Merge pull request #5684 from belforte/maxWall

introduce MaxWallTimeMinsRun. Fix #5683
dmwm · Jun 3, 2018 · 94f4987 · 94f4987
2 parents 99e7034 + a0550be
commit 94f4987
Show file tree

Hide file tree

Showing 4 changed files with 10 additions and 8 deletions.
diff --git a/src/python/TaskWorker/Actions/DagmanCreator.py b/src/python/TaskWorker/Actions/DagmanCreator.py
@@ -160,14 +160,14 @@
                   ((JobStatus =?= 1) && (time() - EnteredCurrentStatus > 7*24*60*60)) || \
                   ((JobStatus =?= 2) && ( \
                      (MemoryUsage > RequestMemory) || \
-                     (MaxWallTimeMins*60 < time() - EnteredCurrentStatus) || \
+                     (MaxWallTimeMinsRun*60 < time() - EnteredCurrentStatus) || \
                      (DiskUsage > %(max_disk_space)s))) || \
                      (time() > CRAB_TaskEndTime) || \
                   ((JobStatus =?= 1) && (time() > (x509UserProxyExpiration + 86400)))
 +PeriodicRemoveReason = ifThenElse(time() - EnteredCurrentStatus > 7*24*60*60 && isUndefined(MemoryUsage), "Removed due to idle time limit", \
                           ifThenElse(time() > x509UserProxyExpiration, "Removed job due to proxy expiration", \
                             ifThenElse(MemoryUsage > RequestMemory, "Removed due to memory use", \
-                              ifThenElse(MaxWallTimeMins*60 < time() - EnteredCurrentStatus, "Removed due to wall clock limit", \
+                              ifThenElse(MaxWallTimeMinsRun*60 < time() - EnteredCurrentStatus, "Removed due to wall clock limit", \
                                 ifThenElse(DiskUsage >  %(max_disk_space)s, "Removed due to disk usage", \
                                   ifThenElse(time() > CRAB_TaskEndTime, "Removed due to reached CRAB_TaskEndTime", \
                                   "Removed due to job being held"))))))

diff --git a/src/python/TaskWorker/Actions/DagmanSubmitter.py b/src/python/TaskWorker/Actions/DagmanSubmitter.py
@@ -80,6 +80,7 @@
     ('RequestMemory', 'tm_maxmemory'),
     ('RequestCpus', 'tm_numcores'),
     ('MaxWallTimeMins', 'tm_maxjobruntime'),
+    ('MaxWallTimeMinsRun', 'tm_maxjobruntime'),
     ('MaxWallTimeMinsProbe', 'maxproberuntime'),
     ('MaxWallTimeMinsTail', 'maxtailruntime'),
     ('JobPrio', 'tm_priority'),

diff --git a/src/python/TaskWorker/Actions/PreJob.py b/src/python/TaskWorker/Actions/PreJob.py
@@ -305,8 +305,8 @@ def alter_submit(self, crab_retry):
                 maxjobruntime = int(str(self.task_ad.lookup('MaxWallTimeMinsProbe')))
             elif 'MaxWallTimeMinsTail' in self.task_ad and self.stage == 'tail':
                 maxjobruntime = int(str(self.task_ad.lookup('MaxWallTimeMinsTail')))
-            elif 'MaxWallTimeMins' in self.task_ad:
-                maxjobruntime = int(str(self.task_ad.lookup('MaxWallTimeMins')))
+            elif 'MaxWallTimeMinsRun' in self.task_ad:
+                maxjobruntime = int(str(self.task_ad.lookup('MaxWallTimeMinsRun')))
             if 'RequestMemory' in self.task_ad:
                 maxmemory = int(str(self.task_ad.lookup('RequestMemory')))
             if 'RequestCpus' in self.task_ad:
@@ -341,7 +341,8 @@ def alter_submit(self, crab_retry):
         new_submit_text += '+CRAB_TransferOutputs = {0}\n+CRAB_SaveLogsFlag = {1}\n'.format(saveoutputs, savelogs)
         if maxjobruntime is not None:
             new_submit_text += '+EstimatedWallTimeMins = %s\n' % str(maxjobruntime)
-            new_submit_text += '+MaxWallTimeMins = %s\n' % str(maxjobruntime)
+            new_submit_text += '+MaxWallTimeMinsRun = %s\n' % str(maxjobruntime)  # how long it can run
+            new_submit_text += '+MaxWallTimeMins = %s\n' % str(maxjobruntime)     # how long a slot can it match to
         if maxmemory is not None:
             new_submit_text += '+RequestMemory = %s\n' % (str(maxmemory))
         if numcores is not None:

diff --git a/src/python/TaskWorker/Actions/RetryJob.py b/src/python/TaskWorker/Actions/RetryJob.py
@@ -403,11 +403,11 @@ def execute_internal(self, logger, reqname, job_return_code, dag_retry, crab_ret
         if used_job_ad:
             #We can determine walltime and max memory from job ad.
             self.ad = job_ad
-            if 'MaxWallTimeMins' in self.ad:
+            if 'MaxWallTimeMinsRun' in self.ad:
                 try:
-                    MAX_WALLTIME = int(self.ad['MaxWallTimeMins']) * 60
+                    MAX_WALLTIME = int(self.ad['MaxWallTimeMinsRun']) * 60
                 except:
-                    msg = "Unable to get MaxWallTimeMins from job classads. Using the default."
+                    msg = "Unable to get MaxWallTimeMinsRun from job classads. Using the default MAX_WALLTIME."
                     self.logger.debug(msg)
             if 'RequestMemory' in self.ad:
                 try: