You've already forked mariadb-columnstore-engine
mirror of
https://github.com/mariadb-corporation/mariadb-columnstore-engine.git
synced 2025-10-25 21:17:42 +03:00
Reformat all code to coding standard
This commit is contained in:
Andrew Hutchings
@@ -38,13 +38,13 @@ pthread_mutex_t CPU_LOCK;
|
||||
|
||||
/**
|
||||
* constants define
|
||||
*/
|
||||
*/
|
||||
|
||||
const std::string FE_MOUNT_DIR = "/var/log/mariadb/columnstore/"; // FE mount dir
|
||||
const int MONITOR_FREQ = 5; // monitor frequency in sec
|
||||
const int LOG_FREQ = 900; // log frequency in sec
|
||||
const int RESOURCE_DEBUG = false;
|
||||
static unsigned int usage[LOG_FREQ/MONITOR_FREQ];
|
||||
static unsigned int usage[LOG_FREQ / MONITOR_FREQ];
|
||||
static int usageCount = 0;
|
||||
|
||||
/*****************************************************************************************
|
||||
@@ -55,258 +55,274 @@ static int usageCount = 0;
|
||||
*****************************************************************************************/
|
||||
void cpuMonitor()
|
||||
{
|
||||
ServerMonitor serverMonitor;
|
||||
ServerMonitor serverMonitor;
|
||||
|
||||
// register for Heartbeat monitoring
|
||||
/* try {
|
||||
ProcHeartbeat procheartbeat;
|
||||
procheartbeat.registerHeartbeat(CPU_HEARTBEAT_ID);
|
||||
}
|
||||
catch (exception& ex)
|
||||
{
|
||||
string error = ex.what();
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on registerHeartbeat: ");
|
||||
args.add(error);
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
catch(...)
|
||||
{
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on sendHeartbeat: Caught unknown exception!");
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
*/
|
||||
int periodCount = 5;
|
||||
float cpuPeriod[periodCount];
|
||||
int periodCounter = 0;
|
||||
float averageCpuUsage = 0;
|
||||
currentCpuUsage = 0;
|
||||
// register for Heartbeat monitoring
|
||||
/* try {
|
||||
ProcHeartbeat procheartbeat;
|
||||
procheartbeat.registerHeartbeat(CPU_HEARTBEAT_ID);
|
||||
}
|
||||
catch (exception& ex)
|
||||
{
|
||||
string error = ex.what();
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on registerHeartbeat: ");
|
||||
args.add(error);
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
catch(...)
|
||||
{
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on sendHeartbeat: Caught unknown exception!");
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
*/
|
||||
int periodCount = 5;
|
||||
float cpuPeriod[periodCount];
|
||||
int periodCounter = 0;
|
||||
float averageCpuUsage = 0;
|
||||
currentCpuUsage = 0;
|
||||
|
||||
// set defaults
|
||||
unsigned int cpuCritical = 0,
|
||||
cpuMajor = 0,
|
||||
cpuMinor = 0,
|
||||
cpuMinorClear = 0;
|
||||
// set defaults
|
||||
unsigned int cpuCritical = 0,
|
||||
cpuMajor = 0,
|
||||
cpuMinor = 0,
|
||||
cpuMinorClear = 0;
|
||||
|
||||
// initial cpu Period table
|
||||
for (int i =0;i < periodCount; i++)
|
||||
{
|
||||
cpuPeriod[i] = 0;
|
||||
}
|
||||
// initial cpu Period table
|
||||
for (int i = 0; i < periodCount; i++)
|
||||
{
|
||||
cpuPeriod[i] = 0;
|
||||
}
|
||||
|
||||
while(true)
|
||||
{
|
||||
// Get CPU usage water mark from server configuration and compare
|
||||
ModuleTypeConfig moduleTypeConfig;
|
||||
Oam oam;
|
||||
try {
|
||||
oam.getSystemConfig(moduleTypeConfig);
|
||||
cpuCritical = moduleTypeConfig.ModuleCPUCriticalThreshold;
|
||||
cpuMajor = moduleTypeConfig.ModuleCPUMajorThreshold;
|
||||
cpuMinor = moduleTypeConfig.ModuleCPUMinorThreshold;
|
||||
cpuMinorClear = moduleTypeConfig.ModuleCPUMinorClearThreshold;
|
||||
} catch (...)
|
||||
{
|
||||
sleep(5);
|
||||
continue;
|
||||
}
|
||||
while (true)
|
||||
{
|
||||
// Get CPU usage water mark from server configuration and compare
|
||||
ModuleTypeConfig moduleTypeConfig;
|
||||
Oam oam;
|
||||
|
||||
if (RESOURCE_DEBUG)
|
||||
cout << "critical water: " << moduleTypeConfig.ModuleCPUCriticalThreshold << endl;
|
||||
try
|
||||
{
|
||||
oam.getSystemConfig(moduleTypeConfig);
|
||||
cpuCritical = moduleTypeConfig.ModuleCPUCriticalThreshold;
|
||||
cpuMajor = moduleTypeConfig.ModuleCPUMajorThreshold;
|
||||
cpuMinor = moduleTypeConfig.ModuleCPUMinorThreshold;
|
||||
cpuMinorClear = moduleTypeConfig.ModuleCPUMinorClearThreshold;
|
||||
}
|
||||
catch (...)
|
||||
{
|
||||
sleep(5);
|
||||
continue;
|
||||
}
|
||||
|
||||
pthread_mutex_lock(&CPU_LOCK);
|
||||
//
|
||||
// get Process and System CPU usage
|
||||
//
|
||||
serverMonitor.getCPUdata();
|
||||
if (RESOURCE_DEBUG)
|
||||
cout << "critical water: " << moduleTypeConfig.ModuleCPUCriticalThreshold << endl;
|
||||
|
||||
// store and get average
|
||||
cpuPeriod[periodCounter] = currentCpuUsage;
|
||||
averageCpuUsage = 0;
|
||||
for (int i =0;i < periodCount; i++)
|
||||
{
|
||||
averageCpuUsage += cpuPeriod[i];
|
||||
}
|
||||
averageCpuUsage = averageCpuUsage / periodCount;
|
||||
pthread_mutex_lock(&CPU_LOCK);
|
||||
//
|
||||
// get Process and System CPU usage
|
||||
//
|
||||
serverMonitor.getCPUdata();
|
||||
|
||||
// store and get average
|
||||
cpuPeriod[periodCounter] = currentCpuUsage;
|
||||
averageCpuUsage = 0;
|
||||
|
||||
for (int i = 0; i < periodCount; i++)
|
||||
{
|
||||
averageCpuUsage += cpuPeriod[i];
|
||||
}
|
||||
|
||||
averageCpuUsage = averageCpuUsage / periodCount;
|
||||
|
||||
// serverMonitor.logCPUactive(averageCpuUsage);
|
||||
if (CPU_DEBUG) {
|
||||
cout << "Current CPU Usage: " << currentCpuUsage << endl;
|
||||
cout << "Average CPU Usage: " << averageCpuUsage << endl;
|
||||
}
|
||||
if (CPU_DEBUG)
|
||||
{
|
||||
cout << "Current CPU Usage: " << currentCpuUsage << endl;
|
||||
cout << "Average CPU Usage: " << averageCpuUsage << endl;
|
||||
}
|
||||
|
||||
if (averageCpuUsage >= cpuCritical && cpuCritical > 0 ) {
|
||||
serverMonitor.sendResourceAlarm("CPU", CPU_USAGE_HIGH, SET, (int) averageCpuUsage);
|
||||
}
|
||||
else if (averageCpuUsage >= cpuMajor && cpuMajor > 0 )
|
||||
serverMonitor.sendResourceAlarm("CPU", CPU_USAGE_MED, SET, (int) averageCpuUsage);
|
||||
else if (averageCpuUsage >= cpuMinor && cpuMinor > 0 )
|
||||
serverMonitor.sendResourceAlarm("CPU", CPU_USAGE_LOW, SET, (int) averageCpuUsage);
|
||||
else if (averageCpuUsage >= cpuMinorClear && cpuMinorClear > 0 ) {
|
||||
serverMonitor.checkCPUAlarm("CPU", CPU_USAGE_LOW);
|
||||
//Log this event
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Current CPU usage = ");
|
||||
args.add((int) currentCpuUsage);
|
||||
args.add(", Average CPU usage = ");
|
||||
args.add((int) averageCpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
}
|
||||
else
|
||||
serverMonitor.checkCPUAlarm("CPU");
|
||||
if (averageCpuUsage >= cpuCritical && cpuCritical > 0 )
|
||||
{
|
||||
serverMonitor.sendResourceAlarm("CPU", CPU_USAGE_HIGH, SET, (int) averageCpuUsage);
|
||||
}
|
||||
else if (averageCpuUsage >= cpuMajor && cpuMajor > 0 )
|
||||
serverMonitor.sendResourceAlarm("CPU", CPU_USAGE_MED, SET, (int) averageCpuUsage);
|
||||
else if (averageCpuUsage >= cpuMinor && cpuMinor > 0 )
|
||||
serverMonitor.sendResourceAlarm("CPU", CPU_USAGE_LOW, SET, (int) averageCpuUsage);
|
||||
else if (averageCpuUsage >= cpuMinorClear && cpuMinorClear > 0 )
|
||||
{
|
||||
serverMonitor.checkCPUAlarm("CPU", CPU_USAGE_LOW);
|
||||
//Log this event
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Current CPU usage = ");
|
||||
args.add((int) currentCpuUsage);
|
||||
args.add(", Average CPU usage = ");
|
||||
args.add((int) averageCpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
}
|
||||
else
|
||||
serverMonitor.checkCPUAlarm("CPU");
|
||||
|
||||
//
|
||||
// check CPU usage by process
|
||||
//
|
||||
ProcessCPUList::iterator p = pcl.begin();
|
||||
while(p != pcl.end())
|
||||
{
|
||||
string processName = (*p).processName;
|
||||
double cpuUsage = (*p).usedPercent;
|
||||
p++;
|
||||
//
|
||||
// check CPU usage by process
|
||||
//
|
||||
ProcessCPUList::iterator p = pcl.begin();
|
||||
|
||||
if (CPU_DEBUG) {
|
||||
cout << "Process Name : " << processName << endl;
|
||||
cout << "CPU Usage: " << cpuUsage << endl;
|
||||
}
|
||||
while (p != pcl.end())
|
||||
{
|
||||
string processName = (*p).processName;
|
||||
double cpuUsage = (*p).usedPercent;
|
||||
p++;
|
||||
|
||||
// check if a Calpont Process, if so alarm is over thresholds
|
||||
// if not, just log if over thresholds
|
||||
if (cpuUsage >= cpuCritical && cpuCritical > 0) {
|
||||
/* try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
if (CPU_DEBUG)
|
||||
{
|
||||
cout << "Process Name : " << processName << endl;
|
||||
cout << "CPU Usage: " << cpuUsage << endl;
|
||||
}
|
||||
|
||||
serverMonitor.sendResourceAlarm(processName, CPU_USAGE_HIGH, SET, (int) cpuUsage);
|
||||
}
|
||||
catch (...) {
|
||||
*/ LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Process");
|
||||
args.add(processName);
|
||||
args.add(" above Critical CPU threshold with a percentage of ");
|
||||
args.add((int) cpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
// check if a Calpont Process, if so alarm is over thresholds
|
||||
// if not, just log if over thresholds
|
||||
if (cpuUsage >= cpuCritical && cpuCritical > 0)
|
||||
{
|
||||
/* try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
|
||||
serverMonitor.sendResourceAlarm(processName, CPU_USAGE_HIGH, SET, (int) cpuUsage);
|
||||
}
|
||||
catch (...) {
|
||||
*/ LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Process");
|
||||
args.add(processName);
|
||||
args.add(" above Critical CPU threshold with a percentage of ");
|
||||
args.add((int) cpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
// }
|
||||
}
|
||||
else if (cpuUsage >= cpuMajor && cpuMajor > 0) {
|
||||
/* try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
}
|
||||
else if (cpuUsage >= cpuMajor && cpuMajor > 0)
|
||||
{
|
||||
/* try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
|
||||
serverMonitor.sendResourceAlarm(processName, CPU_USAGE_MED, SET, (int) cpuUsage);
|
||||
}
|
||||
catch (...) {
|
||||
*/ LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Process");
|
||||
args.add(processName);
|
||||
args.add(" above Major CPU threshold with a percentage of ");
|
||||
args.add((int) cpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
serverMonitor.sendResourceAlarm(processName, CPU_USAGE_MED, SET, (int) cpuUsage);
|
||||
}
|
||||
catch (...) {
|
||||
*/ LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Process");
|
||||
args.add(processName);
|
||||
args.add(" above Major CPU threshold with a percentage of ");
|
||||
args.add((int) cpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
// }
|
||||
}
|
||||
else if (cpuUsage >= cpuMinor && cpuMinor > 0) {
|
||||
/* try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
}
|
||||
else if (cpuUsage >= cpuMinor && cpuMinor > 0)
|
||||
{
|
||||
/* try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
|
||||
serverMonitor.sendResourceAlarm(processName, CPU_USAGE_LOW, SET, (int) cpuUsage);
|
||||
}
|
||||
catch (...) {
|
||||
*/ LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Process");
|
||||
args.add(processName);
|
||||
args.add(" above Minor CPU threshold with a percentage of ");
|
||||
args.add((int) cpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
serverMonitor.sendResourceAlarm(processName, CPU_USAGE_LOW, SET, (int) cpuUsage);
|
||||
}
|
||||
catch (...) {
|
||||
*/ LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Process");
|
||||
args.add(processName);
|
||||
args.add(" above Minor CPU threshold with a percentage of ");
|
||||
args.add((int) cpuUsage);
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
// }
|
||||
}
|
||||
/* else if (cpuUsage >= cpuMinorClear) {
|
||||
try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
}
|
||||
|
||||
serverMonitor.checkCPUAlarm(processName, CPU_USAGE_LOW);
|
||||
}
|
||||
catch (...) {}
|
||||
}
|
||||
else
|
||||
serverMonitor.checkCPUAlarm(processName);
|
||||
*/ }
|
||||
/* else if (cpuUsage >= cpuMinorClear) {
|
||||
try {
|
||||
t = oam.getMyProcessStatus(processID);
|
||||
processName = boost::get<1>(t);
|
||||
|
||||
// send heartbeat message
|
||||
/* try {
|
||||
ProcHeartbeat procheartbeat;
|
||||
procheartbeat.sendHeartbeat(CPU_HEARTBEAT_ID);
|
||||
serverMonitor.checkCPUAlarm(processName, CPU_USAGE_LOW);
|
||||
}
|
||||
catch (...) {}
|
||||
}
|
||||
else
|
||||
serverMonitor.checkCPUAlarm(processName);
|
||||
*/
|
||||
}
|
||||
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Sent Heartbeat Msg");
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
}
|
||||
catch (exception& ex)
|
||||
{
|
||||
string error = ex.what();
|
||||
if ( error.find("Disabled") == string::npos ) {
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on sendHeartbeat: ");
|
||||
args.add(error);
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
}
|
||||
catch(...)
|
||||
{
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on sendHeartbeat: Caught unknown exception!");
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
*/
|
||||
// send heartbeat message
|
||||
/* try {
|
||||
ProcHeartbeat procheartbeat;
|
||||
procheartbeat.sendHeartbeat(CPU_HEARTBEAT_ID);
|
||||
|
||||
pthread_mutex_unlock(&CPU_LOCK);
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("Sent Heartbeat Msg");
|
||||
msg.format(args);
|
||||
ml.logInfoMessage(msg);
|
||||
}
|
||||
catch (exception& ex)
|
||||
{
|
||||
string error = ex.what();
|
||||
if ( error.find("Disabled") == string::npos ) {
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on sendHeartbeat: ");
|
||||
args.add(error);
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
}
|
||||
catch(...)
|
||||
{
|
||||
LoggingID lid(SERVER_MONITOR_LOG_ID);
|
||||
MessageLog ml(lid);
|
||||
Message msg;
|
||||
Message::Args args;
|
||||
args.add("EXCEPTION ERROR on sendHeartbeat: Caught unknown exception!");
|
||||
msg.format(args);
|
||||
ml.logErrorMessage(msg);
|
||||
}
|
||||
*/
|
||||
|
||||
// sleep, 10 minutes
|
||||
sleep(MONITOR_PERIOD*10);
|
||||
pthread_mutex_unlock(&CPU_LOCK);
|
||||
|
||||
++periodCounter;
|
||||
if ( periodCounter >= periodCount )
|
||||
periodCounter = 0;
|
||||
// sleep, 10 minutes
|
||||
sleep(MONITOR_PERIOD * 10);
|
||||
|
||||
} // end of while loop
|
||||
++periodCounter;
|
||||
|
||||
if ( periodCounter >= periodCount )
|
||||
periodCounter = 0;
|
||||
|
||||
} // end of while loop
|
||||
}
|
||||
|
||||
/******************************************************************************************
|
||||
@@ -317,126 +333,146 @@ void cpuMonitor()
|
||||
******************************************************************************************/
|
||||
void ServerMonitor::checkCPUAlarm(string alarmItem, ALARMS alarmID)
|
||||
{
|
||||
Oam oam;
|
||||
ServerMonitor serverMonitor;
|
||||
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";
|
||||
}
|
||||
// get current server name
|
||||
string serverName;
|
||||
oamModuleInfo_t st;
|
||||
|
||||
switch (alarmID) {
|
||||
case ALARM_NONE: // clear all alarms set if any found
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_HIGH, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_HIGH);
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_MED, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_MED);
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_LOW, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_LOW);
|
||||
break;
|
||||
case CPU_USAGE_LOW: // clear high and medium alarms set if any found
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_HIGH, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_HIGH);
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_MED, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_MED);
|
||||
break;
|
||||
case CPU_USAGE_MED: // clear high alarms set if any found
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_HIGH, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_HIGH);
|
||||
break;
|
||||
default: // none to clear
|
||||
break;
|
||||
} // end of switch
|
||||
return;
|
||||
try
|
||||
{
|
||||
st = oam.getModuleInfo();
|
||||
serverName = boost::get<0>(st);
|
||||
}
|
||||
catch (...)
|
||||
{
|
||||
serverName = "Unknown Server";
|
||||
}
|
||||
|
||||
switch (alarmID)
|
||||
{
|
||||
case ALARM_NONE: // clear all alarms set if any found
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_HIGH, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_HIGH);
|
||||
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_MED, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_MED);
|
||||
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_LOW, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_LOW);
|
||||
|
||||
break;
|
||||
|
||||
case CPU_USAGE_LOW: // clear high and medium alarms set if any found
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_HIGH, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_HIGH);
|
||||
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_MED, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_MED);
|
||||
|
||||
break;
|
||||
|
||||
case CPU_USAGE_MED: // clear high alarms set if any found
|
||||
if ( serverMonitor.checkActiveAlarm(CPU_USAGE_HIGH, serverName, alarmItem) )
|
||||
// alarm set, clear it
|
||||
clearAlarm(alarmItem, CPU_USAGE_HIGH);
|
||||
|
||||
break;
|
||||
|
||||
default: // none to clear
|
||||
break;
|
||||
} // end of switch
|
||||
|
||||
return;
|
||||
}
|
||||
|
||||
/*****************************************************************************************
|
||||
* @brief logCPUactive
|
||||
*
|
||||
* purpose: Log Peak and Average CPU usage
|
||||
* purpose: Log Peak and Average CPU usage
|
||||
*
|
||||
*****************************************************************************************/
|
||||
void ServerMonitor::logCPUactive (unsigned int cpuUsage)
|
||||
{
|
||||
ServerMonitor serverMonitor;
|
||||
ServerMonitor serverMonitor;
|
||||
|
||||
// determin the active log file name
|
||||
string usageLogFileName = FE_MOUNT_DIR;
|
||||
usageLogFileName = usageLogFileName + "cpu.log";
|
||||
|
||||
if (RESOURCE_DEBUG)
|
||||
cout << usageLogFileName << endl;
|
||||
|
||||
fstream usageLogFile;
|
||||
usageLogFile.open (usageLogFileName.c_str(), ios::in|ios::out);
|
||||
// determin the active log file name
|
||||
string usageLogFileName = FE_MOUNT_DIR;
|
||||
usageLogFileName = usageLogFileName + "cpu.log";
|
||||
|
||||
if (usageLogFile.fail())
|
||||
{
|
||||
ofstream file (usageLogFileName.c_str());
|
||||
file.close();
|
||||
usageLogFile.open(usageLogFileName.c_str(), ios::in|ios::out);
|
||||
if (!usageLogFile) cout << "--error" << endl;
|
||||
}
|
||||
|
||||
// get the counter
|
||||
usageLogFile.seekg(0, ios::beg);
|
||||
usageLogFile.read (reinterpret_cast<char *>(&usageCount), sizeof (int));
|
||||
if (usageLogFile.eof()) usageLogFile.clear();
|
||||
if (RESOURCE_DEBUG)
|
||||
cout << usageLogFileName << endl;
|
||||
|
||||
// new iteration
|
||||
if (usageCount == 0)
|
||||
{
|
||||
usageLogFile.seekp(0, ios::beg);
|
||||
usageLogFile.write (reinterpret_cast<char *>(&usageCount), sizeof (int));
|
||||
}
|
||||
usageCount ++;
|
||||
|
||||
// append new usage data to the end
|
||||
usageLogFile.seekp (0, ios::end);
|
||||
usageLogFile.write (reinterpret_cast<char *>(&cpuUsage), sizeof (int));
|
||||
|
||||
if (RESOURCE_DEBUG)
|
||||
cout << "usage: " << usageCount << endl;
|
||||
|
||||
// calculate peak and average if it's time to log usage data
|
||||
if (usageCount >= LOG_FREQ / MONITOR_FREQ)
|
||||
{
|
||||
usageLogFile.seekg (4, ios::beg);
|
||||
usageLogFile.read ((char*)usage, sizeof(unsigned int) * LOG_FREQ/MONITOR_FREQ);
|
||||
if (usageLogFile.eof()) usageLogFile.clear();
|
||||
if (RESOURCE_DEBUG)
|
||||
{
|
||||
for (int i = 0; i < usageCount; i++)
|
||||
{
|
||||
cout << usage [i] << endl;
|
||||
}
|
||||
}
|
||||
serverMonitor.logCPUstat(usageCount);
|
||||
|
||||
// delete the file
|
||||
usageLogFile.close();
|
||||
unlink (usageLogFileName.c_str());
|
||||
}
|
||||
|
||||
// else, update usageCount
|
||||
else
|
||||
{
|
||||
usageLogFile.seekp(0, ios::beg);
|
||||
usageLogFile.write (reinterpret_cast<char *>(&usageCount), sizeof (int));
|
||||
usageLogFile.close();
|
||||
}
|
||||
fstream usageLogFile;
|
||||
usageLogFile.open (usageLogFileName.c_str(), ios::in | ios::out);
|
||||
|
||||
if (usageLogFile.fail())
|
||||
{
|
||||
ofstream file (usageLogFileName.c_str());
|
||||
file.close();
|
||||
usageLogFile.open(usageLogFileName.c_str(), ios::in | ios::out);
|
||||
|
||||
if (!usageLogFile) cout << "--error" << endl;
|
||||
}
|
||||
|
||||
// get the counter
|
||||
usageLogFile.seekg(0, ios::beg);
|
||||
usageLogFile.read (reinterpret_cast<char*>(&usageCount), sizeof (int));
|
||||
|
||||
if (usageLogFile.eof()) usageLogFile.clear();
|
||||
|
||||
// new iteration
|
||||
if (usageCount == 0)
|
||||
{
|
||||
usageLogFile.seekp(0, ios::beg);
|
||||
usageLogFile.write (reinterpret_cast<char*>(&usageCount), sizeof (int));
|
||||
}
|
||||
|
||||
usageCount ++;
|
||||
|
||||
// append new usage data to the end
|
||||
usageLogFile.seekp (0, ios::end);
|
||||
usageLogFile.write (reinterpret_cast<char*>(&cpuUsage), sizeof (int));
|
||||
|
||||
if (RESOURCE_DEBUG)
|
||||
cout << "usage: " << usageCount << endl;
|
||||
|
||||
// calculate peak and average if it's time to log usage data
|
||||
if (usageCount >= LOG_FREQ / MONITOR_FREQ)
|
||||
{
|
||||
usageLogFile.seekg (4, ios::beg);
|
||||
usageLogFile.read ((char*)usage, sizeof(unsigned int) * LOG_FREQ / MONITOR_FREQ);
|
||||
|
||||
if (usageLogFile.eof()) usageLogFile.clear();
|
||||
|
||||
if (RESOURCE_DEBUG)
|
||||
{
|
||||
for (int i = 0; i < usageCount; i++)
|
||||
{
|
||||
cout << usage [i] << endl;
|
||||
}
|
||||
}
|
||||
|
||||
serverMonitor.logCPUstat(usageCount);
|
||||
|
||||
// delete the file
|
||||
usageLogFile.close();
|
||||
unlink (usageLogFileName.c_str());
|
||||
}
|
||||
|
||||
// else, update usageCount
|
||||
else
|
||||
{
|
||||
usageLogFile.seekp(0, ios::beg);
|
||||
usageLogFile.write (reinterpret_cast<char*>(&usageCount), sizeof (int));
|
||||
usageLogFile.close();
|
||||
}
|
||||
}
|
||||
|
||||
/*****************************************************************************************
|
||||
@@ -447,27 +483,30 @@ void ServerMonitor::logCPUactive (unsigned int cpuUsage)
|
||||
*****************************************************************************************/
|
||||
void ServerMonitor::logCPUstat (int usageCount)
|
||||
{
|
||||
unsigned int max = 0;
|
||||
unsigned int sum = 0;
|
||||
float average = 0.0;
|
||||
for (int i = 0; i < usageCount; i++)
|
||||
{
|
||||
if (usage[i] > max)
|
||||
max = usage[i];
|
||||
sum += usage[i];
|
||||
}
|
||||
if ( usageCount == 0 )
|
||||
average=0;
|
||||
else
|
||||
average = sum / usageCount;
|
||||
|
||||
// Call system log api to store stats.
|
||||
// for now, write on local for testing purpose.
|
||||
string statFileName = FE_MOUNT_DIR;
|
||||
statFileName = statFileName + "cpustat.log";
|
||||
ofstream file (statFileName.c_str(), ios::app);
|
||||
file << max << " " << average << endl;
|
||||
file.close();
|
||||
unsigned int max = 0;
|
||||
unsigned int sum = 0;
|
||||
float average = 0.0;
|
||||
|
||||
for (int i = 0; i < usageCount; i++)
|
||||
{
|
||||
if (usage[i] > max)
|
||||
max = usage[i];
|
||||
|
||||
sum += usage[i];
|
||||
}
|
||||
|
||||
if ( usageCount == 0 )
|
||||
average = 0;
|
||||
else
|
||||
average = sum / usageCount;
|
||||
|
||||
// Call system log api to store stats.
|
||||
// for now, write on local for testing purpose.
|
||||
string statFileName = FE_MOUNT_DIR;
|
||||
statFileName = statFileName + "cpustat.log";
|
||||
ofstream file (statFileName.c_str(), ios::app);
|
||||
file << max << " " << average << endl;
|
||||
file.close();
|
||||
}
|
||||
|
||||
/*****************************************************************************************
|
||||
@@ -478,54 +517,62 @@ void ServerMonitor::logCPUstat (int usageCount)
|
||||
*****************************************************************************************/
|
||||
void ServerMonitor::getCPUdata()
|
||||
{
|
||||
pcl.clear();
|
||||
pcl.clear();
|
||||
|
||||
system("top -b -n1 | head -12 | awk '{print $9,$12}' | tail -5 > /tmp/columnstore_tmp_files/processCpu");
|
||||
system("top -b -n1 | head -12 | awk '{print $9,$12}' | tail -5 > /tmp/columnstore_tmp_files/processCpu");
|
||||
|
||||
ifstream oldFile1 ("/tmp/columnstore_tmp_files/processCpu");
|
||||
ifstream oldFile1 ("/tmp/columnstore_tmp_files/processCpu");
|
||||
|
||||
// read top 5 users
|
||||
int i = 0;
|
||||
char line[400];
|
||||
while (oldFile1.getline(line, 400))
|
||||
{
|
||||
string buf = line;
|
||||
string::size_type pos = buf.find (' ',0);
|
||||
if (pos != string::npos) {
|
||||
processCPU pc;
|
||||
pc.processName = buf.substr(pos+1,80);
|
||||
pc.usedPercent = atol(buf.substr(0,pos).c_str());
|
||||
pcl.push_back(pc);
|
||||
i++;
|
||||
}
|
||||
}
|
||||
oldFile1.close();
|
||||
// read top 5 users
|
||||
int i = 0;
|
||||
char line[400];
|
||||
|
||||
//
|
||||
// get and check Total CPU usage
|
||||
//
|
||||
system("top -b -n 6 -d 1 | awk '{print $5}' | grep %id > /tmp/columnstore_tmp_files/systemCpu");
|
||||
while (oldFile1.getline(line, 400))
|
||||
{
|
||||
string buf = line;
|
||||
string::size_type pos = buf.find (' ', 0);
|
||||
|
||||
ifstream oldFile ("/tmp/columnstore_tmp_files/systemCpu");
|
||||
if (pos != string::npos)
|
||||
{
|
||||
processCPU pc;
|
||||
pc.processName = buf.substr(pos + 1, 80);
|
||||
pc.usedPercent = atol(buf.substr(0, pos).c_str());
|
||||
pcl.push_back(pc);
|
||||
i++;
|
||||
}
|
||||
}
|
||||
|
||||
float systemIdle = 0;
|
||||
// skip first line in file, and average the next 5 entries which contains idle times
|
||||
oldFile.getline(line, 400);
|
||||
int count = 0;
|
||||
while (oldFile.getline(line, 400))
|
||||
{
|
||||
string buf = line;
|
||||
string::size_type pos = buf.find ('%',0);
|
||||
if (pos != string::npos) {
|
||||
systemIdle = systemIdle + atol(buf.substr(0,pos-1).c_str());
|
||||
count++;
|
||||
}
|
||||
}
|
||||
oldFile.close();
|
||||
oldFile1.close();
|
||||
|
||||
if ( count == 0 )
|
||||
currentCpuUsage = 0;
|
||||
else
|
||||
currentCpuUsage = 100 - (systemIdle / count);
|
||||
//
|
||||
// get and check Total CPU usage
|
||||
//
|
||||
system("top -b -n 6 -d 1 | awk '{print $5}' | grep %id > /tmp/columnstore_tmp_files/systemCpu");
|
||||
|
||||
ifstream oldFile ("/tmp/columnstore_tmp_files/systemCpu");
|
||||
|
||||
float systemIdle = 0;
|
||||
// skip first line in file, and average the next 5 entries which contains idle times
|
||||
oldFile.getline(line, 400);
|
||||
int count = 0;
|
||||
|
||||
while (oldFile.getline(line, 400))
|
||||
{
|
||||
string buf = line;
|
||||
string::size_type pos = buf.find ('%', 0);
|
||||
|
||||
if (pos != string::npos)
|
||||
{
|
||||
systemIdle = systemIdle + atol(buf.substr(0, pos - 1).c_str());
|
||||
count++;
|
||||
}
|
||||
}
|
||||
|
||||
oldFile.close();
|
||||
|
||||
if ( count == 0 )
|
||||
currentCpuUsage = 0;
|
||||
else
|
||||
currentCpuUsage = 100 - (systemIdle / count);
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user