Gaudi Framework, version v20r2
	Generated: 18 Jul 2008

ApMon.h File Reference

---

Detailed Description

Declarations for the ApMon class.

The ApMon class can be used for sending monitoring data to one or more destination hosts that run MonALISA.

Definition in file ApMon.h.

#include <sys/types.h>
#include <sys/stat.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdexcept>
#include <ctype.h>
#include <time.h>
#include <sys/socket.h>
#include <sys/ioctl.h>
#include <sys/utsname.h>
#include <sys/time.h>
#include <arpa/inet.h>
#include <netinet/in.h>
#include <net/if.h>
#include <rpc/rpc.h>
#include <netdb.h>
#include <unistd.h>
#include <pthread.h>
#include <pwd.h>
#include <grp.h>
#include <linux/param.h>

Include dependency graph for ApMon.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	ConfURLs
	Data structure which holds the configuration URLs. More...
struct	MonitoredJob
	Data structure which holds information about a job monitored by ApMon. More...
class	ApMon
	Data structure used for sending monitoring data to a MonaLisa module. More...
Defines
#define	XDR_STRING   0
	Used to code the string data type.
#define	XDR_INT32   2
	Used to code the 4 bytes integer data type.
#define	XDR_REAL32   4
	Used to code the 4 bytes real data type.
#define	XDR_REAL64   5
	Used to code the 8 bytes real data type.
#define	MAX_DGRAM_SIZE   8192
	Maximum UDP datagram size.
#define	MAX_STRING_LEN   512
	Maximum string length (for hostnames).
#define	MAX_STRING_LEN1   (MAX_STRING_LEN + 1)
#define	RET_SUCCESS   0
	Function return value (succes).
#define	RET_ERROR   -1
	Function return value (error).
#define	PROCUTILS_ERROR   -2
#define	RET_NOT_SENT   -3
	A datagram was not sent because the number of datagrams that can be sent per second is limited.
#define	MAX_N_DESTINATIONS   30
	Maximum number of destinations hosts to which we send the parameters.
#define	DEFAULT_PORT   8884
	The default port on which MonALISa listens.
#define	MAX_HEADER_LENGTH   40
	Maximum header length.
#define	FILE_INIT   1
	Indicates that the object was initialized from a file.
#define	LIST_INIT   2
	Indicates that the object was initialized from a list.
#define	DIRECT_INIT   3
	Indicates that the object was initialized directly.
#define	JOB_MONITOR_INTERVAL   20
	Time interval (in sec) at which job monitoring datagrams are sent.
#define	SYS_MONITOR_INTERVAL   20
	Time interval (in sec) at which system monitoring datagams are sent.
#define	RECHECK_INTERVAL   600
	Time interval (in sec) at which the configuration files are checked for changes.
#define	GEN_MONITOR_INTERVALS   10
	The number of time intervals at which ApMon sends general system monitoring information (considering the time intervals at which ApMon sends system monitoring information).
#define	MAX_MONITORED_JOBS   30
	The maximum number of jobs that can be monitored.
#define	MAX_SYS_PARAMS   30
	The maximum number of system parameters.
#define	MAX_GEN_PARAMS   30
	The maximum number of general system parameters.
#define	MAX_JOB_PARAMS   30
	The maximum number of job parameters.
#define	MAX_MSG_RATE   20
	The maxim number of mesages per second that will be sent to MonALISA.
#define	NLETTERS   26
#define	TWO_BILLION   2000000000
#define	APMON_VERSION   "2.2.1"
Functions
void *	bkTask (void *param)
	This function is executed in a background thread and has two roles: it automatically sends the system/job monitoring parameters (if the user requested) and it checks the configuration file/URLs for changes.

---

Define Documentation

#define APMON_VERSION   "2.2.1"

Definition at line 155 of file ApMon.h.

Referenced by ApMon::sendTimedParameters().

#define DEFAULT_PORT   8884

The default port on which MonALISa listens.

Definition at line 114 of file ApMon.h.

Referenced by ApMon::addToDestinations().

#define DIRECT_INIT   3

Indicates that the object was initialized directly.

Definition at line 126 of file ApMon.h.

Referenced by ApMon::setConfRecheck().

#define FILE_INIT   1

Indicates that the object was initialized from a file.

Definition at line 118 of file ApMon.h.

Referenced by ApMon::ApMon(), and bkTask().

#define GEN_MONITOR_INTERVALS   10

The number of time intervals at which ApMon sends general system monitoring information (considering the time intervals at which ApMon sends system monitoring information).

Definition at line 138 of file ApMon.h.

Referenced by ApMon::setGenMonitoring().

#define JOB_MONITOR_INTERVAL   20

Time interval (in sec) at which job monitoring datagrams are sent.

Definition at line 128 of file ApMon.h.

Referenced by ApMon::initMonitoring(), and ApMon::setJobMonitoring().

#define LIST_INIT   2

Indicates that the object was initialized from a list.

Definition at line 123 of file ApMon.h.

Referenced by ApMon::constructFromList().

#define MAX_DGRAM_SIZE   8192

Maximum UDP datagram size.

Definition at line 102 of file ApMon.h.

Referenced by ApMon::arrayInit(), ApMon::encodeParams(), and ApMon::sendTimedParameters().

#define MAX_GEN_PARAMS   30

The maximum number of general system parameters.

Definition at line 144 of file ApMon.h.

#define MAX_HEADER_LENGTH   40

Maximum header length.

Definition at line 115 of file ApMon.h.

Referenced by ApMon::encodeParams(), and ApMon::sendTimedParameters().

#define MAX_JOB_PARAMS   30

The maximum number of job parameters.

Definition at line 146 of file ApMon.h.

#define MAX_MONITORED_JOBS   30

The maximum number of jobs that can be monitored.

Definition at line 140 of file ApMon.h.

Referenced by ApMon::addJobToMonitor(), and ApMon::arrayInit().

#define MAX_MSG_RATE   20

The maxim number of mesages per second that will be sent to MonALISA.

Definition at line 149 of file ApMon.h.

Referenced by ApMon::initMonitoring().

#define MAX_N_DESTINATIONS   30

Maximum number of destinations hosts to which we send the parameters.

Definition at line 111 of file ApMon.h.

Referenced by ApMon::initialize(), and ApMon::parseConf().

#define MAX_STRING_LEN   512

Maximum string length (for hostnames).

Definition at line 103 of file ApMon.h.

Referenced by ApMon::arrayInit(), ProcUtils::countOpenFiles(), ProcUtils::getBootTime(), apmon_mon_utils::getChildren(), ProcUtils::getCPUInfo(), ProcUtils::getCPUUsage(), ApMon::getDestFromWeb(), ProcUtils::getMemUsed(), ProcUtils::getNetInfo(), ProcUtils::getNetworkInterfaces(), ProcUtils::getNumCPUs(), ProcUtils::getSwapPages(), ProcUtils::getSysMem(), apmon_utils::httpRequest(), apmon_utils::isPrivateAddress(), apmon_utils::parse_URL(), ApMon::parseConf(), ApMon::parseXApMonLine(), apmon_mon_utils::readJobInfo(), and apmon_utils::urlModified().

#define MAX_STRING_LEN1   (MAX_STRING_LEN + 1)

Definition at line 104 of file ApMon.h.

Referenced by ApMon::parseConf(), apmon_mon_utils::readJobInfo(), and apmon_utils::urlModified().

#define MAX_SYS_PARAMS   30

The maximum number of system parameters.

Definition at line 142 of file ApMon.h.

#define NLETTERS   26

Definition at line 151 of file ApMon.h.

Referenced by ProcUtils::getProcesses().

#define PROCUTILS_ERROR   -2

Definition at line 107 of file ApMon.h.

Referenced by apmon_mon_utils::readJobInfo(), ApMon::sendGeneralInfo(), ApMon::sendSysInfo(), ApMon::updateGeneralInfo(), and ApMon::updateSysInfo().

#define RECHECK_INTERVAL   600

Time interval (in sec) at which the configuration files are checked for changes.

Definition at line 133 of file ApMon.h.

Referenced by bkTask(), ApMon::initMonitoring(), and ApMon::setConfRecheck().

#define RET_ERROR   -1

Function return value (error).

Definition at line 106 of file ApMon.h.

Referenced by apmon_mon_utils::getChildren(), ProcUtils::getNetstatInfo(), ProcUtils::getProcesses(), apmon_mon_utils::parsePSTime(), apmon_mon_utils::readJobDiskUsage(), apmon_mon_utils::readJobInfo(), ApMon::sendGeneralInfo(), ApMon::sendOneJobInfo(), ApMon::sendSysInfo(), ApMon::sendTimedParameters(), apmon_utils::sizeEval(), ApMon::updateJobInfo(), ApMon::updateSysInfo(), and apmon_utils::xdrSize().

#define RET_NOT_SENT   -3

A datagram was not sent because the number of datagrams that can be sent per second is limited.

Definition at line 108 of file ApMon.h.

Referenced by ApMon::sendTimedParameters().

#define RET_SUCCESS   0

Function return value (succes).

Definition at line 105 of file ApMon.h.

Referenced by ApMon::initMonitoring(), ApMon::sendOneJobInfo(), ApMon::sendSysInfo(), and ApMon::sendTimedParameters().

#define SYS_MONITOR_INTERVAL   20

Time interval (in sec) at which system monitoring datagams are sent.

Definition at line 130 of file ApMon.h.

Referenced by ApMon::initMonitoring(), and ApMon::setSysMonitoring().

#define TWO_BILLION   2000000000

Definition at line 153 of file ApMon.h.

Referenced by ApMon::sendTimedParameters().

#define XDR_INT32   2

Used to code the 4 bytes integer data type.

Definition at line 97 of file ApMon.h.

Referenced by ApMon::encodeParams(), apmon_utils::logParameters(), ApMon::sendParameter(), ApMon::sendTimedParameters(), apmon_utils::sizeEval(), and apmon_utils::xdrSize().

#define XDR_REAL32   4

Used to code the 4 bytes real data type.

Definition at line 99 of file ApMon.h.

Referenced by ApMon::encodeParams(), apmon_utils::logParameters(), ApMon::sendParameter(), apmon_utils::sizeEval(), and apmon_utils::xdrSize().

#define XDR_REAL64   5

Used to code the 8 bytes real data type.

Definition at line 100 of file ApMon.h.

Referenced by ApMon::encodeParams(), apmon_utils::logParameters(), ApMon::sendGeneralInfo(), ApMon::sendOneJobInfo(), ApMon::sendParameter(), ApMon::sendSysInfo(), apmon_utils::sizeEval(), and apmon_utils::xdrSize().

#define XDR_STRING   0

Used to code the string data type.

Definition at line 95 of file ApMon.h.

Referenced by ApMon::encodeParams(), apmon_utils::logParameters(), ApMon::sendGeneralInfo(), ApMon::sendParameter(), ApMon::sendTimedParameters(), apmon_utils::sizeEval(), and apmon_utils::xdrSize().

---

Function Documentation

void* bkTask

(

void *

param

)

This function is executed in a background thread and has two roles: it automatically sends the system/job monitoring parameters (if the user requested) and it checks the configuration file/URLs for changes.

(this is done in a separate thread).

Definition at line 913 of file ApMon.cpp.

References bkTask(), ApMon::confChangedCond, FILE_INIT, INFO, JOB_INFO_SEND, apmon_utils::logger(), RECHECK_CONF, RECHECK_INTERVAL, SYS_INFO_SEND, apmon_utils::urlModified(), WARNING, and std::runtime_error::what().

Referenced by bkTask().

                          { 
#else
DWORD WINAPI bkTask(void *param) {
#endif
  struct stat st;
#ifndef WIN32
  struct timespec delay;
#else
  DWORD delay;
#endif
  bool resourceChanged, haveChange;
  int nextOp = -1, i, ret;
  int generalInfoCount;
  time_t crtTime, timeRemained;
  time_t nextRecheck = 0, nextJobInfoSend = 0, nextSysInfoSend = 0;
  ApMon *apm = (ApMon *)param;
  char logmsg[200];

  logger(INFO, "[Starting background thread...]");
  apm -> bkThreadStarted = true;

  crtTime = time(NULL);

  pthread_mutex_lock(&(apm -> mutexBack));
  if (apm -> confCheck) {
    nextRecheck = crtTime + apm -> crtRecheckInterval;
    //sprintf(logmsg, "###1 crt %ld interv %ld recheck %ld ", crtTime,
    //   apm -> crtRecheckInterval, nextRecheck);
    //logger(FINE, logmsg);
    //fflush(stdout);
  }
  if (apm -> jobMonitoring)
    nextJobInfoSend = crtTime + apm -> jobMonitorInterval;
  if (apm -> sysMonitoring)
    nextSysInfoSend = crtTime + apm -> sysMonitorInterval;
  pthread_mutex_unlock(&(apm -> mutexBack));
  
  timeRemained = -1;
  generalInfoCount = 0;

  while (1) {
    pthread_mutex_lock(&apm -> mutexBack);
    if (apm -> stopBkThread) {
//      printf("### stopBkThread \n");
      pthread_mutex_unlock(&apm -> mutexBack);
      break;
    }
    pthread_mutex_unlock(&apm -> mutexBack);

    //sprintf(logmsg, "### 2 recheck %ld sys %ld ", nextRecheck, 
    //    nextSysInfoSend);
    //logger(FINE, logmsg);

    /* determine the next operation that must be performed */
    if (nextRecheck > 0 && (nextJobInfoSend <= 0 || 
                            nextRecheck <= nextJobInfoSend)) {
      if (nextSysInfoSend <= 0 || nextRecheck <= nextSysInfoSend) {
        nextOp = RECHECK_CONF;
        timeRemained = nextRecheck - crtTime;
      } else {
        nextOp = SYS_INFO_SEND;
        timeRemained = nextSysInfoSend - crtTime;
      }
    } else {
      if (nextJobInfoSend > 0 && (nextSysInfoSend <= 0 || 
                                  nextJobInfoSend <= nextSysInfoSend)) {
        nextOp = JOB_INFO_SEND;
        timeRemained = nextJobInfoSend - crtTime;
      } else if (nextSysInfoSend > 0) {
        nextOp = SYS_INFO_SEND;
        timeRemained = nextSysInfoSend - crtTime;
      }
    }

    if (timeRemained == -1)
      timeRemained = RECHECK_INTERVAL;

#ifndef WIN32
    /* the moment when the next operation should be performed */
    delay.tv_sec = crtTime + timeRemained;
    delay.tv_nsec = 0;
#else
    delay = (/*crtTime +*/ timeRemained) * 1000;  // this is in millis
#endif

    pthread_mutex_lock(&(apm -> mutexBack));

    pthread_mutex_lock(&(apm -> mutexCond));
    /* check for changes in the settings */
    haveChange = false;
    if (apm -> jobMonChanged || apm -> sysMonChanged || apm -> recheckChanged)
      haveChange = true;
    if (apm -> jobMonChanged) {
      if (apm -> jobMonitoring) 
        nextJobInfoSend = crtTime + apm -> jobMonitorInterval;
      else
        nextJobInfoSend = -1;
      apm -> jobMonChanged = false;
    }
    if (apm -> sysMonChanged) {
      if (apm -> sysMonitoring) 
        nextSysInfoSend = crtTime + apm -> sysMonitorInterval;
      else
        nextSysInfoSend = -1;
      apm -> sysMonChanged = false;
    }
    if (apm -> recheckChanged) {
      if (apm -> confCheck) {
        nextRecheck = crtTime + apm -> crtRecheckInterval;
      }
      else
        nextRecheck = -1;
      apm -> recheckChanged = false;
    }
    pthread_mutex_unlock(&(apm -> mutexBack));

    if (haveChange) {
      pthread_mutex_unlock(&(apm -> mutexCond));
      continue;
    }
    
    /* wait until the next operation should be performed or until
       a change in the settings occurs */
#ifndef WIN32
    ret = pthread_cond_timedwait(&(apm -> confChangedCond), 
                                &(apm -> mutexCond), &delay);
    pthread_mutex_unlock(&(apm -> mutexCond));
#else
    pthread_mutex_unlock(&(apm -> mutexCond));
    ret = WaitForSingleObject(apm->confChangedCond, delay);
#endif
    if (ret == ETIMEDOUT) {
//      printf("### ret TIMEDOUT\n");
      /* now perform the operation */
      if (nextOp == JOB_INFO_SEND) {
        apm -> sendJobInfo();
        crtTime = time(NULL);
        nextJobInfoSend = crtTime + apm -> getJobMonitorInterval();
      }
      
      if (nextOp == SYS_INFO_SEND) {
        apm -> sendSysInfo();
        if (apm -> getGenMonitoring()) {
          if (generalInfoCount <= 1)
            apm -> sendGeneralInfo();
          generalInfoCount = (generalInfoCount + 1) % apm -> genMonitorIntervals;
        }
        crtTime = time(NULL);
        nextSysInfoSend = crtTime + apm -> getSysMonitorInterval();
      }

      if (nextOp == RECHECK_CONF) {
        //logger(FINE, "### recheck conf");
        resourceChanged = false;
        try {
          if (apm -> initType == FILE_INIT) {
            sprintf(logmsg, "Checking for modifications for file %s ", 
                    apm -> initSources[0]);
            logger(INFO, logmsg);
            stat(apm -> initSources[0], &st);
            if (st.st_mtime > apm -> lastModifFile) {
              sprintf(logmsg, "File %s modified ", apm -> initSources[0]);
              logger(INFO, logmsg);
              resourceChanged = true;
            }
          }

          // check the configuration URLs
          for (i = 0; i < apm -> confURLs.nConfURLs; i++) {
            sprintf(logmsg, "[Checking for modifications for URL %s ] ", 
                   apm -> confURLs.vURLs[i]);
            logger(INFO, logmsg);
            if (urlModified(apm -> confURLs.vURLs[i], apm -> confURLs.lastModifURLs[i])) {
              sprintf(logmsg, "URL %s modified ", apm -> confURLs.vURLs[i]);
              logger(INFO, logmsg);
              resourceChanged = true;
              break;
            }
          }

          if (resourceChanged) {
            logger(INFO, "Reloading configuration...");
            if (apm -> initType == FILE_INIT)
              apm -> initialize(apm -> initSources[0], false);
            else
              apm -> initialize(apm -> nInitSources, apm -> initSources, false);
          }
          apm -> setCrtRecheckInterval(apm -> getRecheckInterval());
        } catch (runtime_error &err) {
          logger(WARNING, err.what());
          logger(WARNING, "Increasing the time interval for reloading the configuration...");
          apm -> setCrtRecheckInterval(apm -> getRecheckInterval() * 5);
        }
        crtTime = time(NULL);
        nextRecheck = crtTime + apm -> getCrtRecheckInterval();
        //sleep(apm -> getCrtRecheckInterval());
      }
    }
 
  } // while

#ifndef WIN32
  return NULL; // it doesn't matter what we return here
#else
  return 0;
#endif
}

---