add more checks around the getsystems

oam/install_scripts/columnstore

@@ -81,7 +81,7 @@ start() {
 		exit 0
 	fi
 
-	(sudo mkdir -p /var/lock/subsys && touch /var/lock/subsys/columnstore) >/dev/null 2>&1
+	(sudo mkdir -p /var/lock/subsys && sudo chmod 777 /var/lock/subsys && sudo touch /var/lock/subsys/columnstore) >/dev/null 2>&1
 
 	if [ -x $InstallDir/bin/columnstore.pre-start ]; then
 		$InstallDir/bin/columnstore.pre-start

oam/install_scripts/post-install

@@ -7,11 +7,11 @@
 prefix=/usr/local
 installdir=$prefix/mariadb/columnstore
 rpmmode=install
-user=$USER
-sudo=sudo
-if [ -z "$user" ]; then
 user=root
 sudo=" "
+if [ $USER != "root"]; then
+	user=$USER
+        sudo="sudo "
 fi
 
 quiet=0

oam/install_scripts/pre-uninstall

@@ -7,12 +7,12 @@
 prefix=/usr/local
 installdir=$prefix/mariadb/columnstore
 rpmmode=install
-user=$USER
-sudo=sudo
-if [ -z "$user" ]; then
-	user=root
 sudo=" "
+if [ $USER != "root"]; then
+        user=$USER
+        sudo="sudo "
 fi
+
 quiet=0
 
 for arg in "$@"; do

oam/oamcpp/liboamcpp.cpp

@@ -9271,7 +9271,7 @@ namespace oam
 			catch(...)
 			{
 				processor.shutdown();
-				throw std::runtime_error("error");
+				throw std::runtime_error("write error");
 			}
 			
 
@@ -9282,7 +9282,7 @@ namespace oam
 			catch(...)
 			{
 				processor.shutdown();
-				throw std::runtime_error("error");
+				throw std::runtime_error("read error");
 			}
 
 			ByteStream::byte returnRequestType;
@@ -9649,6 +9649,8 @@ namespace oam
 				return true;
 			}
 		}
+      	writeLog("checkSystemRunning - system reported down", LOG_TYPE_DEBUG );
+
 		return false;
 	}
 } //namespace oam

oamapps/mcsadmin/mcsadmin.cpp

@@ -4683,7 +4683,7 @@ int processCommand(string* arguments)
 			}
 			cout << "     ";
 			cout.width(20);
-			cout << "---------------";
+			//cout << "---------------";
 			if ( AmazonElasticIPCount > 0 )
 			{
 				cout.width(20);

procmgr/processmanager.cpp

@@ -2631,8 +2631,7 @@ void processMSG(messageqcpp::IOSocket* cfIos)
 
              	case PROCESSALARM:
                 {
-                    string dbroot;
+                    log.writeLog(__LINE__,  "MSG RECEIVED: Process Alarm Message");
-                    msg >> dbroot;
 
     				ByteStream::byte alarmID;
     				std::string componentID;
@@ -2839,6 +2838,18 @@ void processMSG(messageqcpp::IOSocket* cfIos)
 								processManager.restartProcessType("DDLProc");
 								processManager.restartProcessType("DMLProc");
 								sleep(1);
+
+								string DMLmodule = config.OAMParentName();
+								if ( config.ServerInstallType() != oam::INSTALL_COMBINE_DM_UM_PM ) {
+						        	string PrimaryUMModuleName;
+        							try {
+            							oam.getSystemConfig("PrimaryUMModuleName", PrimaryUMModuleName);
+        							}
+        							catch(...) {}
+									if ( !PrimaryUMModuleName.empty() )
+										DMLmodule = PrimaryUMModuleName;
+								}
+
 								// Wait for DMLProc to be ACTIVE
 								BRM::DBRM dbrm;
 								state = AUTO_OFFLINE;
@@ -2848,7 +2859,7 @@ void processMSG(messageqcpp::IOSocket* cfIos)
 									|| state == oam::AUTO_INIT
 									|| state == oam::ROLLBACK_INIT)
 								{
-									oam.getProcessStatus("DMLProc", config.OAMParentName(), procstat);
+									oam.getProcessStatus("DMLProc", DMLmodule, procstat);
 									state = procstat.ProcessOpState;
 									if ( procstat.ProcessOpState == oam::ACTIVE)
 										break;

procmon/main.cpp

@@ -138,7 +138,9 @@ int main(int argc, char **argv)
 
 	if ( cloud == "amazon-ec2" ) {
 		if(!aMonitor.amazonIPCheck()) {
-			string cmd = startup::StartUp::installDir() + "/bin/infinidb stop > /dev/null 2>&1";
+			log.writeLog(__LINE__, "ERROR: amazonIPCheck failed, exiting", LOG_TYPE_CRITICAL);
+			sleep(2);
+			string cmd = startup::StartUp::installDir() + "/bin/columnstore stop > /dev/null 2>&1";
 			system(cmd.c_str());
 			exit(1);
 		}
@@ -377,7 +379,7 @@ int main(int argc, char **argv)
 	if ( config.OAMParentName() == oam::UnassignedName ) {
 		cerr << endl << "OAMParentModuleName == oam::UnassignedName, exiting " << endl;
 		log.writeLog(__LINE__, "OAMParentModuleName == oam::UnassignedName, exiting", LOG_TYPE_CRITICAL);
-		exit (-1);
+		exit (1);
 	}
 
 	//check if module is in a DISABLED state
@@ -562,8 +564,8 @@ int main(int argc, char **argv)
 			}
 			catch(...)
 			{
-				log.writeLog(__LINE__, "Problem getting the ParentOAMModuleName key from the Calpont System Configuration file", LOG_TYPE_ERROR);
+				log.writeLog(__LINE__, "Problem getting the ParentOAMModuleName key from the Columnstore System Configuration file", LOG_TYPE_CRITICAL);
-				exit(-1);
+				exit(1);
 			}
 		}
 
@@ -1673,7 +1675,7 @@ static void statusControlThread()
 	fProcStatMapreg.swap(region);
 	fShmProcessStatus = static_cast<shmProcessStatus*>(fProcStatMapreg.get_address());
 	if (fShmProcessStatus == 0) {
-		log.writeLog(__LINE__, "*****ProcessStatusTable shmat failed.", LOG_TYPE_ERROR);
+		log.writeLog(__LINE__, "*****ProcessStatusTable shmat failed.", LOG_TYPE_CRITICAL);
 		exit(1);
 	}
 
@@ -1735,7 +1737,7 @@ static void statusControlThread()
 	fSysStatMapreg.swap(region2);
 	fShmSystemStatus = static_cast<shmDeviceStatus*>(fSysStatMapreg.get_address());
 	if (fShmSystemStatus == 0) {
-		log.writeLog(__LINE__, "*****SystemStatusTable shmat failed.", LOG_TYPE_ERROR);
+		log.writeLog(__LINE__, "*****SystemStatusTable shmat failed.", LOG_TYPE_CRITICAL);
 		exit(1);
 	}
 
@@ -1839,7 +1841,7 @@ static void statusControlThread()
 	fShmNICStatus = static_cast<shmDeviceStatus*>(fNICStatMapreg.get_address());
 
 	if (fShmNICStatus == 0) {
-		log.writeLog(__LINE__, "*****NICStatusTable shmat failed.", LOG_TYPE_ERROR);
+		log.writeLog(__LINE__, "*****NICStatusTable shmat failed.", LOG_TYPE_CRITICAL);
 		exit(1);
 	}
 
@@ -1925,7 +1927,7 @@ static void statusControlThread()
 	fShmExtDeviceStatus = static_cast<shmDeviceStatus*>(fExtStatMapreg.get_address());
 
 	if (fShmExtDeviceStatus == 0) {
-		log.writeLog(__LINE__, "*****ExtDeviceStatusTable shmat failed.", LOG_TYPE_ERROR);
+		log.writeLog(__LINE__, "*****ExtDeviceStatusTable shmat failed.", LOG_TYPE_CRITICAL);
 		exit(1);
 	}
 
@@ -2003,7 +2005,7 @@ static void statusControlThread()
 	fShmDbrootStatus = static_cast<shmDeviceStatus*>(fdDbrootStatMapreg.get_address());
 
 	if (fShmDbrootStatus == 0) {
-		log.writeLog(__LINE__, "*****DbrootStatusTable shmat failed.", LOG_TYPE_ERROR);
+		log.writeLog(__LINE__, "*****DbrootStatusTable shmat failed.", LOG_TYPE_CRITICAL);
 		exit(1);
 	}
 

procmon/processmonitor.cpp

@@ -2597,7 +2597,7 @@ pid_t ProcessMonitor::startProcess(string processModuleType, string processName,
 		//Update Process Status: Mark Process INIT state 
 		updateProcessInfo(processName, FAILED, newProcessID);
 
-		exit(oam::API_FAILURE);
+		return (oam::API_FAILURE);
 	} 
 
 	return newProcessID;
@@ -2812,6 +2812,11 @@ void sendProcessThread(sendProcessInfo_t* t)
 	try {
 		oam.setProcessStatus(processName, config.moduleName(), state, PID);
 	}
+    catch (exception& ex)
+    {
+      	string error = ex.what();
+      	log.writeLog(__LINE__, "EXCEPTION ERROR on setProcessStatus: " + error, LOG_TYPE_ERROR);
+	}
 	catch(...)
 	{
 		log.writeLog(__LINE__, "EXCEPTION ERROR on setProcessStatus: Caught unknown exception!", LOG_TYPE_ERROR );