MCOL-1137 - fixed master/slave setup after failvover

oam/install_scripts/disable-rep-columnstore.sh

@@ -56,7 +56,7 @@ checkForError
 #
 # Run reset slave command
 #
-echo "Run start slave command" >>/tmp/disable-rep-status.log
+echo "Run reset slave command" >>/tmp/disable-rep-status.log
 cat >/tmp/idb_disable-rep.sql <<EOD
 reset slave;
 EOD

procmgr/main.cpp

@@ -1645,7 +1645,7 @@ void pingDeviceThread()
 											DeviceNetworkConfig devicenetworkconfig;
 											devicenetworkconfig.DeviceName = moduleName;
 											devicenetworklist.push_back(devicenetworkconfig);
-											processManager.setMySQLReplication(devicenetworklist);
+											processManager.setMySQLReplication(devicenetworklist, oam::UnassignedName, false, true);
 										}
 									}
 									else
@@ -1833,6 +1833,9 @@ void pingDeviceThread()
 								if (LANOUTAGEACTIVE)
 									break;
 	
+								// if not disabled and amazon, skip
+								if (opState != oam::AUTO_DISABLED )
+								{	
 									//Log failure, issue alarm, set moduleOpState 
 									Configuration config;
 									log.writeLog(__LINE__, "module is down: " + moduleName, LOG_TYPE_CRITICAL);
@@ -1893,11 +1896,10 @@ void pingDeviceThread()
 											log.writeLog(__LINE__, "EXCEPTION ERROR on autoMovePmDbroot: Caught unknown exception!", LOG_TYPE_ERROR);
 										}
 									}
+								}
 								
 								// if Cloud Instance
-								// state = running, then instance is rebooting, monitor for recovery
+								// state = terminate, remove/addmodule to launch new instance
-								// state = stopped, then try starting, if fail, remove/addmodule to launch new instance
-								// state = terminate or nothing, remove/addmodule to launch new instance
 								if ( amazon ) {
 									if ( moduleName.find("um") == 0 )
 									{
@@ -2104,7 +2106,9 @@ void pingDeviceThread()
 										}
 									}
 
-									if ( moduleName.find("pm") == 0 )
+									if ( ( moduleName.find("pm") == 0 ) &&
+									    ( opState != oam::AUTO_DISABLED ) )
+
 									{
 										// resume the dbrm
 										oam.dbrmctl("resume");
@@ -2166,6 +2170,10 @@ void pingDeviceThread()
 									}
 								}
 
+								// if disabled and amazon, break out
+								if ( (opState == oam::AUTO_DISABLED ) && amazon )
+								      break;
+	
 								//start SIMPLEX runtype processes on a SIMPLEX runtype module
 								string moduletype = moduleName.substr(0,MAX_MODULE_TYPE_SIZE);
 		

procmgr/processmanager.cpp

@@ -9715,8 +9715,11 @@ std::string ProcessManager::getStandbyModule()
 
 			if ( systemprocessstatus.processstatus[i].ProcessName == "ProcessManager" &&
 				systemprocessstatus.processstatus[i].ProcessOpState == oam::COLD_STANDBY )
+			{
 				// Found a ProcessManager in a COLD_STANDBY state
 				newStandbyModule = systemprocessstatus.processstatus[i].Module;
+				break;
+			}
 
 			if ( systemprocessstatus.processstatus[i].ProcessName == "ProcessManager" &&
 				systemprocessstatus.processstatus[i].ProcessOpState == oam::MAN_OFFLINE &&
@@ -10346,6 +10349,18 @@ int ProcessManager::setMySQLReplication(oam::DeviceNetworkList devicenetworklist
 				if ( remoteModuleName == masterModule )
 					continue;
 
+				// skip disabled modules
+				int opState = oam::ACTIVE;
+				bool degraded;
+				try {
+					oam.getModuleStatus(remoteModuleName, opState, degraded);
+				}
+				catch(...)
+				{}
+
+				if (opState == oam::MAN_DISABLED || opState == oam::AUTO_DISABLED)
+					continue;
+
 				// don't do PMs unless PMwithUM flag is set
 				if ( config.ServerInstallType() != oam::INSTALL_COMBINE_DM_UM_PM ) {
 					string moduleType = remoteModuleName.substr(0,MAX_MODULE_TYPE_SIZE);
@@ -10423,6 +10438,18 @@ int ProcessManager::setMySQLReplication(oam::DeviceNetworkList devicenetworklist
 				if ( remoteModuleName == masterModule )
 					continue;
 
+				// skip disabled modules
+				int opState = oam::ACTIVE;
+				bool degraded;
+				try {
+					oam.getModuleStatus(remoteModuleName, opState, degraded);
+				}
+				catch(...)
+				{}
+
+				if (opState == oam::MAN_DISABLED || opState == oam::AUTO_DISABLED)
+					continue;
+
 				// don't do PMs unless PMwithUM flag is set
 				if ( config.ServerInstallType() != oam::INSTALL_COMBINE_DM_UM_PM ) {
 					string moduleType = remoteModuleName.substr(0,MAX_MODULE_TYPE_SIZE);
@@ -10471,6 +10498,18 @@ int ProcessManager::setMySQLReplication(oam::DeviceNetworkList devicenetworklist
 			if ( remoteModuleName == masterModule )
 				continue;
 
+			// skip disabled modules
+			int opState = oam::ACTIVE;
+			bool degraded;
+			try {
+				oam.getModuleStatus(remoteModuleName, opState, degraded);
+			}
+			catch(...)
+			{}
+
+			if (opState == oam::MAN_DISABLED || opState == oam::AUTO_DISABLED)
+				continue;
+
 				ByteStream msg1;
 			ByteStream::byte requestID = oam::SLAVEREP;
 			if ( !enable ) {

procmon/processmonitor.cpp

@@ -4950,6 +4950,16 @@ int ProcessMonitor::runMasterRep(std::string& masterLogFile, std::string& master
 		{
 			string moduleName =  (*pt).DeviceName;
 
+			//skip if local module or module is not ACTIVE
+			if ( moduleName == config.moduleName() )
+				continue;
+			  
+			int opState = oam::ACTIVE;
+			bool degraded;
+			oam.getModuleStatus(moduleName, opState, degraded);
+			if (opState != oam::ACTIVE)
+				continue;
+
 			bool passwordError = false;
 
 			string moduleType = systemModuleTypeConfig.moduletypeconfig[i].ModuleType;