/******************************************************************************************
* @brief sendAlarm
*
* purpose: send a trap and log the process information
*
******************************************************************************************/
void ServerMonitor::sendAlarm(string alarmItem, ALARMS alarmID, int action, float sensorValue)
{
ServerMonitor serverMonitor;
Oam oam;
//Log this event
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add(alarmItem);
args.add(", sensor value out-of-range: ");
args.add(sensorValue);
// get current server name
string moduleName;
oamModuleInfo_t st;
try {
st = oam.getModuleInfo();
moduleName = boost::get<0>(st);
}
catch (...) {
moduleName = "Unknown Server";
}
// check if there is an active alarm above the reporting theshold
// that needs to be cleared
serverMonitor.checkAlarm(alarmItem, alarmID);
// check if Alarm is already active, don't resend
if ( !( oam.checkActiveAlarm(alarmID, moduleName, alarmItem)) ) {
SNMPManager alarmMgr;
// send alarm
alarmMgr.sendAlarmReport(alarmItem.c_str(), alarmID, action);
args.add(", Alarm set: ");
args.add(alarmID);
}
// output log
msg.format(args);
ml.logWarningMessage(msg);
return;
}
/******************************************************************************************
* @brief checkDiskAlarm
*
* purpose: check to see if an alarm(s) is set on Disk and clear if so
*
******************************************************************************************/
void ServerMonitor::checkDiskAlarm(string alarmItem, ALARMS alarmID)
{
Oam oam;
ServerMonitor serverMonitor;
// get current server name
string serverName;
oamModuleInfo_t st;
try {
st = oam.getModuleInfo();
serverName = boost::get<0>(st);
}
catch (...) {
serverName = "Unknown Server";
}
switch (alarmID) {
case NO_ALARM: // clear all alarms set if any found
if ( oam.checkActiveAlarm(DISK_USAGE_HIGH, serverName, alarmItem) )
// alarm set, clear it
clearAlarm(alarmItem, DISK_USAGE_HIGH);
if ( oam.checkActiveAlarm(DISK_USAGE_MED, serverName, alarmItem) )
// alarm set, clear it
clearAlarm(alarmItem, DISK_USAGE_MED);
if ( oam.checkActiveAlarm(DISK_USAGE_LOW, serverName, alarmItem) )
// alarm set, clear it
clearAlarm(alarmItem, DISK_USAGE_LOW);
break;
case DISK_USAGE_LOW: // clear high and medium alarms set if any found
if ( oam.checkActiveAlarm(DISK_USAGE_HIGH, serverName, alarmItem) )
// alarm set, clear it
clearAlarm(alarmItem, DISK_USAGE_HIGH);
if ( oam.checkActiveAlarm(DISK_USAGE_MED, serverName, alarmItem) )
// alarm set, clear it
clearAlarm(alarmItem, DISK_USAGE_MED);
break;
case DISK_USAGE_MED: // clear high alarms set if any found
if ( oam.checkActiveAlarm(DISK_USAGE_HIGH, serverName, alarmItem) )
// alarm set, clear it
clearAlarm(alarmItem, DISK_USAGE_HIGH);
break;
default: // none to clear
break;
} // end of switch
return;
}
/******************************************************************************************
* @brief sendResourceAlarm
*
* purpose: send a trap and log the process information
*
******************************************************************************************/
bool ServerMonitor::sendResourceAlarm(string alarmItem, ALARMS alarmID, int action, int usage)
{
ServerMonitor serverMonitor;
Oam oam;
//Log this event
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add(alarmItem);
args.add(" usage at percentage of ");
args.add(usage);
// get current module name
string moduleName;
oamModuleInfo_t st;
try {
st = oam.getModuleInfo();
moduleName = boost::get<0>(st);
}
catch (...) {
moduleName = "Unknown Server";
}
// check if there is an active alarm above the reporting theshold
// that needs to be cleared
if (alarmItem == "CPU")
serverMonitor.checkCPUAlarm(alarmItem, alarmID);
else if (alarmItem == "Local Disk" || alarmItem == "External")
serverMonitor.checkDiskAlarm(alarmItem, alarmID);
else if (alarmItem == "Local Memory")
serverMonitor.checkMemoryAlarm(alarmItem, alarmID);
else if (alarmItem == "Local Swap")
serverMonitor.checkSwapAlarm(alarmItem, alarmID);
// don't issue an alarm on thge dbroots is already issued by this or another server
if ( alarmItem.find(startup::StartUp::installDir() + "/data") == 0 ) {
// check if Alarm is already active from any module, don't resend
if ( !( oam.checkActiveAlarm(alarmID, "*", alarmItem)) ) {
SNMPManager alarmMgr;
// send alarm
alarmMgr.sendAlarmReport(alarmItem.c_str(), alarmID, action);
args.add(", Alarm set: ");
args.add(alarmID);
msg.format(args);
ml.logInfoMessage(msg);
return true;
}
else
return false;
}
else
{
// check if Alarm is already active from this module, don't resend
if ( !( oam.checkActiveAlarm(alarmID, moduleName, alarmItem)) ) {
SNMPManager alarmMgr;
// send alarm
alarmMgr.sendAlarmReport(alarmItem.c_str(), alarmID, action);
args.add(", Alarm set: ");
args.add(alarmID);
msg.format(args);
ml.logInfoMessage(msg);
return true;
}
else
return false;
}
return true;
}
/*****************************************************************************************
* @brief diskMonitor Thread
*
* purpose: Get current Local and External disk usage and report alarms
*
*****************************************************************************************/
void diskMonitor()
{
ServerMonitor serverMonitor;
Oam oam;
SystemConfig systemConfig;
ModuleTypeConfig moduleTypeConfig;
typedef std::vector<std::string> LocalFileSystems;
LocalFileSystems lfs;
struct statvfs buf;
// set defaults
int localDiskCritical = 90,
localDiskMajor = 80,
localDiskMinor = 70,
ExternalDiskCritical = 90,
ExternalDiskMajor = 80,
ExternalDiskMinor = 70;
// get module types
string moduleType;
int moduleID=-1;
string moduleName;
oamModuleInfo_t t;
try {
t = oam.getModuleInfo();
moduleType = boost::get<1>(t);
moduleID = boost::get<2>(t);
moduleName = boost::get<0>(t);
}
catch (exception& e) {}
bool Externalflag = false;
//check for external disk
DBrootList dbrootList;
if (moduleType == "pm") {
systemStorageInfo_t t;
t = oam.getStorageConfig();
if ( boost::get<0>(t) == "external")
Externalflag = true;
// get dbroot list and storage type from config file
DBRootConfigList dbrootConfigList;
oam.getPmDbrootConfig(moduleID, dbrootConfigList);
DBRootConfigList::iterator pt = dbrootConfigList.begin();
for( ; pt != dbrootConfigList.end() ; pt++)
{
int dbrootID = *pt;
string dbroot = "DBRoot" + oam.itoa(dbrootID);
string dbootdir;
try{
oam.getSystemConfig(dbroot, dbootdir);
}
catch(...) {}
if ( dbootdir.empty() || dbootdir == "" )
continue;
DBrootData dbrootData;
dbrootData.dbrootDir = dbootdir;
dbrootData.downFlag = false;
dbrootList.push_back(dbrootData);
}
}
string cloud = oam::UnassignedName;
try {
oam.getSystemConfig( "Cloud", cloud);
}
catch(...) {
cloud = oam::UnassignedName;
}
//get Gluster Config setting
string GlusterConfig = "n";
try {
oam.getSystemConfig( "GlusterConfig", GlusterConfig);
}
catch(...)
{
GlusterConfig = "n";
}
int diskSpaceCheck = 0;
while(true)
{
SystemStatus systemstatus;
try {
oam.getSystemStatus(systemstatus);
}
catch (exception& ex)
{}
if (systemstatus.SystemOpState != oam::ACTIVE ) {
sleep(5);
continue;
}
// Get Local/External Disk Mount points to monitor and associated thresholds
try {
oam.getSystemConfig (moduleTypeConfig);
localDiskCritical = moduleTypeConfig.ModuleDiskCriticalThreshold;
localDiskMajor = moduleTypeConfig.ModuleDiskMajorThreshold;
localDiskMinor = moduleTypeConfig.ModuleDiskMinorThreshold;
DiskMonitorFileSystems::iterator p = moduleTypeConfig.FileSystems.begin();
for( ; p != moduleTypeConfig.FileSystems.end() ; p++)
{
string fs = *p;
lfs.push_back(fs);
if (DISK_DEBUG) {
//Log this event
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Local Config File System to monitor =");
args.add(fs);
msg.format(args);
ml.logDebugMessage(msg);
}
}
} catch (...)
{
sleep(5);
continue;
}
// get External info
try
{
oam.getSystemConfig(systemConfig);
} catch (...)
{
sleep(5);
continue;
}
if (Externalflag) {
// get External info
try
{
ExternalDiskCritical = systemConfig.ExternalCriticalThreshold;
ExternalDiskMajor = systemConfig.ExternalMajorThreshold;
ExternalDiskMinor = systemConfig.ExternalMinorThreshold;
} catch (...)
{
sleep(5);
continue;
}
}
//check for local file systems
LocalFileSystems::iterator p = lfs.begin();
while(p != lfs.end())
{
string deviceName = *p;
++p;
string fileName;
// check local
if ( deviceName == "/") {
fileName = deviceName + "usr/local/Calpont/releasenum";
}
else
{
fileName = deviceName + "/000.dir";
}
uint64_t totalBlocks;
uint64_t usedBlocks;
if (!statvfs(fileName.c_str(), &buf)) {
uint64_t blksize, blocks, freeblks, free;
blksize = buf.f_bsize;
blocks = buf.f_blocks;
freeblks = buf.f_bfree;
totalBlocks = blocks * blksize;
free = freeblks * blksize;
usedBlocks = totalBlocks - free;
}
else
continue;
int64_t diskUsage = 0;
if ( totalBlocks == 0 ) {
diskUsage = 0;
//Log this event
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Total Disk Usage is set to 0");
msg.format(args);
ml.logWarningMessage(msg);
}
else
diskUsage = (usedBlocks / (totalBlocks / 100)) + 1;
SMSystemDisk sd;
sd.deviceName = deviceName;
sd.usedPercent = diskUsage;
sd.totalBlocks = totalBlocks;
sd.usedBlocks = usedBlocks;
sdl.push_back(sd);
if (DISK_DEBUG)
cout << "Disk Usage for " << deviceName << " is " << diskUsage << endl;
if ( diskSpaceCheck == 0 )
{
if (diskUsage >= localDiskCritical && localDiskCritical > 0 ) {
//adjust if over 100%
if ( diskUsage > 100 )
diskUsage = 100;
if ( serverMonitor.sendResourceAlarm(deviceName, DISK_USAGE_HIGH, SET, (int) diskUsage) )
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Local Disk above Critical Disk threshold with a percentage of ");
args.add((int) diskUsage);
msg.format(args);
ml.logInfoMessage(msg);
}
}
else if (diskUsage >= localDiskMajor && localDiskMajor > 0 ) {
if (serverMonitor.sendResourceAlarm(deviceName, DISK_USAGE_MED, SET, (int) diskUsage))
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Local Disk above Major Disk threshold with a percentage of ");
args.add((int) diskUsage);
msg.format(args);
ml.logInfoMessage(msg);
}
}
else if (diskUsage >= localDiskMinor && localDiskMinor > 0 ) {
if ( serverMonitor.sendResourceAlarm(deviceName, DISK_USAGE_LOW, SET, (int) diskUsage))
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Local Disk above Minor Disk threshold with a percentage of ");
args.add((int) diskUsage);
msg.format(args);
ml.logInfoMessage(msg);
}
}
else
serverMonitor.checkDiskAlarm(deviceName);
}
//check for external file systems/devices
if (Externalflag ||
(!Externalflag && GlusterConfig == "y" && moduleType == "pm") ){
try
{
DBRootConfigList dbrootConfigList;
oam.getPmDbrootConfig(moduleID, dbrootConfigList);
DBRootConfigList::iterator pt = dbrootConfigList.begin();
for( ; pt != dbrootConfigList.end() ; pt++)
{
int dbroot = *pt;
string deviceName = systemConfig.DBRoot[dbroot-1];
string fileName = deviceName + "/000.dir";
if (DISK_DEBUG) {
//Log this event
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("DBRoots monitoring");
args.add(dbroot);
args.add(" ,file system =" );
args.add(fileName);
msg.format(args);
ml.logDebugMessage(msg);
}
uint64_t totalBlocks;
uint64_t usedBlocks;
if (!statvfs(fileName.c_str(), &buf)) {
uint64_t blksize, blocks, freeblks, free;
blksize = buf.f_bsize;
blocks = buf.f_blocks;
freeblks = buf.f_bfree;
totalBlocks = blocks * blksize;
free = freeblks * blksize;
usedBlocks = totalBlocks - free;
}
else
{
SMSystemDisk sd;
sd.deviceName = deviceName;
sd.usedPercent = 0;
sd.totalBlocks = 0;
sd.usedBlocks = 0;
sdl.push_back(sd);
continue;
}
int diskUsage = 0;
if ( totalBlocks == 0 ) {
diskUsage = 0;
//Log this event
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Total Disk Usage is set to 0");
msg.format(args);
ml.logWarningMessage(msg);
}
else
diskUsage = (usedBlocks / (totalBlocks / 100)) + 1;
SMSystemDisk sd;
sd.deviceName = deviceName;
sd.usedPercent = diskUsage;
sd.totalBlocks = totalBlocks;
sd.usedBlocks = usedBlocks;
sdl.push_back(sd);
if (DISK_DEBUG)
cout << "Disk Usage for " << deviceName << " is " << diskUsage << endl;
if (diskUsage >= ExternalDiskCritical && ExternalDiskCritical > 0 ) {
//adjust if over 100%
if ( diskUsage > 100 )
diskUsage = 100;
if ( serverMonitor.sendResourceAlarm(deviceName, DISK_USAGE_HIGH, SET, diskUsage))
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Disk usage for");
args.add(deviceName);
args.add(" above Critical Disk threshold with a percentage of ");
args.add((int) diskUsage);
msg.format(args);
ml.logInfoMessage(msg);
}
}
else if (diskUsage >= ExternalDiskMajor && ExternalDiskMajor > 0 ) {
if ( serverMonitor.sendResourceAlarm(deviceName, DISK_USAGE_MED, SET, diskUsage))
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Disk usage for");
args.add(deviceName);
args.add(" above Major Disk threshold with a percentage of ");
args.add((int) diskUsage);
msg.format(args);
ml.logInfoMessage(msg);
}
}
else if (diskUsage >= ExternalDiskMinor && ExternalDiskMinor > 0 ) {
if ( serverMonitor.sendResourceAlarm(deviceName, DISK_USAGE_LOW, SET, diskUsage))
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Disk usage for");
args.add(deviceName);
args.add(" above Minor Disk threshold with a percentage of ");
args.add((int) diskUsage);
msg.format(args);
ml.logInfoMessage(msg);
}
}
else
serverMonitor.checkDiskAlarm(deviceName);
}
}
catch (exception& e)
{
cout << endl << "**** getPmDbrootConfig Failed : " << e.what() << endl;
}
}
}
//check OAM dbroot test flag to validate dbroot exist if on pm
if ( moduleName.find("pm") != string::npos ) {
//check OAM dbroot test flag to validate dbroot exist
if ( dbrootList.size() != 0 ) {
DBrootList::iterator p = dbrootList.begin();
while ( p != dbrootList.end() )
{
//get dbroot directory
string dbrootDir = (*p).dbrootDir;
string dbrootName;
string dbrootID;
//get dbroot name
string::size_type pos = dbrootDir.rfind("/",80);
if (pos != string::npos)
dbrootName = dbrootDir.substr(pos+1,80);
//get ID
dbrootID = dbrootName.substr(4,80);
string fileName = dbrootDir + "/OAMdbrootCheck";
// retry in case we hit the remount window
for ( int retry = 0 ; ; retry++ )
{
bool fail = false;
//first test, check if OAMdbrootCheck exists
ifstream file (fileName.c_str());
if (!file)
fail = true;
else
{ //second test for amazon, check volume status
if ( cloud != oam::UnassignedName ) {
string volumeNameID = "PMVolumeName" + dbrootID;
string volumeName = oam::UnassignedName;
try {
oam.getSystemConfig( volumeNameID, volumeName);
}
catch(...)
{}
if ( volumeName.empty() || volumeName == oam::UnassignedName )
fail = false;
else
{
string status = oam.getEC2VolumeStatus(volumeName);
if ( status == "attached" )
fail = false;
else
{
fail = true;
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("dbroot monitoring: Volume not attached");
args.add(volumeName);
args.add("/");
args.add(dbrootName);
msg.format(args);
ml.logCriticalMessage(msg);
}
}
}
else
fail = false;
}
if (fail) {
//double check system status before reporting any error BUG 5078
SystemStatus systemstatus;
try {
oam.getSystemStatus(systemstatus);
}
catch (exception& ex)
{}
if (systemstatus.SystemOpState != oam::ACTIVE ) {
break;
}
if ( retry < 10 ) {
sleep(3);
continue;
}
else
{
if ( !(*p).downFlag ) {
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("dbroot monitoring: Lost access to ");
args.add(dbrootDir);
msg.format(args);
ml.logCriticalMessage(msg);
oam.sendDeviceNotification(dbrootName, DBROOT_DOWN, moduleName);
(*p).downFlag = true;
try{
oam.setDbrootStatus(dbrootID, oam::AUTO_OFFLINE);
}
catch (exception& ex)
{}
break;
}
}
}
else
{
if ( (*p).downFlag ) {
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("dbroot monitoring: Access back to ");
args.add(dbrootDir);
msg.format(args);
ml.logInfoMessage(msg);
oam.sendDeviceNotification(dbrootName, DBROOT_UP, moduleName);
(*p).downFlag = false;
try{
oam.setDbrootStatus(dbrootID, oam::ACTIVE);
}
catch (exception& ex)
{}
}
file.close();
break;
}
}
p++;
}
}
}
//do Gluster status check, if configured
if ( GlusterConfig == "y")
{
bool pass = true;
string errmsg = "unknown";
try {
string arg1 = "";
string arg2 = "";
int ret = oam.glusterctl(oam::GLUSTER_STATUS, arg1, arg2, errmsg);
if ( ret != 0 )
{
cerr << "FAILURE: Status check error: " + errmsg << endl;
pass = false;
}
}
catch (exception& e)
{
cerr << endl << "**** glusterctl API exception: " << e.what() << endl;
cerr << "FAILURE: Status check error" << endl;
pass = false;
}
catch (...)
{
cerr << endl << "**** glusterctl API exception: UNKNOWN" << endl;
cerr << "FAILURE: Status check error" << endl;
pass = false;
}
if ( !pass )
{ // issue log and alarm
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Gluster Status check failure error msg: ");
args.add(errmsg);
msg.format(args);
ml.logWarningMessage(msg);
serverMonitor.sendResourceAlarm(errmsg, GLUSTER_DISK_FAILURE, SET, 0);
}
}
// sleep 10 seconds
sleep(MONITOR_PERIOD/6);
//check disk space every 10 minutes
diskSpaceCheck++;
if ( diskSpaceCheck >= 60 )
diskSpaceCheck = 0;
lfs.clear();
sdl.clear();
} // end of while loop
}
/*****************************************************************************************
* @brief setSNMPModuleName API
*
* purpose: Set SNMP Module name in the snmpdx.conf file
*
*****************************************************************************************/
void SNMPManager::setSNMPModuleName ()
{
// get current Module name
Oam oam;
string ModuleName;
oamModuleInfo_t st;
try {
st = oam.getModuleInfo();
ModuleName = boost::get<0>(st);
}
catch (...) {
ModuleName = "Unknown Report Module";
}
string agentName = SUB_AGENT;
string fileName;
makeFileName (agentName, fileName);
vector <string> lines;
ifstream oldFile (fileName.c_str());
if (!oldFile) throw runtime_error ("No configuration file found");
char line[200];
string buf;
string newLine;
string newLine1;
string delimiters = " ";
while (oldFile.getline(line, 200))
{
buf = line;
string::size_type pos = buf.find("ModuleNameStub",0);
if (pos != string::npos)
{
newLine = buf.substr(0, pos);
newLine.append(ModuleName);
string::size_type pos1 = buf.find("|",pos);
if (pos1 != string::npos)
{
newLine1 = buf.substr(pos1, 200);
newLine.append(newLine1);
}
buf = newLine;
}
// output to temp file
lines.push_back(buf);
}
oldFile.close();
unlink (fileName.c_str());
ofstream newFile (fileName.c_str());
// create new file
int fd = open(fileName.c_str(), O_RDWR|O_CREAT, 0666);
// Aquire an exclusive lock
if (flock(fd,LOCK_EX) == -1) {
throw runtime_error ("Lock SNMP configuration file error");
}
copy(lines.begin(), lines.end(), ostream_iterator<string>(newFile, "\n"));
newFile.close();
// Release lock
if (flock(fd,LOCK_UN) == -1)
{
throw runtime_error ("Release lock SNMP configuration file error");
}
close(fd);
}
int main(int argc, char** argv)
{
int c;
string pname(argv[0]);
bool vflg = false;
bool dflg = false;
bool xflg = false;
string configFile;
opterr = 0;
while ((c = getopt(argc, argv, "c:vdxh")) != EOF)
switch (c)
{
case 'v':
vflg = true;
break;
case 'd':
dflg = true;
break;
case 'c':
configFile = optarg;
break;
case 'x':
xflg = true;
break;
case 'h':
case '?':
default:
usage(pname);
return (c == 'h' ? 0 : 1);
break;
}
if ((argc - optind) < 3)
{
usage(pname);
return 1;
}
#ifdef COMMUNITY_KEYRANGE
//No OAM in CE...
dflg = true;
#endif
Oam oam;
oamModuleInfo_t t;
bool parentOAMModuleFlag = true;
string parentOAMModule = " ";
int serverInstallType = oam::INSTALL_COMBINE_DM_UM_PM;
//get local module info; validate running on Active Parent OAM Module
try {
t = oam.getModuleInfo();
parentOAMModuleFlag = boost::get<4>(t);
parentOAMModule = boost::get<3>(t);
serverInstallType = boost::get<5>(t);
}
catch (exception&) {
parentOAMModuleFlag = true;
}
if (!dflg && !parentOAMModuleFlag)
{
cerr << "Exiting, setConfig can only be run on the Active "
"OAM Parent Module '" << parentOAMModule << "'" << endl;
return 2;
}
Config* cf;
if (configFile.length() > 0)
cf = Config::makeConfig(configFile);
else
cf = Config::makeConfig();
if (vflg)
cout << "Using config file: " << cf->configFile() << endl;
if (xflg)
cf->delConfig(argv[optind + 0], argv[optind + 1]);
else
cf->setConfig(argv[optind + 0], argv[optind + 1], argv[optind + 2]);
cf->write();
if (dflg || serverInstallType == oam::INSTALL_COMBINE_DM_UM_PM)
return 0;
//get number of pms
string count = cf->getConfig("PrimitiveServers", "Count");
try {
oam.distributeConfigFile();
//sleep to give time for change to be distributed
sleep(atoi(count.c_str()));
}
catch (...) {
return 1;
}
return 0;
}
int main(int argc, char *argv[])
{
Oam oam;
string installDir(startup::StartUp::installDir());
Config* sysConfig = Config::makeConfig();
string SystemSection = "SystemConfig";
string InstallSection = "Installation";
bool HARDWARE = false;
bool SOFTWARE = false;
bool CONFIG = false;
bool DBMS = false;
bool RESOURCE = false;
bool LOG = false;
bool BULKLOG = false;
bool HADOOP = false;
//get current time and date
time_t now;
now = time(NULL);
struct tm tm;
localtime_r(&now, &tm);
char timestamp[200];
strftime (timestamp, 200, "%m:%d:%y-%H:%M:%S", &tm);
currentDate = timestamp;
char helpArg[3] = "-h";
// Get System Name
try{
oam.getSystemConfig("SystemName", systemName);
}
catch(...)
{
systemName = "unassigned";
}
// get Local Module Name and Server Install Indicator
string singleServerInstall;
oamModuleInfo_t st;
try {
st = oam.getModuleInfo();
localModule = boost::get<0>(st);
}
catch (...) {
cout << endl << "**** Failed : Failed to read Local Module Name" << endl;
exit(-1);
}
try{
oam.getSystemConfig("SingleServerInstall", singleServerInstall);
}
catch(...)
{
singleServerInstall = "y";
}
if (argc == 1) {
argv[1] = &helpArg[0];
argc = 2;
}
string DataFilePlugin;
try{
DataFilePlugin = sysConfig->getConfig(SystemSection, "DataFilePlugin");
}
catch(...)
{
cout << "ERROR: Problem accessing InfiniDB configuration file" << endl;
exit(-1);
}
for( int i = 1; i < argc; i++ )
{
if( string("-h") == argv[i] ) {
cout << endl;
cout << "'calpontSupport' generates a Set of System Support Report Files in a tar file" << endl;
cout << "called calpontSupportReport.'system-name'.tar.gz in the local directory." << endl;
cout << "It should be run on the server with the DBRM front-end." << endl;
cout << "Check the Admin Guide for additional information." << endl;
cout << endl;
cout << "Usage: calpontSupport [-h][-a][-hw][-s][-c][-db][-r][-l][-bl][-lc][-p 'root-password'][-mp 'mysql-root-password'][-de]";
// if hdfs set up print the hadoop option
if (!DataFilePlugin.empty())
cout << "[-hd]";
cout << endl;
cout << " -h help" << endl;
cout << " -a Output all Reports (excluding Bulk Logs Reports)" << endl;
cout << " -hw Output Hardware Reports only" << endl;
cout << " -s Output Software Reports only" << endl;
cout << " -c Output Configuration/Status Reports only" << endl;
cout << " -db Output DBMS Reports only" << endl;
cout << " -r Output Resource Reports only" << endl;
cout << " -l Output Calpont Log/Alarms Reports only" << endl;
cout << " -bl Output Calpont Bulk Log Reports only" << endl;
cout << " -lc Output Reports for Local Server only" << endl;
cout << " -p password (multi-server systems), root-password or 'ssh' to use 'ssh keys'" << endl;
cout << " -mp mysql root user password" << endl;
cout << " -de Debug Flag" << endl;
// if hdfs set up print the hadoop option
if (!DataFilePlugin.empty())
cout << " -hd Output hadoop reports only" << endl;
exit (0);
}
else
{
if( string("-a") == argv[i] ) {
HARDWARE = true;
SOFTWARE = true;
CONFIG = true;
DBMS = true;
RESOURCE = true;
LOG = true;
HADOOP = (DataFilePlugin.empty()? false : true);
}
else if( string("-hw") == argv[i] )
HARDWARE = true;
else if( string("-s") == argv[i] )
SOFTWARE = true;
else if( string("-c") == argv[i] )
CONFIG = true;
else if( string("-db") == argv[i] )
DBMS = true;
else if( string("-r") == argv[i] )
RESOURCE = true;
else if( string("-l") == argv[i] )
LOG = true;
else if( string("-bl") == argv[i] )
BULKLOG = true;
else if( string("-lc") == argv[i] )
LOCAL = true;
else if( string("-p") == argv[i] ) {
i++;
if ( argc == i ) {
cout << "ERROR: missing root password argument" << endl;
exit(-1);
}
rootPassword = argv[i];
//add single quote for special characters
if ( rootPassword != "ssh" )
{
rootPassword = "******" + rootPassword + "'";
}
}
else if( string("-mp") == argv[i] ) {
i++;
if ( argc == i ) {
cout << "ERROR: missing mysql root user password argument" << endl;
exit(-1);
}
mysqlpw = argv[i];
mysqlpw = "'" + mysqlpw + "'";
}
else if( string("-de") == argv[i] )
debug_flag = "1";
else if ( string("-hd") == argv[i] )
{
HADOOP = (DataFilePlugin.empty()? false : true);
}
else
{
cout << "Invalid Option of '" << argv[i] << "', run with '-h' for help" << endl;
exit (1);
}
}
}
//default to -a if nothing is set
if ( !HARDWARE && !SOFTWARE && !CONFIG && !DBMS && !RESOURCE && !LOG && !BULKLOG && !HADOOP)
{
HARDWARE = true;
SOFTWARE = true;
CONFIG = true;
DBMS = true;
RESOURCE = true;
LOG = true;
HADOOP = (DataFilePlugin.empty()? false : true);
}
//get Parent OAM Module Name and setup of it's Custom OS files
string PrimaryUMModuleName;
try{
PrimaryUMModuleName = sysConfig->getConfig(SystemSection, "PrimaryUMModuleName");
}
catch(...)
{
cout << "ERROR: Problem getting Parent OAM Module Name" << endl;
exit(-1);
}
if ( PrimaryUMModuleName == "unassigned" )
PrimaryUMModuleName = localModule;
if ( (localModule != PrimaryUMModuleName) && DBMS )
{
char* pcommand = 0;
char *p;
string argument = "n";
while(true)
{
cout << endl << "You selected to get the DBMS data." << endl;
cout << "You need to run the calpontSupport command on module '" << PrimaryUMModuleName << "' to get that information." << endl;
cout << "Or you can proceed on to get all data except the DBMS." << endl;
pcommand = readline(" Do you want to proceed: (y or n) [n]: ");
if (pcommand && *pcommand) {
p = strtok(pcommand," ");
argument = p;
free(pcommand);
pcommand = 0;
}
if (pcommand) {
free(pcommand);
pcommand = 0;
}
if( argument == "y") {
cout << endl;
break;
}
else if( argument == "n")
exit (1);
}
}
//get number of worker-nodes, will tell us if a single server system
//get Parent OAM Module Name and setup of it's Custom OS files
try{
string NumWorkers = sysConfig->getConfig("DBRM_Controller", "NumWorkers");
if ( NumWorkers == "1" )
singleServerInstall = "y";
}
catch(...)
{}
if ( singleServerInstall == "n" && !LOCAL)
if ( HARDWARE || SOFTWARE || CONFIG || RESOURCE || LOG || HADOOP )
if ( rootPassword.empty() ) {
cout << "ERROR: Multi-Module System, Password Argument required or use '-lc' option, check help for more information" << endl;
exit(-1);
}
//get Parent OAM Module Name and setup of it's Custom OS files
//string parentOAMModuleName;
ChildModule parentOAMModule;
try{
parentOAMModule.moduleName = sysConfig->getConfig(SystemSection, "ParentOAMModuleName");
}
catch(...)
{
cout << "ERROR: Problem getting Parent OAM Module Name" << endl;
exit(-1);
}
//Get list of configured system modules
SystemModuleTypeConfig sysModuleTypeConfig;
try{
oam.getSystemConfig(sysModuleTypeConfig);
}
catch(...)
{
cout << "ERROR: Problem reading the Calpont System Configuration file" << endl;
exit(-1);
}
string ModuleSection = "SystemModuleConfig";
for ( unsigned int i = 0 ; i < sysModuleTypeConfig.moduletypeconfig.size(); i++)
{
string moduleType = sysModuleTypeConfig.moduletypeconfig[i].ModuleType;
int moduleCount = sysModuleTypeConfig.moduletypeconfig[i].ModuleCount;
if ( moduleCount == 0 )
//no modules equipped for this Module Type, skip
continue;
//get IP addresses and Host Names
DeviceNetworkList::iterator listPT = sysModuleTypeConfig.moduletypeconfig[i].ModuleNetworkList.begin();
for( ; listPT != sysModuleTypeConfig.moduletypeconfig[i].ModuleNetworkList.end() ; listPT++)
{
string moduleName = (*listPT).DeviceName;
HostConfigList::iterator pt1 = (*listPT).hostConfigList.begin();
string moduleIPAddr = (*pt1).IPAddr;
string moduleHostName = (*pt1).HostName;
if ( moduleName == localModule) {
localModuleHostName = moduleHostName;
}
//save Child modules
if ( moduleName != localModule && moduleType != "xm") {
childmodule.moduleName = moduleName;
childmodule.moduleIP = moduleIPAddr;
childmodule.hostName = moduleHostName;
childmodulelist.push_back(childmodule);
}
if (moduleName == parentOAMModule.moduleName)
{
parentOAMModule.moduleIP = moduleIPAddr;
parentOAMModule.hostName = moduleHostName;
parentOAMModule.moduleName = moduleName;
}
}
} //end of i for loop
// create a clean Calpont Support Report
system("rm -f *_configReport.txt");
system("rm -f *_dbmsReport.txt");
system("rm -f *_hardwareReport.txt");
system("rm -f *_logReport.txt");
system("rm -f *_bulklogReport.txt");
system("rm -f *_resourceReport.txt");
system("rm -f *_softwareReport.txt");
system("rm -f hadoopReport.txt");
//
// Software
//
if ( SOFTWARE ) {
string reportType = "software";
pthread_t reportthread;
int status = pthread_create (&reportthread, NULL, (void*(*)(void*)) &reportThread, &reportType);
if ( status != 0 ) {
cout << "ERROR: reportthread: pthread_create failed, return status = " + oam.itoa(status);
}
sleep(5);
}
//
// Configuration
//
if ( CONFIG ) {
string reportType = "config";
pthread_t reportthread;
int status = pthread_create (&reportthread, NULL, (void*(*)(void*)) &reportThread, &reportType);
if ( status != 0 ) {
cout << "ERROR: reportthread: pthread_create failed, return status = " + oam.itoa(status);
}
sleep(5);
}
//
// Alarms and Calpont Logs
//
if ( LOG ) {
string reportType = "log";
pthread_t reportthread;
int status = pthread_create (&reportthread, NULL, (void*(*)(void*)) &reportThread, &reportType);
if ( status != 0 ) {
cout << "ERROR: reportthread: pthread_create failed, return status = " + oam.itoa(status);
}
sleep(5);
}
//
// Bulk Logs
//
if ( BULKLOG ) {
string reportType = "bulklog";
pthread_t reportthread;
int status = pthread_create (&reportthread, NULL, (void*(*)(void*)) &reportThread, &reportType);
if ( status != 0 ) {
cout << "ERROR: reportthread: pthread_create failed, return status = " + oam.itoa(status);
}
sleep(5);
}
//
// Hardware
//
if ( HARDWARE ) {
string reportType = "hardware";
pthread_t reportthread;
int status = pthread_create (&reportthread, NULL, (void*(*)(void*)) &reportThread, &reportType);
if ( status != 0 ) {
cout << "ERROR: reportthread: pthread_create failed, return status = " + oam.itoa(status);
}
sleep(5);
}
//
// Resources
//
if ( RESOURCE ) {
string reportType = "resource";
pthread_t reportthread;
int status = pthread_create (&reportthread, NULL, (void*(*)(void*)) &reportThread, &reportType);
if ( status != 0 ) {
cout << "ERROR: reportthread: pthread_create failed, return status = " + oam.itoa(status);
}
sleep(5);
}
//
// DBMS
//
if ( DBMS ) {
system("rm -f calpontSupportReport.txt;touch calpontSupportReport.txt");
title();
system("echo '=======================================================================' >> calpontSupportReport.txt");
system("echo '= DBMS Report =' >> calpontSupportReport.txt");
system("echo '=======================================================================' >> calpontSupportReport.txt");
// run DBMS report on local server
cout << "Get dbms report data for " << localModule << endl;
bool FAILED = false;
if ( localModule != PrimaryUMModuleName )
{
cout << " FAILED: run calpontSupport on '" << PrimaryUMModuleName << "' to get the dbrm report" << endl;
FAILED = true;
}
else
{
// check if mysql is supported and get info
string calpontMysql = installDir + "/mysql/bin/mysql --defaults-file=" + installDir + "/mysql/my.cnf -u root ";
string cmd = calpontMysql + " -e 'status' > /tmp/idbmysql.log 2>&1";
system(cmd.c_str());
//check for mysql password set
string pwprompt = " ";
if (oam.checkLogStatus("/tmp/idbmysql.log", "ERROR 1045") ) {
cout << "NOTE: MySQL root user password is set" << endl;
//needs a password, was password entered on command line
if ( mysqlpw == " " )
{ //go check my.cnf
string file = installDir + "/mysql/my.cnf";
ifstream oldFile (file.c_str());
vector <string> lines;
char line[200];
string buf;
while (oldFile.getline(line, 200))
{
buf = line;
string::size_type pos = buf.find("password",0);
if (pos != string::npos) {
string::size_type pos1 = buf.find("=",0);
if (pos1 != string::npos) {
pos = buf.find("#",0);
if (pos == string::npos) {
//password arg in my.cnf, go get password
cout << "NOTE: Using password from my.cnf" << endl;
mysqlpw = buf.substr(pos1+1,80);
cout << mysqlpw << endl;
break;
}
}
}
}
oldFile.close();
if ( mysqlpw == " " )
{
cout << "NOTE: No password provide on command line or found uncommented in my.cnf" << endl;
cout << endl;
string prompt = " *** Enter MySQL password > ";
mysqlpw = getpass(prompt.c_str());
}
}
//check for mysql password set
pwprompt = "--password="******" -e 'status' > /tmp/idbmysql.log 2>&1";
system(cmd.c_str());
if (oam.checkLogStatus("/tmp/idbmysql.log", "ERROR 1045") ) {
cout << "FAILED: Failed login using MySQL root user password '" << mysqlpw << "'" << endl;
FAILED = true;
}
}
if (!FAILED)
{
// check if mysql is supported and get info
string calpontMysql = installDir + "/mysql/bin/mysql --defaults-file=" + installDir + "/mysql/my.cnf -u root " + pwprompt;
string cmd = calpontMysql + " -V > /dev/null 2>&1";
int ret = system(cmd.c_str());
if ( WEXITSTATUS(ret) == 0) {
// run DBMS report info
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS InfiniDB Mysql Version ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " -e status ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " -e 'status' >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql Calpont System Column ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " -e desc calpontsys.syscolumn ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " -e 'desc calpontsys.syscolumn;' >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql Calpont System Table ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " -e desc calpontsys.systable ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " -e 'desc calpontsys.systable;' >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql Calpont System Catalog Data ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " calpontsys < " + installDir + "/mysql/dumpcat_mysql.sql ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " calpontsys < " + installDir + "/mysql/dumpcat_mysql.sql >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql Calpont System Table Data ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " -e select * from calpontsys.systable ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " -e 'select * from calpontsys.systable;' >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql Calpont Usernames ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " -e show databases ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " -e 'show databases;' >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql InfiniDB variables ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# " + calpontMysql + " show variables ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = calpontMysql + " -e 'show variables;' >> calpontSupportReport.txt";
system(cmd.c_str());
}
}
}
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** Database Size Report ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
string cmd = "echo '################# /bin/databaseSizeReport ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = installDir + "/bin/databaseSizeReport >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** DBMS Mysql InfiniDB config file ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# cat /mysql/my.cnf ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "cat " + installDir + "/mysql/my.cnf 2>/dev/null >> calpontSupportReport.txt";
system(cmd.c_str());
system("echo ' ' >> calpontSupportReport.txt");
system("echo '******************** Active Queries ********************' >> calpontSupportReport.txt");
system("echo ' ' >> calpontSupportReport.txt");
cmd = "echo '################# mcsadmin getActiveSqlStatement ################# ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "echo ' ' >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = installDir + "/bin/mcsadmin getActiveSqlStatement >> calpontSupportReport.txt";
system(cmd.c_str());
cmd = "cat calpontSupportReport.txt > " + localModule + "_dbmsReport.txt";
system(cmd.c_str());
}
//
// HADOOP
//
if (HADOOP)
{
if (LOCAL || childmodulelist.empty())
{
cout << "Get hadoop report data" << endl;
string cmd = installDir + "/bin/hadoopReport.sh " + localModule + " " + installDir + "\n";
cmd += " mv -f /tmp/hadoopReport.txt .";
FILE* pipe = popen(cmd.c_str(), "r");
if (!pipe)
{
cout << "Failed to get a pipe for hadoop health check commands" << endl;
exit(-1);
}
pclose(pipe);
}
else
{
// only get hadoop report from parentOAMModule, because it's consistant view.
parentmodulelist.push_back(parentOAMModule);
threadInfo_t *st = new threadInfo_t;
ChildModuleList::iterator iter = parentmodulelist.begin();
*st = boost::make_tuple(iter, "hadoop");
pthread_t hdthread;
int status = pthread_create (&hdthread, NULL, (void*(*)(void*)) &childReportThread, st);
if ( status != 0 )
{
cout << "ERROR: childreportthread: pthread_create failed, return status = " + oam.itoa(status) << endl;
}
}
}
//wait for all threads to complete
sleep(5);
int wait = 0;
while (true)
{
//cout << "check " << runningThreads << endl;
if (runningThreads < 1)
break;
sleep(2);
wait++;
// give it 60 minutes to complete
if ( wait >= 3600 * 5)
{
cout << "Timed out (60 minutes) waiting for Requests to complete" << endl;
}
}
system("rm -f calpontSupportReport.txt");
system("unix2dos *Report.txt > /dev/null 2>&1");
system("rm -rf calpontSupportReport;mkdir calpontSupportReport;mv *Report.txt calpontSupportReport/. > /dev/null 2>&1;mv *Report.tar.gz calpontSupportReport/. > /dev/null 2>&1");
string cmd = "tar -zcf calpontSupportReport." + systemName + ".tar.gz calpontSupportReport/*";
system(cmd.c_str());
cout << endl << "Calpont Support Script Successfully completed, files located in calpontSupportReport." + systemName + ".tar.gz" << endl;
}
void procmonMonitor()
{
ServerMonitor serverMonitor;
Oam oam;
//wait before monitoring is started
sleep(60);
// get current server name
string moduleName;
oamModuleInfo_t st;
try {
st = oam.getModuleInfo();
moduleName = boost::get<0>(st);
}
catch (...) {
// Critical error, Log this event and exit
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("Failed to read local module Info");
msg.format(args);
ml.logCriticalMessage(msg);
exit(-1);
}
string msgPort = moduleName + "_ProcessMonitor";
int heartbeatCount = 0;
// loop forever monitoring Local Process Monitor
while(true)
{
ByteStream msg;
ByteStream::byte requestID = LOCALHEARTBEAT;
msg << requestID;
try
{
MessageQueueClient mqRequest(msgPort);
mqRequest.write(msg);
// wait 10 seconds for response
ByteStream::byte returnACK;
ByteStream::byte returnRequestID;
ByteStream::byte requestStatus;
ByteStream receivedMSG;
struct timespec ts = { 10, 0 };
try {
receivedMSG = mqRequest.read(&ts);
if (receivedMSG.length() > 0) {
receivedMSG >> returnACK;
receivedMSG >> returnRequestID;
receivedMSG >> requestStatus;
if ( returnACK == oam::ACK && returnRequestID == requestID) {
// ACK for this request
heartbeatCount = 0;
}
}
else
{
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("procmonMonitor: ProcMon Msg timeout!!!");
msg.format(args);
ml.logWarningMessage(msg);
heartbeatCount++;
if ( heartbeatCount > 2 ) {
//Process Monitor not responding, restart it
system("pkill ProcMon");
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("procmonMonitor: Restarting ProcMon");
msg.format(args);
ml.logWarningMessage(msg);
sleep(60);
heartbeatCount = 0;
}
}
mqRequest.shutdown();
}
catch (SocketClosed &ex) {
string error = ex.what();
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("procmonMonitor: EXCEPTION ERROR on mqRequest.read: " + error);
msg.format(args);
ml.logErrorMessage(msg);
}
catch (...) {
LoggingID lid(SERVER_MONITOR_LOG_ID);
MessageLog ml(lid);
Message msg;
Message::Args args;
args.add("procmonMonitor: EXCEPTION ERROR on mqRequest.read: Caught unknown exception");
msg.format(args);
ml.logErrorMessage(msg);
}
}
catch (exception& ex)
