ClusterCockpit/cc-condor-sync
1
0
Fork 0
mirror of https://gitlab.cs.uni-saarland.de/hpc/cc-condor-sync.git synced 2024-12-26 13:39:05 +01:00
Code Issues Packages Projects Releases Wiki Activity

Offset ArrayJobId with submitnode id..

Browse Source
This commit is contained in:
Joachim Meyer 2022-11-09 14:27:53 +01:00
parent 9571f3cda6
commit 09334ab4f1
1 changed files with 16 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
htcondor-clustercockpit-push.py
View File

@ -130,6 +130,10 @@ class CondorSync:
command = "%s -j %s --json" % (self.config['slurm']['sacct'], jobid) command = "%s -j %s --json" % (self.config['slurm']['sacct'], jobid)
return json.loads(self._exec(command)) return json.loads(self._exec(command))
def _getSubmitNodeId(self, globalJobId):
job_id_parts = globalJobId.split('#')
return self.config['htcondor']['submitnodes'][job_id_parts[0].split('.')[0]]
def _jobIdInCC(self, jobid): def _jobIdInCC(self, jobid):
for job in self.ccData['jobs']: for job in self.ccData['jobs']:
if jobid == job['jobId']: if jobid == job['jobId']:
@ -138,10 +142,8 @@ class CondorSync:
def _jobIdToInt(self, globalJobId): def _jobIdToInt(self, globalJobId):
job_id_parts = globalJobId.split('#') job_id_parts = globalJobId.split('#')
submit = self.config['htcondor']['submitnodes'][job_id_parts[0].split('.')[
0]]
cluster_id, proc_id = [int(id) for id in job_id_parts[1].split('.')] cluster_id, proc_id = [int(id) for id in job_id_parts[1].split('.')]
return cluster_id << 32 | ((proc_id & 0x3FFFFFFF) << 2) | submit return cluster_id << 32 | ((proc_id & 0x3FFFFFFF) << 2) | self._getSubmitNodeId(globalJobId)
def _jobRunning(self, jobid): def _jobRunning(self, jobid):
for job in self.slurmJobData['jobs']: for job in self.slurmJobData['jobs']:
@ -271,7 +273,8 @@ class CondorSync:
# is this part of an array job? # is this part of an array job?
if job['Cluster'] > 0: if job['Cluster'] > 0:
data.update({"arrayJobId": job['Cluster']}) # todo: offset by submit node! data.update(
{"arrayJobId": job['Cluster'] * 10 + self._getSubmitNodeId(job['GlobalJobId'])})
num_acc = 0 num_acc = 0
for node in nodelist: for node in nodelist:
@ -320,7 +323,8 @@ class CondorSync:
if 'GlobalJobID' in job: if 'GlobalJobID' in job:
globalJobId = job['GlobalJobId'] globalJobId = job['GlobalJobId']
else: else:
globalJobId = "%s#%d.%d#%d" % (self.submit_node, job['Cluster'], job['Proc'], int(time.time())) globalJobId = "%s#%d.%d#%d" % (
self.submit_node, job['Cluster'], job['Proc'], int(time.time()))
print("INFO: Stop job %s" % globalJobId) print("INFO: Stop job %s" % globalJobId)
jobId = self._jobIdToInt(globalJobId) jobId = self._jobIdToInt(globalJobId)
@ -382,9 +386,11 @@ class CondorSync:
if isinstance(job['ToE']['When'], int): if isinstance(job['ToE']['When'], int):
data['stopTime'] = job['ToE']['When'] data['stopTime'] = job['ToE']['When']
else: else:
data['stopTime'] = int(time.mktime(dateparser.parse(job['ToE']['When']).timetuple())) data['stopTime'] = int(time.mktime(
dateparser.parse(job['ToE']['When']).timetuple()))
else: else:
data['stopTime'] = int(time.mktime(dateparser.parse(job['EventTime']).timetuple())) data['stopTime'] = int(time.mktime(
dateparser.parse(job['EventTime']).timetuple()))
if 'JobCurrentStartDate' in job: if 'JobCurrentStartDate' in job:
data['startTime'] = job['JobCurrentStartDate'] data['startTime'] = job['JobCurrentStartDate']
@ -440,8 +446,8 @@ class CondorSync:
elif event is tailf.Truncated: elif event is tailf.Truncated:
print("File was truncated") print("File was truncated")
else: else:
assert False, "unreachable" # currently. more events may be introduced later assert False, "unreachable" # currently. more events may be introduced later
time.sleep(5) # save CPU cycles time.sleep(5) # save CPU cycles
with open(self.config['htcondor']['eventlog'], 'r', encoding='utf-8') as f: with open(self.config['htcondor']['eventlog'], 'r', encoding='utf-8') as f:
eventlog = f.read() eventlog = f.read()