def stop_job(self):
while True:
# first get the entity version
job_info = self.get_job_info()
version = job_info.status.version.value
request = stateless_svc.StopJobRequest(
job_id=v1alpha_peloton.JobID(value=self.job_id),
version=v1alpha_peloton.EntityVersion(value=version),
)
try:
self.client.stateless_svc.StopJob(
request,
metadata=self.client.jobmgr_metadata,
timeout=default_timeout,
)
except grpc.RpcError as e:
# if entity version is incorrect, just retry
if (
e.code() == grpc.StatusCode.ABORTED
and INVALID_ENTITY_VERSION_ERR_MESSAGE in e.details()
):
continue
raise
break
def stop(self, ranges=None, entity_version=None):
"""
Stops certain pods based on the ranges.
If ranges is not provided then it stops the job
Job level stop does not support range.
We are using pod api for range operation.
We do this for backward compatibility of existing tests
:param ranges: the instance ranges to stop
:param entity_version: the entity version of the job, for concurrency control.
If entity_version is provided, stop will use the provided value,
and raise an exception if version is wrong.
if entity_version is not provided, stop will query job runtime to
get config version and retry until version is correct.
:return: stop response from the API
"""
# wait for job manager leader
self.wait_for_jobmgr_available()
if ranges is None:
job_entity_version = (entity_version or self.entity_version
or self.get_status().version.value)
while True:
request = stateless_svc.StopJobRequest(
job_id=v1alpha_peloton.JobID(value=self.job_id),
version=v1alpha_peloton.EntityVersion(
value=job_entity_version),
)
try:
resp = self.client.stateless_svc.StopJob(
request,
metadata=self.client.jobmgr_metadata,
timeout=self.config.rpc_timeout_sec,
)
except grpc.RpcError as e:
# if entity version is incorrect, get entity version from job status
# and try again.
if (e.code() == grpc.StatusCode.ABORTED and
INVALID_ENTITY_VERSION_ERR_MESSAGE in e.details()
and entity_version is None):
job_entity_version = (entity_version or
self.get_status().version.value)
continue
raise
break
self.entity_version = resp.version.value
log.info("job stopped, new entity version: %s",
self.entity_version)
return resp
for pod_range in ranges:
for pod_id in range(getattr(pod_range, "from"), pod_range.to):
pod_name = self.job_id + "-" + str(pod_id)
request = pod_svc.StopPodRequest(
pod_name=v1alpha_peloton.PodName(value=pod_name))
self.client.pod_svc.StopPod(
request,
metadata=self.client.jobmgr_metadata,
timeout=self.config.rpc_timeout_sec,
)
log.info("stopping pods in job {0} with ranges {1}".format(
self.job_id, ranges))
return pod_svc.StopPodResponse()