Improves pgxc_ctl behavior to

1) When issuing pg_ctl start.   Added -w option for better synchronization.

2) At pinging coordinator/datanode, when ping does not detect the node,
   retries three times with 100 millisecond interval just in case.

Author: koichi-szk

contrib/pgxc_ctl/coord_cmd.c

@@ -275,12 +275,12 @@ cmd_t *prepare_initCoordinatorSlave(char *nodeName)
 	 * a master which can handle the request.   So GTM should be running. We can test all of them by
 	 * single 'select 1' command.
 	 */
-	if (pingNode(aval(VAR_coordMasterServers)[idx], aval(VAR_coordPorts)[idx]) != 0)
+	if (pingNode(aval(VAR_coordMasterServers)[idx], aval(VAR_coordPorts)[idx]) != -2)
 	{
 		/* Master is not running. Must start it first */
 		appendCmdEl(cmdBuildDir, (cmdStartMaster = initCmd(aval(VAR_coordMasterServers)[idx])));
 		snprintf(newCommand(cmdStartMaster), MAXLINE,
-				 "pg_ctl start -Z coordinator -D %s -o -i",
+				 "pg_ctl start -Z coordinator -D %s -o -i -w",
 				 aval(VAR_coordMasterDirs)[idx]);
 	}
 	/*
@@ -977,7 +977,7 @@ int add_coordinatorMaster(char *name, char *host, int port, int pooler, char *di
 				   aval(VAR_coordPorts)[0], aval(VAR_coordMasterServers)[0], pgdumpall_out);
 
 	/* Start the new coordinator */
-	doImmediate(host, NULL, "pg_ctl start -Z restoremode -D %s -o -i", dir);
+	doImmediate(host, NULL, "pg_ctl start -Z restoremode -D %s -o -i -w", dir);
 
 	/* Restore the backup */
 	doImmediateRaw("psql -h %s -p %d -d %s -f %s", host, port, sval(VAR_defaultDatabase), pgdumpall_out);
@@ -1151,7 +1151,7 @@ int add_coordinatorSlave(char *name, char *host, char *dir, char *archDir)
 	doImmediate(aval(VAR_coordMasterServers)[idx], NULL, 
 				"pg_ctl stop -Z coordinator -D %s -m fast", aval(VAR_coordMasterDirs)[idx]);
 	doImmediate(aval(VAR_coordMasterServers)[idx], NULL, 
-				"pg_ctl start -Z coordinator -D %s", aval(VAR_coordMasterDirs)[idx]);
+				"pg_ctl start -Z coordinator -D %s -w", aval(VAR_coordMasterDirs)[idx]);
 #if 0
 	/* pg_basebackup */
 	doImmediate(host, NULL, "pg_basebackup -p %s -h %s -D %s -x",
@@ -1239,7 +1239,7 @@ int add_coordinatorSlave(char *name, char *host, char *dir, char *archDir)
 	fclose(f);
 
 	/* Start the slave */
-	doImmediate(host, NULL, "pg_ctl start -Z coordinator -D %s", dir);
+	doImmediate(host, NULL, "pg_ctl start -Z coordinator -D %s -w", dir);
 	return 0;
 }
 
@@ -1502,7 +1502,7 @@ cmd_t *prepare_startCoordinatorMaster(char *nodeName)
 	}
 	cmd = cmdPgCtl = initCmd(aval(VAR_coordMasterServers)[idx]);
 	snprintf(newCommand(cmdPgCtl), MAXLINE,
-			 "pg_ctl start -Z coordinator -D %s -o -i",
+			 "pg_ctl start -Z coordinator -D %s -o -i -w",
 			 aval(VAR_coordMasterDirs)[idx]);
 	return(cmd);
 }
@@ -1562,7 +1562,7 @@ cmd_t *prepare_startCoordinatorSlave(char *nodeName)
 	}
 	cmd = cmdPgCtlStart = initCmd(aval(VAR_coordSlaveServers)[idx]);
 	snprintf(newCommand(cmdPgCtlStart), MAXLINE,
-			 "pg_ctl start -Z coordinator -D %s -o -i",
+			 "pg_ctl start -Z coordinator -D %s -o -i -w",
 			 aval(VAR_coordSlaveDirs)[idx]);
 
 	/* Postgresql.conf at the Master */

contrib/pgxc_ctl/datanode_cmd.c

@@ -370,7 +370,7 @@ cmd_t *prepare_startDatanodeMaster(char *nodeName)
 	}
 	cmdStartDatanodeMaster = initCmd(aval(VAR_datanodeMasterServers)[idx]);
 	snprintf(newCommand(cmdStartDatanodeMaster), MAXLINE,
-			 "pg_ctl start -Z datanode -D %s -o -i", aval(VAR_datanodeMasterDirs)[idx]);
+			 "pg_ctl start -Z datanode -D %s -o -i -w", aval(VAR_datanodeMasterDirs)[idx]);
 	return(cmdStartDatanodeMaster);
 }
 
@@ -429,7 +429,7 @@ cmd_t *prepare_startDatanodeSlave(char *nodeName)
 
 	cmd = cmdStartDatanodeSlave = initCmd(aval(VAR_datanodeSlaveServers)[idx]);
 	snprintf(newCommand(cmdStartDatanodeSlave), MAXLINE,
-			 "pg_ctl start -Z datanode -D %s",
+			 "pg_ctl start -Z datanode -D %s -w",
 			 aval(VAR_datanodeSlaveDirs)[idx]);
 
 	/* Change the master to synchronous mode */
@@ -986,7 +986,7 @@ int add_datanodeMaster(char *name, char *host, int port, char *dir)
 				   aval(VAR_datanodePorts)[0], aval(VAR_datanodeMasterServers)[0], pgdumpall_out);
 
 	/* Start the new datanode */
-	doImmediate(host, NULL, "pg_ctl start -Z restoremode -D %s -o -i", dir);
+	doImmediate(host, NULL, "pg_ctl start -Z restoremode -D %s -o -i -w", dir);
 
 	/* Restore the backup */
 	doImmediateRaw("psql -h %s -p %d -d %s -f %s", host, port, sval(VAR_defaultDatabase), pgdumpall_out);
@@ -1156,7 +1156,7 @@ int add_datanodeSlave(char *name, char *host, char *dir, char *archDir)
 	doImmediate(aval(VAR_coordMasterServers)[idx], NULL, 
 				"pg_ctl stop -Z datanode -D %s -m fast", aval(VAR_datanodeMasterDirs)[idx]);
 	doImmediate(aval(VAR_coordMasterServers)[idx], NULL, 
-				"pg_ctl start -Z datanode -D %s", aval(VAR_datanodeMasterDirs)[idx]);
+				"pg_ctl start -Z datanode -D %s -w", aval(VAR_datanodeMasterDirs)[idx]);
 	/* pg_basebackup */
 	doImmediate(host, NULL, "pg_basebackup -p %s -h %s -D %s -x",
 				aval(VAR_datanodePorts)[idx], aval(VAR_datanodeMasterServers)[idx], dir);
@@ -1198,7 +1198,7 @@ int add_datanodeSlave(char *name, char *host, char *dir, char *archDir)
 			aval(VAR_datanodeArchLogDirs)[idx], aval(VAR_datanodeArchLogDirs)[idx]);
 	fclose(f);
 	/* Start the slave */
-	doImmediate(host, NULL, "pg_ctl start -Z datanode -D %s", dir);
+	doImmediate(host, NULL, "pg_ctl start -Z datanode -D %s -w", dir);
 	return 0;
 }
 

contrib/pgxc_ctl/utils.c

@@ -28,10 +28,12 @@
 #include "config.h"
 #include "variables.h"
 #include "varnames.h"
+#include "sys/time.h"
 
 static int Malloc_ed = 0;
 static int Strdup_ed = 0;
 static int Freed = 0;
+static void myUsleep(long microsec);
 
 void *Malloc(size_t size)
 {
@@ -308,6 +310,9 @@ int pingNode(char *host, char *port)
 	PGPing status;
 	char conninfo[MAXLINE+1];
 	char editBuf[MAXPATH+1];
+#define RETRY 3
+#define sleepMicro 100*1000	/* 100 millisec */
+	int retry;
 
 	conninfo[0] = 0;
 	if (host)
@@ -322,14 +327,22 @@ int pingNode(char *host, char *port)
 	}
 	if (conninfo[0])
 	{
-		status = PQping(conninfo);
-		if (status == PQPING_OK)
-			return 0;
-		else
-			return 1;
+		for (retry = RETRY; retry; retry--){
+			status = PQping(conninfo);
+			if (status == PQPING_OK)
+				return 0;
+			else
+			{
+				myUsleep(sleepMicro);
+				continue;
+			}
+		}
+		return 1;
 	}
 	else
 		return -1;
+#undef RETRY
+#undef sleepMicro
 }
 
 void trimNl(char *s)
@@ -378,4 +391,15 @@ char *getIpAddress(char *hostName)
 	trimNl(ipAddr);
 	return ipAddr;
 }
-	
+
+static void myUsleep(long microsec)
+{
+	struct timeval delay;
+
+	if (microsec <= 0)
+		return;
+
+	delay.tv_sec = microsec / 1000000L;
+	delay.tv_usec = microsec % 1000000L;
+	(void) select(0, NULL, NULL, NULL, &delay);
+}