From 984ce7420b2ed41ee0033579dc3af2f50ae4b9d1 Mon Sep 17 00:00:00 2001
From: Ian Barwick <ian@2ndquadrant.com>
Date: Wed, 6 Feb 2019 13:31:02 +0900
Subject: [PATCH] "daemon status": emit warning if WAL replay is paused

Specifically, if WAL replay is paused *and* WAL is pending replay,
this node cannot be promoted until WAL replay is unpaused. In this
state it is not a suitable promotion candidate in a failover situation.
---
 dbutils.c              |  2 +-
 dbutils.h              |  2 ++
 repmgr-action-daemon.c | 21 ++++++++++++++++++++-
 3 files changed, 23 insertions(+), 2 deletions(-)

diff --git a/dbutils.c b/dbutils.c
index e2da25de..b6b9a24f 100644
--- a/dbutils.c
+++ b/dbutils.c
@@ -5046,7 +5046,7 @@ is_wal_replay_paused(PGconn *conn, bool check_pending_wal)
 
 	if (PQresultStatus(res) != PGRES_TUPLES_OK)
 	{
-		log_db_error(conn, query.data, _("unable to execute \"%s\""), query.data);
+		log_db_error(conn, query.data, _("unable to execute WAL replay pause query"));
 	}
 	else
 	{
diff --git a/dbutils.h b/dbutils.h
index 43e193a9..d86432c6 100644
--- a/dbutils.h
+++ b/dbutils.h
@@ -347,9 +347,11 @@ typedef struct RepmgrdInfo {
 	char pid_file[MAXLEN];
 	bool pg_running;
 	char pg_running_text[MAXLEN];
+	RecoveryType recovery_type;
 	bool running;
 	char repmgrd_running[MAXLEN];
 	bool paused;
+	bool wal_paused_pending_wal;
 } RepmgrdInfo;
 
 
diff --git a/repmgr-action-daemon.c b/repmgr-action-daemon.c
index b97ec8b9..0c724fc0 100644
--- a/repmgr-action-daemon.c
+++ b/repmgr-action-daemon.c
@@ -66,6 +66,7 @@ do_daemon_status(void)
 	int i;
 	RepmgrdInfo **repmgrd_info;
 	ItemList	warnings = {NULL, NULL};
+	bool		connection_error_found = false;
 
 	/* Connect to local database to obtain cluster connection data */
 	log_verbose(LOG_INFO, _("connecting to database"));
@@ -108,14 +109,18 @@ do_daemon_status(void)
 		repmgrd_info[i] = pg_malloc0(sizeof(RepmgrdInfo));
 		repmgrd_info[i]->node_id = cell->node_info->node_id;
 		repmgrd_info[i]->pid = UNKNOWN_PID;
+		repmgrd_info[i]->recovery_type = RECTYPE_UNKNOWN;
 		repmgrd_info[i]->paused = false;
 		repmgrd_info[i]->running = false;
 		repmgrd_info[i]->pg_running = true;
+		repmgrd_info[i]->wal_paused_pending_wal = false;
 
 		cell->node_info->conn = establish_db_connection_quiet(cell->node_info->conninfo);
 
 		if (PQstatus(cell->node_info->conn) != CONNECTION_OK)
 		{
+			connection_error_found = true;
+
 			if (runtime_options.verbose)
 			{
 				char		error[MAXLEN];
@@ -166,6 +171,20 @@ do_daemon_status(void)
 
 			repmgrd_info[i]->paused = repmgrd_is_paused(cell->node_info->conn);
 
+			repmgrd_info[i]->recovery_type = get_recovery_type(cell->node_info->conn);
+
+			if (repmgrd_info[i]->recovery_type == RECTYPE_STANDBY)
+			{
+				repmgrd_info[i]->wal_paused_pending_wal = is_wal_replay_paused(cell->node_info->conn, true);
+
+				if (repmgrd_info[i]->wal_paused_pending_wal == true)
+				{
+					item_list_append_format(&warnings,
+											"WAL replay is paused on node \"%s\" (ID: %i) with WAL replay pending; this node cannot be promoted",
+											cell->node_info->node_name, cell->node_info->node_id);
+				}
+			}
+
 			PQfinish(cell->node_info->conn);
 		}
 
@@ -244,7 +263,7 @@ do_daemon_status(void)
 			printf(_("  - %s\n"), cell->string);
 		}
 
-		if (runtime_options.verbose == false)
+		if (runtime_options.verbose == false && connection_error_found == true)
 		{
 			log_hint(_("execute with --verbose option to see connection error messages"));
 		}