Fixed small bug when stopping the scheduler; re-enabled the test (after...

Fixed small bug when stopping the scheduler; re-enabled the test (after leaving it more time to execute).

gridtk/local.py

@@ -94,13 +94,14 @@ class JobManagerLocal(JobManager):
     self.lock()
 
     job, array_job = self._job_and_array(job_id, array_id)
-    if job.status == 'executing':
-      logger.info("Reset job '%s' in the database" % job)
-      job.status = 'submitted'
+    if job is not None:
+      if job.status == 'executing':
+        logger.info("Reset job '%s' in the database" % job)
+        job.status = 'submitted'
 
-    if array_job is not None and array_job.status == 'executing':
-      logger.debug("Reset array job '%s' in the database" % array_job)
-      array_job.status = 'submitted'
+      if array_job is not None and array_job.status == 'executing':
+        logger.debug("Reset array job '%s' in the database" % array_job)
+        array_job.status = 'submitted'
 
     self.session.commit()
     self.unlock()
@@ -142,14 +143,14 @@ class JobManagerLocal(JobManager):
     self.lock()
     # get the files to write to
     job, array_job = self._job_and_array(job_id, array_id)
-    if no_log:
+    if job is None or no_log:
       out, err = None, None
-    elif array_job:
+    elif array_job is not None:
       out, err = array_job.std_out_file(), array_job.std_err_file()
     else:
       out, err = job.std_out_file(), job.std_err_file()
 
-    log_dir = job.log_dir if not no_log else None
+    log_dir = job.log_dir if not no_log and job is not None else None
     job_id = job.id
     array_id = array_job.id if array_job else None
     self.unlock()

gridtk/manager.py

@@ -78,7 +78,12 @@ class JobManager:
   def _job_and_array(self, job_id, array_id = None):
     # get the job (and the array job) with the given id(s)
     job = self.get_jobs((job_id,))
-    assert (len(job) == 1)
+    if len(job) > 1:
+      logger.error("%d jobs with the same ID '%d' were detected in the database"%(len(job), job_id))
+    elif not len(job):
+      logger.warn("Job with ID '%d' was not found in the database."%job_id)
+      return (None, None)
+
     job = job[0]
     unique_id = job.unique
 

gridtk/sge.py

@@ -142,5 +142,5 @@ class JobManagerSGE(JobManager):
         logger.info("Stopped job '%s' in the SGE grid." % job)
         job.status = 'submitted'
 
-    self.session.commit()
+      self.session.commit()
     self.unlock()

gridtk/tests/__init__.py

@@ -29,6 +29,7 @@ class GridTKTest(unittest.TestCase):
     # make sure that all scheduler jobs are stopped after exiting
     if self.scheduler_job:
       self.scheduler_job.send_signal(signal.SIGINT)
+      self.scheduler_job.wait()
     # Clean up the mess that we created
     import shutil
     shutil.rmtree(self.temp_dir)
@@ -36,8 +37,6 @@ class GridTKTest(unittest.TestCase):
   def test01_local(self):
     # This test executes all commands of the local grid manager and asserts that everything is fine
 
-    raise nose.plugins.skip.SkipTest("[AA, 17.08.2013] Skipping test because it does not work on nightlies until MG is back from holidays")
-
     # first, add some commands to the database
     script_1 = pkg_resources.resource_filename('gridtk.tests', 'test_script.sh')
     script_2 = pkg_resources.resource_filename('gridtk.tests', 'test_array.sh')
@@ -79,7 +78,7 @@ class GridTKTest(unittest.TestCase):
     self.scheduler_job = subprocess.Popen(['./bin/jman', '--local', '--database', self.database, 'run-scheduler', '--sleep-time', '5', '--parallel', '2'])
 
     # sleep some time to assure that the scheduler was able to start the first job
-    time.sleep(2)
+    time.sleep(4)
     # ... and kill the scheduler
     self.scheduler_job.send_signal(signal.SIGINT)
     self.scheduler_job = None
@@ -100,10 +99,10 @@ class GridTKTest(unittest.TestCase):
     jman.main(['./bin/jman', '--local', '--database', self.database, 'resubmit', '--job-id', '1', '--running-jobs'])
 
     # now, start the local execution of the job in a parallel job
-    self.scheduler_job = subprocess.Popen(['./bin/jman', '--local', '--database', self.database, 'run-scheduler', '--sleep-time', '4', '--parallel', '2'])
+    self.scheduler_job = subprocess.Popen(['./bin/jman', '--local', '--database', self.database, 'run-scheduler', '--sleep-time', '5', '--parallel', '2'])
 
     # sleep some time to assure that the scheduler was able to finish the first and start the second job
-    time.sleep(6)
+    time.sleep(9)
     # ... and kill the scheduler
     self.scheduler_job.send_signal(signal.SIGINT)
     self.scheduler_job = None
@@ -138,9 +137,9 @@ class GridTKTest(unittest.TestCase):
     assert os.path.exists(self.log_dir)
 
     # now, let the scheduler run all jobs
-    self.scheduler_job = subprocess.Popen(['./bin/jman', '--local', '--database', self.database, 'run-scheduler', '--sleep-time', '0.1', '--parallel', '2', '--die-when-finished'])
+    self.scheduler_job = subprocess.Popen(['./bin/jman', '--local', '--database', self.database, 'run-scheduler', '--sleep-time', '1', '--parallel', '2', '--die-when-finished'])
     # ... and kill the scheduler
-    time.sleep(3)
+    time.sleep(10)
     assert self.scheduler_job.poll() is not None
     self.scheduler_job = None