Reverted cpu monitoring, removed threshold which is useless

PILOTVERSION

@@ -1 +1 @@
-3.9.2.29
+3.9.2.30

pilot/util/constants.py

@@ -28,7 +28,7 @@
 RELEASE = '3'   # released number should be fixed at 3 for Pilot 3
 VERSION = '9'   # version number is '1' for first release, '0' until then, increased for bigger updates
 REVISION = '2'  # revision number should be reset to '0' for every new version release, increased for small updates
-BUILD = '29'     # build number should be reset to '1' for every new development cycle
+BUILD = '30'     # build number should be reset to '1' for every new development cycle
 
 SUCCESS = 0
 FAILURE = 1

pilot/util/monitoring.py

@@ -116,12 +116,14 @@ def job_monitor_tasks(job: JobData, mt: MonitoringTime, args: object) -> tuple[i
         else:
             _cpuconsumptiontime = int(round(cpuconsumptiontime))
             if _cpuconsumptiontime > 0:
+                # make sure there are no sudden jumps in the cpuconsumptiontime
+                #factor = _cpuconsumptiontime / job.cpuconsumptiontime
                 job.cpuconsumptiontime = int(round(cpuconsumptiontime))
                 job.cpuconversionfactor = 1.0
                 logger.info(f'(instant) CPU consumption time for pid={job.pid}: {cpuconsumptiontime} (rounded to {job.cpuconsumptiontime})')
             elif _cpuconsumptiontime == -1:
                 logger.warning('could not get CPU consumption time')
-            elif _cpuconsumptiontime == 0.0:
+            elif _cpuconsumptiontime == 0:
                 logger.warning(f'process {job.pid} can no longer be monitored (due to stat problems) - aborting')
                 return 0, ""
             else:

pilot/util/processes.py

@@ -552,6 +552,7 @@ def get_cpu_consumption_time(t0: tuple) -> float:
 def get_instant_cpu_consumption_time(pid: int) -> float:
     """
     Return the CPU consumption time (system+user time) for a given process, by parsing /prod/pid/stat.
+
     Note 1: the function returns 0.0 if the pid is not set.
     Note 2: the function must sum up all the user+system times for both the main process (pid) and the child
     processes, since the main process is most likely spawning new processes.
@@ -574,21 +575,14 @@ def get_instant_cpu_consumption_time(pid: int) -> float:
         if os.path.exists(path):
             try:
                 with open(path, "r", encoding="utf-8") as fp:
-                    _read = fp.read()
-                    fields = _read.split(' ')[13:17]
+                    fields = fp.read().split(' ')[13:17]
                     utime, stime, cutime, cstime = [(float(f) / hz) for f in fields]
             except IOError as exc:
                 logger.warning(f'exception caught: {exc} (ignored)')
 
     if utime and stime and cutime and cstime:
         # sum up all the user+system times for both the main process (pid) and the child processes
         cpu_consumption_time = utime + stime + cutime + cstime
-        max_threshold = 1e6
-        if cpu_consumption_time > max_threshold:
-            logger.warning(f'CPU consumption time={cpu_consumption_time} for pid={pid} exceeds sanity threshold={max_threshold} (reset to 1.0)')
-            logger.warning(f"utime={utime} stime={stime} cutime={cutime} cstime={cstime} hz={hz}")
-            logger.warning(f"fp.read()={_read}")
-            cpu_consumption_time = 1.0
     else:
         cpu_consumption_time = 0.0