS)
 	if (!shared_state)
 		PG_RETURN_NULL();
 
-	/* A primary node cannot have an upstream ID */
-	if (!RecoveryInProgress())
-		PG_RETURN_INT32(UNKNOWN_NODE_ID);
-
 	LWLockAcquire(shared_state->lock, LW_SHARED);
 	upstream_node_id = shared_state->upstream_node_id;
 	LWLockRelease(shared_state->lock);
diff --git a/repmgrd-physical.c b/repmgrd-physical.c
index 31fbed6c..d2b6b04a 100644
--- a/repmgrd-physical.c
+++ b/repmgrd-physical.c
@@ -3633,15 +3633,25 @@ do_election(NodeInfoList *sibling_nodes, int *new_primary_id)
 
 		if (sibling_replication_info.upstream_last_seen >= 0 && sibling_replication_info.upstream_last_seen < (config_file_options.monitor_interval_secs * 2))
 		{
-			nodes_with_primary_still_visible++;
-			log_notice(_("node %i last saw primary node %i second(s) ago, considering primary still visible"),
-					   cell->node_info->node_id,
-					   sibling_replication_info.upstream_last_seen);
-			appendPQExpBuffer(&nodes_with_primary_visible,
-							  " - node \"%s\" (ID: %i): %i second(s) ago\n",
-							  cell->node_info->node_name,
-							  cell->node_info->node_id,
-							  sibling_replication_info.upstream_last_seen);
+			if (sibling_replication_info.upstream_node_id != upstream_node_info.node_id)
+			{
+				log_warning(_("assumed sibling node %i monitoring different upstream node %i"),
+							cell->node_info->node_id,
+							sibling_replication_info.upstream_node_id);
+
+			}
+			else
+			{
+				nodes_with_primary_still_visible++;
+				log_notice(_("node %i last saw primary node %i second(s) ago, considering primary still visible"),
+						   cell->node_info->node_id,
+						   sibling_replication_info.upstream_last_seen);
+				appendPQExpBuffer(&nodes_with_primary_visible,
+								  " - node \"%s\" (ID: %i): %i second(s) ago\n",
+								  cell->node_info->node_name,
+								  cell->node_info->node_id,
+								  sibling_replication_info.upstream_last_seen);
+			}
 		}
 		else
 		{