Initial implementation of "failover_validation_command"

configfile.c

@@ -362,6 +362,7 @@ _parse_config(t_configuration_options *options, ItemList *error_list, ItemList *
 	options->sibling_nodes_disconnect_timeout = DEFAULT_SIBLING_NODES_DISCONNECT_TIMEOUT;
 	options->connection_check_type = CHECK_PING;
 	options->primary_visibility_consensus = false;
+	memset(options->failover_validation_command, 0, sizeof(options->failover_validation_command));
 
 	/*-------------
 	 * witness settings
@@ -644,6 +645,8 @@ _parse_config(t_configuration_options *options, ItemList *error_list, ItemList *
 		}
 		else if (strcmp(name, "primary_visibility_consensus") == 0)
 			options->primary_visibility_consensus = parse_bool(value, name, error_list);
+		else if (strcmp(name, "failover_validation_command") == 0)
+			strncpy(options->failover_validation_command, value, sizeof(options->failover_validation_command));
 
 		/* witness settings */
 		else if (strcmp(name, "witness_sync_interval") == 0)

configfile.h

@@ -145,6 +145,7 @@ typedef struct
 	int			sibling_nodes_disconnect_timeout;
 	ConnectionCheckType connection_check_type;
 	bool		primary_visibility_consensus;
+	char		failover_validation_command[MAXPGPATH];
 
 	/* BDR settings */
 	bool		bdr_local_monitoring_only;
@@ -189,7 +190,7 @@ typedef struct
 		/* node information */ \
 		UNKNOWN_NODE_ID, "", "", "", "", "", "", "", REPLICATION_TYPE_PHYSICAL,	\
 		/* log settings */ \
-		"", "", "", DEFAULT_LOG_STATUS_INTERVAL,	\
+		"", "", "", DEFAULT_LOG_STATUS_INTERVAL, \
 		/* standby clone settings */ \
 		false, "", "", { NULL, NULL }, "", false, "", false, "", \
 		/* standby promote settings */ \
@@ -216,7 +217,7 @@ typedef struct
         false, -1, \
 		DEFAULT_ASYNC_QUERY_TIMEOUT, \
 		DEFAULT_PRIMARY_NOTIFICATION_TIMEOUT, \
-		-1, "", false, DEFAULT_SIBLING_NODES_DISCONNECT_TIMEOUT, CHECK_PING, true, \
+		-1, "", false, DEFAULT_SIBLING_NODES_DISCONNECT_TIMEOUT, CHECK_PING, true, "", \
 		/* BDR settings */ \
 		false, DEFAULT_BDR_RECOVERY_TIMEOUT, \
 		/* service settings */ \

repmgrd-physical.c

@@ -23,7 +23,6 @@
 #include "repmgrd.h"
 #include "repmgrd-physical.h"
 
-
 typedef enum
 {
 	FAILOVER_STATE_UNKNOWN = -1,
@@ -88,7 +87,8 @@ static void update_monitoring_history(void);
 static void handle_sighup(PGconn **conn, t_server_type server_type);
 
 static const char *format_failover_state(FailoverState failover_state);
-
+static const char * format_failover_state(FailoverState failover_state);
+static void parse_failover_validation_command(const char *template,  t_node_info *node_info, PQExpBufferData *out);
 
 void
 handle_sigint_physical(SIGNAL_ARGS)
@@ -3571,7 +3571,58 @@ do_election(void)
 			   candidate_node->node_id);
 
 	if (candidate_node->node_id == local_node_info.node_id)
+	{
+		/*
+		 * If "failover_validation_command" is set, execute that command.
+		 */
+
+		if (config_file_options.failover_validation_command[0] != '\0')
+		{
+			PQExpBufferData failover_validation_command;
+			PQExpBufferData command_output;
+			int return_value = -1;
+
+			initPQExpBuffer(&failover_validation_command);
+			initPQExpBuffer(&command_output);
+
+			parse_failover_validation_command(config_file_options.failover_validation_command,
+											  candidate_node,
+											  &failover_validation_command);
+
+			log_notice(_("executing \"failover_validation_command\""));
+			log_detail("%s", failover_validation_command.data);
+
+			/* we determine success of the command by the value placed into return_value */
+			(void) local_command_return_value(failover_validation_command.data,
+											  &command_output,
+											  &return_value);
+
+			termPQExpBuffer(&failover_validation_command);
+
+			if (return_value != 0)
+				log_warning(_("failover validation command returned %i"), return_value);
+
+			if (command_output.data[0] != '\0')
+			{
+				log_info("output returned by failover validation command:\n%s", command_output.data);
+			}
+			else
+			{
+				log_info(_("no output returned from command"));
+			}
+
+			termPQExpBuffer(&command_output);
+
+			if (return_value != 0)
+			{
+				/* create event here? */
+				log_notice(_("failover validation command returned a non-zero value (%i)"), return_value);
+				return ELECTION_LOST;
+			}
+		}
+
 		return ELECTION_WON;
+	}
 
 	return ELECTION_LOST;
 }
@@ -3782,3 +3833,47 @@ handle_sighup(PGconn **conn, t_server_type server_type)
 
 	got_SIGHUP = false;
 }
+
+
+static void
+parse_failover_validation_command(const char *template,  t_node_info *node_info, PQExpBufferData *out)
+{
+	const char *src_ptr;
+
+	for (src_ptr = template; *src_ptr; src_ptr++)
+	{
+		if (*src_ptr == '%')
+		{
+			switch (src_ptr[1])
+			{
+				case '%':
+					/* %%: replace with % */
+					src_ptr++;
+					appendPQExpBufferChar(out, *src_ptr);
+					break;
+				case 'n':
+					/* %n: node id */
+					src_ptr++;
+					appendPQExpBuffer(out, "%i", node_info->node_id);
+					break;
+				case 'a':
+					/* %a: node name */
+					src_ptr++;
+					appendPQExpBufferStr(out, node_info->node_name);
+					break;
+
+				default:
+					/* otherwise treat the % as not special */
+					appendPQExpBufferChar(out, *src_ptr);
+
+					break;
+			}
+		}
+		else
+		{
+			appendPQExpBufferChar(out, *src_ptr);
+		}
+	}
+
+	return;
+}