class K8sJobActions(BaseJobActions):
"""
Used by K8sLando to handle messages at a job specific context.
"""
def __init__(self, settings):
super(K8sJobActions, self).__init__(settings)
self.cluster_api = settings.get_cluster_api()
self.bespin_job = self.job_api.get_job()
self.manager = JobManager(self.cluster_api, settings.config,
self.bespin_job)
def _set_job_state(self, state):
# Keep cached bespin_job state up to date
super(K8sJobActions, self)._set_job_state(state)
self.bespin_job.state = state
def _set_job_step(self, step):
# Keep cached bespin_job step up to date
super(K8sJobActions, self)._set_job_step(step)
self.bespin_job.step = step
def job_is_at_state_and_step(self, state, step):
return self.bespin_job.state == state and self.bespin_job.step == step
def start_job(self, payload):
"""
Request from user to start running a job. This starts a job to stage user input data into a volume.
:param payload:StartJobPayload contains job_id we should start
"""
self._set_job_state(JobStates.RUNNING)
self._set_job_step(JobSteps.CREATE_VM)
input_files = self.job_api.get_input_files()
input_files_size_in_g = self._calculate_input_data_size_in_g(
input_files)
# The stage data volume contains the workflow, job order, file metadata, and the user's input files.
stage_data_volume_size_in_g = self.config.base_stage_data_volume_size_in_g + input_files_size_in_g
self._show_status("Creating stage data persistent volumes")
self.manager.create_stage_data_persistent_volumes(
stage_data_volume_size_in_g)
self.perform_staging_step(input_files)
@staticmethod
def _calculate_input_data_size_in_g(input_files):
total_bytes = 0
for dds_file in input_files.dds_files:
total_bytes += dds_file.size
for url_file in input_files.url_files:
total_bytes += url_file.size
return math.ceil(float(total_bytes) / (1024.0 * 1024.0 * 1024.0))
def perform_staging_step(self, input_files):
self._set_job_step(JobSteps.STAGING)
self._show_status("Creating Stage data job")
job = self.manager.create_stage_data_job(input_files)
self._show_status("Launched stage data job: {}".format(
job.metadata.name))
def stage_job_complete(self, payload):
"""
Message from worker that a the staging job step is complete and successful.
Sets the job state to RUNNING and puts the run job message into the queue for the worker.
:param payload: JobStepCompletePayload: contains job id and vm_instance_name(unused)
"""
if not self.job_is_at_state_and_step(JobStates.RUNNING,
JobSteps.STAGING):
# ignore request to perform incompatible step
logging.info(
"Ignoring request to run job:{} wrong step/state".format(
self.job_id))
return
self._set_job_step(JobSteps.RUNNING)
self._show_status("Cleaning up after stage data")
self.manager.cleanup_stage_data_job()
self._show_status("Creating volumes for running workflow")
self.manager.create_run_workflow_persistent_volumes()
self.run_workflow_job()
def run_workflow_job(self):
self._show_status("Creating run workflow job")
job = self.manager.create_run_workflow_job()
self._show_status("Launched run workflow job: {}".format(
job.metadata.name))
def run_job_complete(self, payload):
"""
Message from worker that a the run job step is complete and successful.
Sets the job state to STORING_OUTPUT and puts the store output message into the queue for the worker.
:param payload: JobStepCompletePayload: contains job id and vm_instance_name(unused)
"""
if not self.job_is_at_state_and_step(JobStates.RUNNING,
JobSteps.RUNNING):
# ignore request to perform incompatible step
logging.info(
"Ignoring request to store output for job:{} wrong step/state".
format(self.job_id))
return
self.manager.cleanup_run_workflow_job()
self.organize_output_project()
def organize_output_project(self):
self._set_job_step(JobSteps.ORGANIZE_OUTPUT_PROJECT)
self._show_status("Creating organize output project job")
methods_document = self.job_api.get_workflow_methods_document(
self.bespin_job.workflow.methods_document)
methods_content = None
if methods_document:
methods_content = methods_document.content
job = self.manager.create_organize_output_project_job(methods_content)
self._show_status("Launched organize output project job: {}".format(
job.metadata.name))
def organize_output_complete(self, payload):
if not self.job_is_at_state_and_step(JobStates.RUNNING,
JobSteps.ORGANIZE_OUTPUT_PROJECT):
# ignore request to perform incompatible step
logging.info(
"Ignoring request to organize output project for job:{} wrong step/state"
.format(self.job_id))
return
self.manager.cleanup_organize_output_project_job()
self.save_output()
def save_output(self):
store_output_data = self.job_api.get_store_output_job_data()
# get_store_output_job_data
self._set_job_step(JobSteps.STORING_JOB_OUTPUT)
self._show_status("Creating store output job")
job = self.manager.create_save_output_job(
store_output_data.share_dds_ids)
self._show_status("Launched save output job: {}".format(
job.metadata.name))
def store_job_output_complete(self, payload):
"""
Message from worker that a the store output job step is complete and successful.
Records information about the resulting output project and frees cloud resources.
:param payload: JobStepCompletePayload: contains job id and vm_instance_name(unused)
"""
if not self.job_is_at_state_and_step(JobStates.RUNNING,
JobSteps.STORING_JOB_OUTPUT):
# ignore request to perform incompatible step
logging.info(
"Ignoring request to cleanup for job:{} wrong step/state".
format(self.job_id))
return
self.manager.cleanup_save_output_job()
self._set_job_step(JobSteps.RECORD_OUTPUT_PROJECT)
self._show_status("Creating record output project job")
job = self.manager.create_record_output_project_job()
self._show_status("Launched record output project job: {}".format(
job.metadata.name))
def record_output_project_complete(self, payload):
"""
Records the output project id and readme file id that based on the store output pod logs.
"""
if not self.job_is_at_state_and_step(JobStates.RUNNING,
JobSteps.RECORD_OUTPUT_PROJECT):
# ignore request to perform incompatible step
logging.info(
"Ignoring request to cleanup for job:{} wrong step/state".
format(self.job_id))
return
project_id, readme_file_id = self.manager.read_record_output_project_details(
)
self._show_status("Saving project id {} and readme id {}.".format(
project_id, readme_file_id))
self.job_api.save_project_details(project_id, readme_file_id)
self.manager.cleanup_record_output_project_job()
self._show_status("Marking job finished")
self._set_job_step(JobSteps.NONE)
self._set_job_state(JobStates.FINISHED)
def restart_job(self, payload):
"""
Request from user to resume running a job. It will resume based on the value of bespin_job.step
returned from the job api. Canceled jobs will always restart from the beginning
:param payload:RestartJobPayload contains job_id we should restart
"""
full_restart = False
if self.bespin_job.state != JobStates.CANCELED:
self.manager.cleanup_jobs_and_config_maps()
if self.bespin_job.step == JobSteps.STAGING:
self._set_job_state(JobStates.RUNNING)
input_files = self.job_api.get_input_files()
self.perform_staging_step(input_files)
elif self.bespin_job.step == JobSteps.RUNNING:
self._set_job_state(JobStates.RUNNING)
self.run_workflow_job()
elif self.bespin_job.step == JobSteps.ORGANIZE_OUTPUT_PROJECT:
self._set_job_state(JobStates.RUNNING)
self.organize_output_project()
elif self.bespin_job.step == JobSteps.STORING_JOB_OUTPUT:
self._set_job_state(JobStates.RUNNING)
self.save_output()
elif self.bespin_job.step == JobSteps.RECORD_OUTPUT_PROJECT:
self.cannot_restart_step_error(
step_name="record output project")
else:
full_restart = True
else:
full_restart = True
if full_restart:
self.manager.cleanup_all()
self.start_job(None)
def cancel_job(self, payload):
"""
Request from user to cancel a running a job.
Sets status to canceled and terminates the associated jobs, configmaps and pvcs
:param payload: CancelJobPayload: contains job id we should cancel
"""
self._set_job_step(JobSteps.NONE)
self._set_job_state(JobStates.CANCELED)
self._show_status("Canceling job")
self.manager.cleanup_all()
def stage_job_error(self, payload):
"""
Message from watcher that the staging job had an error
:param payload:JobStepErrorPayload: info about error
"""
self._job_step_failed("Staging job failed", payload)
def run_job_error(self, payload):
"""
Message from watcher that the run workflow job had an error
:param payload:JobStepErrorPayload: info about error
"""
self._job_step_failed("Running job failed", payload)
def organize_output_error(self, payload):
"""
Message from watcher that the organize output project job had an error
:param payload:JobStepErrorPayload: info about error
"""
self._job_step_failed("Organize output job failed", payload)
def store_job_output_error(self, payload):
"""
Message from watcher that the store output project job had an error
:param payload:JobStepErrorPayload: info about error
"""
self._job_step_failed("Storing job output failed", payload)
def record_output_project_error(self, payload):
self._job_step_failed("Recording output project failed", payload)
def _job_step_failed(self, message, payload):
self._set_job_state(JobStates.ERRORED)
self._show_status(message)
self._log_error(message=payload.message)
def test_create_run_workflow_job(self):
mock_cluster_api = Mock()
mock_config = Mock(storage_class_name='nfs')
manager = JobManager(cluster_api=mock_cluster_api,
config=mock_config,
job=self.mock_job)
manager.create_run_workflow_job()
# it should have created a job to run the workflow with several volumes mounted
args, kwargs = mock_cluster_api.create_job.call_args
name, batch_spec = args
self.assertEqual(name, 'run-workflow-51-jpb') # job name
self.assertEqual(batch_spec.name,
'run-workflow-51-jpb') # job spec name
self.assertEqual(batch_spec.labels['bespin-job-id'],
'51') # Bespin job id stored in a label
self.assertEqual(batch_spec.labels['bespin-job-step'],
'run_workflow') # store the job step in a label
job_container = batch_spec.container
self.assertEqual(job_container.name,
'run-workflow-51-jpb') # container name
self.assertEqual(job_container.image_name,
self.mock_job.k8s_settings.run_workflow.image_name,
'run workflow image name is based on job settings')
expected_bash_command = 'cwltool --cachedir /bespin/output-data/tmpout/ ' \
'--outdir /bespin/output-data/results/ ' \
'--max-ram 1G --max-cores 2 ' \
'--usage-report /bespin/output-data/job-51-jpb-resource-usage.json ' \
'--stdout /bespin/output-data/bespin-workflow-output.json ' \
'--stderr /bespin/output-data/bespin-workflow-output.log ' \
'/bespin/job-data/workflow/someurl.cwl#main ' \
'/bespin/job-data/job-order.json'.split(' ')
self.assertEqual(
job_container.command, expected_bash_command,
'run workflow command combines job settings and staged files')
self.assertEqual(
job_container.env_dict['CALRISSIAN_POD_NAME'].field_path,
'metadata.name',
'We should store the pod name in a CALRISSIAN_POD_NAME environment variable'
)
self.assertEqual(
job_container.requested_cpu,
self.mock_job.k8s_settings.run_workflow.cpus,
'run workflow requested cpu is based on a job setting')
self.assertEqual(
job_container.requested_memory,
self.mock_job.k8s_settings.run_workflow.memory,
'run workflow requested memory is based on a job setting')
self.assertEqual(len(job_container.volumes), 3)
job_data_volume = job_container.volumes[0]
self.assertEqual(job_data_volume.name, 'job-data-51-jpb')
self.assertEqual(job_data_volume.mount_path, '/bespin/job-data')
self.assertEqual(job_data_volume.volume_claim_name, 'job-data-51-jpb')
self.assertEqual(job_data_volume.read_only, True,
'job data should be a read only volume')
output_data_volume = job_container.volumes[1]
self.assertEqual(output_data_volume.name, 'output-data-51-jpb')
self.assertEqual(output_data_volume.mount_path, '/bespin/output-data')
self.assertEqual(output_data_volume.volume_claim_name,
'output-data-51-jpb')
self.assertEqual(output_data_volume.read_only, False)
system_data_volume = job_container.volumes[2]
self.assertEqual(system_data_volume.name, 'system-data-51-jpb')
self.assertEqual(
system_data_volume.mount_path,
mock_config.run_workflow_settings.system_data_volume.mount_path,
'mount path for the system volume is based on a config setting')
self.assertEqual(
system_data_volume.volume_claim_name, mock_config.
run_workflow_settings.system_data_volume.volume_claim_name,
'pvc name for the system volume is based on a config setting')
self.assertEqual(
system_data_volume.read_only, True,
'system data should be read only for running workflow')
