fix: sadly, remove MJF

DIRACGrid · Oct 11, 2023 · 1968c85 · 1968c85
1 parent 9d357f3
commit 1968c85
Show file tree

Hide file tree

Showing 8 changed files with 11 additions and 451 deletions.
diff --git a/src/DIRAC/Core/Utilities/MJF.py b/src/DIRAC/Core/Utilities/MJF.py
diff --git a/src/DIRAC/Resources/Computing/BatchSystems/TimeLeft/MJFResourceUsage.py b/src/DIRAC/Resources/Computing/BatchSystems/TimeLeft/MJFResourceUsage.py
diff --git a/src/DIRAC/Resources/Computing/BatchSystems/TimeLeft/TimeLeft.py b/src/DIRAC/Resources/Computing/BatchSystems/TimeLeft/TimeLeft.py
@@ -143,10 +143,6 @@ def __getBatchSystemPlugin(self):
                 name = batchSystem
                 break
 
-        if name is None and "MACHINEFEATURES" in os.environ and "JOBFEATURES" in os.environ:
-            # Only use MJF if legacy batch system information not available for now
-            name = "MJF"
-
         if name is None:
             self.log.warn(f"Batch system type for site {DIRAC.siteName()} is not currently supported")
             return S_ERROR("Current batch system is not supported")

diff --git a/src/DIRAC/Resources/Computing/BatchSystems/TimeLeft/test/Test_TimeLeft.py b/src/DIRAC/Resources/Computing/BatchSystems/TimeLeft/test/Test_TimeLeft.py
@@ -54,7 +54,6 @@
     "    00:00:60.00 6267 40713 25469,14249 12/31-20:52:00 -"
 )
 
-MJF_OUT = "0"
 
 SLURM_OUT_0 = "12345,86400,24,3600,03:00:00"
 SLURM_OUT_1 = "12345,86400,24,3600,4-03:00:00"
@@ -72,7 +71,6 @@
     [
         ("LSF", {}, LSF_OUT, 0.0),
         ("LSF", {"bin": "/usr/bin", "hostNorm": 10.0}, LSF_OUT, 0.0),
-        ("MJF", {}, MJF_OUT, 0.0),
         ("SGE", {}, SGE_OUT, 300.0),
         ("SLURM", {}, SLURM_OUT_0, 432000.0),
         ("SLURM", {}, SLURM_OUT_1, 432000.0),

diff --git a/src/DIRAC/WorkloadManagementSystem/JobWrapper/Watchdog.py b/src/DIRAC/WorkloadManagementSystem/JobWrapper/Watchdog.py
@@ -32,7 +32,6 @@
 from DIRAC import S_ERROR, S_OK, gLogger
 from DIRAC.ConfigurationSystem.Client.Config import gConfig
 from DIRAC.ConfigurationSystem.Client.PathFinder import getSystemInstance
-from DIRAC.Core.Utilities import MJF
 from DIRAC.Core.Utilities.Os import getDiskSpace
 from DIRAC.Core.Utilities.Profiler import Profiler
 from DIRAC.Resources.Computing.BatchSystems.TimeLeft.TimeLeft import TimeLeft
@@ -217,7 +216,6 @@ def execute(self):
             and (time.time() - self.initialValues["StartTime"]) > self.wallClockCheckSeconds * self.wallClockCheckCount
         ):
             self.wallClockCheckCount += 1
-            self._performWallClockChecks()
 
         if self.littleTimeLeft:
             # if we have gone over enough iterations query again
@@ -242,35 +240,6 @@ def execute(self):
             # self.log.debug('Application thread is alive: checking count is %s' %(self.checkCount))
             return S_OK()
 
-    #############################################################################
-    def _performWallClockChecks(self):
-        """Watchdog performs the wall clock checks based on MJF. Signals are sent
-        to processes if we need to stop, but function always returns S_OK()
-        """
-        mjf = MJF.MJF()
-
-        try:
-            wallClockSecondsLeft = mjf.getWallClockSecondsLeft()
-        except Exception:
-            # Just stop if we can't get the wall clock seconds left
-            return S_OK()
-
-        jobstartSeconds = mjf.getIntJobFeature("jobstart_secs")
-        if jobstartSeconds is None:
-            # Just stop if we don't know when the job started
-            return S_OK()
-
-        if (int(time.time()) > jobstartSeconds + self.stopSigStartSeconds) and (
-            wallClockSecondsLeft < self.stopSigFinishSeconds + self.wallClockCheckSeconds
-        ):
-            # Need to send the signal! Assume it works to avoid sending the signal more than once
-            self.log.info("Sending signal to JobWrapper children", f"({self.stopSigNumber})")
-            self.stopSigSent = True
-
-            kill_proc_tree(self.wrapperPID, includeParent=False)
-
-        return S_OK()
-
     #############################################################################
     def _performChecks(self):
         """The Watchdog checks are performed at a different period to the checking of the