Store node status in node record struct

dbutils.h

@@ -46,6 +46,12 @@ typedef enum {
 	MS_DEGRADED = 1
 } MonitoringState;
 
+typedef enum {
+	NODE_STATUS_UNKNOWN = -1,
+	NODE_STATUS_UP,
+	NODE_STATUS_DOWN
+} NodeStatus;
+
 /*
  * Struct to store node information
  */
@@ -65,27 +71,29 @@ typedef struct s_node_info
 	bool		  is_ready;
 	bool		  is_visible;
 	XLogRecPtr	  last_wal_receive_lsn;
+	NodeStatus	  node_status;
 	MonitoringState monitoring_state;
 	PGconn		 *conn;
 }	t_node_info;
 
 
 #define T_NODE_INFO_INITIALIZER { \
-  NODE_NOT_FOUND, \
-  NO_UPSTREAM_NODE, \
-  UNKNOWN, \
-  "", \
-  "", \
-  "", \
-  DEFAULT_LOCATION, \
-  DEFAULT_PRIORITY, \
-  true, \
-  "", \
-  false, \
-  false, \
-  InvalidXLogRecPtr, \
-  MS_NORMAL, \
-  NULL \
+	NODE_NOT_FOUND, \
+	NO_UPSTREAM_NODE, \
+	UNKNOWN, \
+	"", \
+	"", \
+	"", \
+	DEFAULT_LOCATION, \
+	DEFAULT_PRIORITY, \
+	true, \
+	"", \
+	false, \
+	false, \
+	InvalidXLogRecPtr, \
+	NODE_STATUS_UNKNOWN, \
+	MS_NORMAL, \
+	NULL \
 }
 
 

repmgrd-bdr.c

@@ -168,6 +168,11 @@ monitor_bdr(void)
 				{
 					if (is_server_available(cell->node_info->conninfo) == false)
 					{
+						instr_time	upstream_node_unreachable_start;
+
+						INSTR_TIME_SET_CURRENT(upstream_node_unreachable_start);
+
+
 						// XXX improve
 						log_warning("connection problem! to node %i", cell->node_info->node_id);
 						do_bdr_failover(&nodes, cell->node_info);
@@ -235,7 +240,6 @@ do_bdr_failover(NodeInfoList *nodes, t_node_info *monitored_node)
 	PGconn	   *next_node_conn = NULL;
 	NodeInfoListCell *cell;
 	PQExpBufferData event_details;
-	RecordStatus  record_status;
 	t_event_info event_info = T_EVENT_INFO_INITIALIZER;
 	t_node_info target_node  = T_NODE_INFO_INITIALIZER;
 
@@ -258,10 +262,14 @@ do_bdr_failover(NodeInfoList *nodes, t_node_info *monitored_node)
 
 		if (PQstatus(next_node_conn) == CONNECTION_OK)
 		{
-			// XXX check if record returned
-			record_status = get_node_record(next_node_conn, cell->node_info->node_id, &target_node);
+			RecordStatus record_status = get_node_record(next_node_conn,
+														 cell->node_info->node_id,
+														 &target_node);
 
-			break;
+			if (record_status == RECORD_FOUND)
+			{
+				break;
+			}
 		}
 
 		next_node_conn = NULL;
@@ -286,6 +294,7 @@ do_bdr_failover(NodeInfoList *nodes, t_node_info *monitored_node)
 	log_debug("this node is the failover handler");
 
 	// check here that the node hasn't come back up...
+
 	log_info(_("connecting to target node %s"), target_node.node_name);
 
 	initPQExpBuffer(&event_details);

repmgrd-physical.c

@@ -135,15 +135,18 @@ do_physical_node_check(void)
 
 
 
-
+/*
+ * repmgrd running on the primary server
+ */
 void
 monitor_streaming_primary(void)
 {
 #ifndef BDR_ONLY
-	NodeStatus	node_status = NODE_STATUS_UP;
 	instr_time	log_status_interval_start;
 	PQExpBufferData event_details;
 
+	local_node_info.node_status = NODE_STATUS_UP;
+
 	reset_node_voting_status();
 
 	/* Log startup event */
@@ -181,7 +184,7 @@ monitor_streaming_primary(void)
 		{
 
 			/* node is down, we were expecting it to be up */
-			if (node_status == NODE_STATUS_UP)
+			if (local_node_info.node_status == NODE_STATUS_UP)
 			{
 				PQExpBufferData event_details;
 				instr_time	local_node_unreachable_start;
@@ -195,7 +198,7 @@ monitor_streaming_primary(void)
 
 				log_warning("%s", event_details.data);
 
-				node_status = NODE_STATUS_UNKNOWN;
+				local_node_info.node_status = NODE_STATUS_UNKNOWN;
 
 				PQfinish(local_conn);
 
@@ -214,9 +217,9 @@ monitor_streaming_primary(void)
 
 				termPQExpBuffer(&event_details);
 
-				local_conn = try_reconnect(local_node_info.conninfo, &node_status);
+				local_conn = try_reconnect(&local_node_info);
 
-				if (node_status == NODE_STATUS_UP)
+				if (local_node_info.node_status == NODE_STATUS_UP)
 				{
 					int		local_node_unreachable_elapsed = calculate_elapsed(local_node_unreachable_start);
 
@@ -279,7 +282,7 @@ monitor_streaming_primary(void)
 
 				if (PQstatus(local_conn) == CONNECTION_OK)
 				{
-					node_status = NODE_STATUS_UP;
+					local_node_info.node_status = NODE_STATUS_UP;
 					monitoring_state = MS_NORMAL;
 
 					initPQExpBuffer(&event_details);
@@ -343,10 +346,10 @@ monitor_streaming_standby(void)
 {
 #ifndef BDR_ONLY
 	RecordStatus record_status;
-	NodeStatus	upstream_node_status = NODE_STATUS_UP;
 	instr_time	log_status_interval_start;
 	PQExpBufferData event_details;
 
+	upstream_node_info.node_status = NODE_STATUS_UP;
 	reset_node_voting_status();
 
 	log_debug("monitor_streaming_standby()");
@@ -470,7 +473,7 @@ monitor_streaming_standby(void)
 		{
 
 			/* upstream node is down, we were expecting it to be up */
-			if (upstream_node_status == NODE_STATUS_UP)
+			if (upstream_node_info.node_status == NODE_STATUS_UP)
 			{
 				instr_time	upstream_node_unreachable_start;
 
@@ -478,7 +481,7 @@ monitor_streaming_standby(void)
 
 				initPQExpBuffer(&event_details);
 
-				upstream_node_status = NODE_STATUS_UNKNOWN;
+				upstream_node_info.node_status = NODE_STATUS_UNKNOWN;
 
 				appendPQExpBuffer(&event_details,
 								  _("unable to connect to upstream node \"%s\" (node ID: %i)"),
@@ -499,9 +502,9 @@ monitor_streaming_standby(void)
 				termPQExpBuffer(&event_details);
 
 				PQfinish(upstream_conn);
-				upstream_conn = try_reconnect(upstream_node_info.conninfo, &upstream_node_status);
+				upstream_conn = try_reconnect(&upstream_node_info);
 
-				if (upstream_node_status == NODE_STATUS_UP)
+				if (upstream_node_info.node_status == NODE_STATUS_UP)
 				{
 					int		upstream_node_unreachable_elapsed = calculate_elapsed(upstream_node_unreachable_start);
 
@@ -524,7 +527,7 @@ monitor_streaming_standby(void)
 				}
 
 				/* still down after reconnect attempt(s) */
-				if (upstream_node_status == NODE_STATUS_DOWN)
+				if (upstream_node_info.node_status == NODE_STATUS_DOWN)
 				{
 					bool failover_done = false;
 
@@ -564,7 +567,7 @@ monitor_streaming_standby(void)
 					// and upstream is now former primary
 					// XXX scan other nodes to see if any has become primary
 
-					upstream_node_status = NODE_STATUS_UP;
+					upstream_node_info.node_status = NODE_STATUS_UP;
 					monitoring_state = MS_NORMAL;
 
 					if (upstream_node_info.type == PRIMARY)

repmgrd.c

@@ -63,9 +63,6 @@ static void handle_sighup(SIGNAL_ARGS);
 static void handle_sigint(SIGNAL_ARGS);
 #endif
 
-
-PGconn *try_reconnect(const char *conninfo, NodeStatus *node_status);
-
 int calculate_elapsed(instr_time start_time);
 void update_registration(PGconn *conn);
 void terminate(int retval);
@@ -612,7 +609,7 @@ show_help(void)
 
 
 PGconn *
-try_reconnect(const char *conninfo, NodeStatus *node_status)
+try_reconnect(t_node_info *node_info)
 {
 	PGconn *conn;
 
@@ -623,7 +620,7 @@ try_reconnect(const char *conninfo, NodeStatus *node_status)
 	for (i = 0; i < max_attempts; i++)
 	{
 		log_info(_("checking state of node, %i of %i attempts"), i, max_attempts);
-		if (is_server_available(conninfo) == true)
+		if (is_server_available(node_info->conninfo) == true)
 		{
 			log_notice(_("node has recovered, reconnecting"));
 
@@ -633,10 +630,10 @@ try_reconnect(const char *conninfo, NodeStatus *node_status)
 			 *  - fall back to degraded monitoring?
 			 *  - make that configurable
 			 */
-			conn = establish_db_connection(conninfo, false);
+			conn = establish_db_connection(node_info->conninfo, false);
 			if (PQstatus(conn) == CONNECTION_OK)
 			{
-				*node_status = NODE_STATUS_UP;
+				node_info->node_status = NODE_STATUS_UP;
 				return conn;
 			}
 
@@ -650,7 +647,8 @@ try_reconnect(const char *conninfo, NodeStatus *node_status)
 
 
 	log_warning(_("unable to reconnect to node after %i attempts"), max_attempts);
-	*node_status = NODE_STATUS_DOWN;
+	node_info->node_status = NODE_STATUS_DOWN;
+
 	return NULL;
 }
 

repmgrd.h

@@ -10,12 +10,6 @@
 #include <time.h>
 #include "portability/instr_time.h"
 
-typedef enum {
-	NODE_STATUS_UNKNOWN = -1,
-	NODE_STATUS_UP,
-	NODE_STATUS_DOWN
-} NodeStatus;
-
 
 extern MonitoringState monitoring_state;
 extern instr_time	degraded_monitoring_start;
@@ -25,7 +19,7 @@ extern t_node_info local_node_info;
 extern PGconn	   *local_conn;
 extern bool			startup_event_logged;
 
-PGconn *try_reconnect(const char *conninfo, NodeStatus *node_status);
+PGconn *try_reconnect(t_node_info *node_info);
 
 int calculate_elapsed(instr_time start_time);
 const char *print_monitoring_state(MonitoringState monitoring_state);