Merge pull request #429 from bernt-matthias/deactivate_job

Better cleanup after external DRMAA and condor jobs

Author: mvdbeek

pulsar/managers/queued_condor.py

@@ -15,10 +15,6 @@
 log = getLogger(__name__)
 
 
-# TODO:
-#  - user_log_sizes and state_cache never expire
-#    elements never expire. This is a small memory
-#    leak that should be fixed.
 class CondorQueueManager(ExternalBaseManager):
     """
     Job manager backend that plugs into Condor.
@@ -92,3 +88,9 @@ def __get_state_from_log(self, external_id, log_file):
         self.user_log_sizes[external_id] = log_size
         self.state_cache[external_id] = state
         return state
+
+    def _deactivate_job(self, job_id: str) -> None:
+        external_id = self._external_id(job_id)
+        self.user_log_sizes.pop(external_id, None)
+        self.state_cache.pop(external_id, None)
+        super()._deactivate_job(job_id)

pulsar/managers/queued_external_drmaa.py

@@ -90,6 +90,12 @@ def __sudo(self, *cmds, **kwargs):
         assert p.returncode == 0, "{}, {}".format(stdout, stderr)
         return stdout
 
+    def _deactivate_job(self, job_id: str) -> None:
+        external_id = self._external_id(job_id)
+        del self.user_map[external_id]
+        self.reclaimed.pop(job_id, None)
+        super()._deactivate_job(job_id)
+
 
 def _handle_default(value, script_name):
     """ There are two potential variants of these scripts,