def rerun_job(cls, job_id, initiator_role, initiator_party_id, component_name):
schedule_logger(job_id=job_id).info(f"try to rerun job {job_id} on initiator {initiator_role} {initiator_party_id}")
jobs = JobSaver.query_job(job_id=job_id, role=initiator_role, party_id=initiator_party_id)
if jobs:
job = jobs[0]
else:
raise RuntimeError(f"can not found job {job_id} on initiator {initiator_role} {initiator_party_id}")
if component_name != job_utils.job_virtual_component_name():
tasks = JobSaver.query_task(job_id=job_id, role=initiator_role, party_id=initiator_party_id, component_name=component_name)
else:
tasks = JobSaver.query_task(job_id=job_id, role=initiator_role, party_id=initiator_party_id)
job_can_rerun = False
dsl_parser = schedule_utils.get_job_dsl_parser(dsl=job.f_dsl,
runtime_conf=job.f_runtime_conf,
train_runtime_conf=job.f_train_runtime_conf)
for task in tasks:
if task.f_status in {TaskStatus.WAITING, TaskStatus.COMPLETE}:
if task.f_status == TaskStatus.WAITING:
job_can_rerun = True
schedule_logger(job_id=job_id).info(f"task {task.f_task_id} {task.f_task_version} on {task.f_role} {task.f_party_id} is {task.f_status}, pass rerun")
else:
# stop old version task
FederatedScheduler.stop_task(job=job, task=task, stop_status=TaskStatus.CANCELED)
FederatedScheduler.clean_task(job=job, task=task, content_type="metrics")
# create new version task
task.f_task_version = task.f_task_version + 1
task.f_run_pid = None
task.f_run_ip = None
FederatedScheduler.create_task(job=job, task=task)
# Save the status information of all participants in the initiator for scheduling
schedule_logger(job_id=job_id).info(f"create task {task.f_task_id} new version {task.f_task_version}")
for _role, _party_ids in job.f_runtime_conf["role"].items():
for _party_id in _party_ids:
if _role == initiator_role and _party_id == initiator_party_id:
continue
JobController.initialize_tasks(job_id, _role, _party_id, False, job.f_runtime_conf["initiator"], RunParameters(**job.f_runtime_conf["job_parameters"]), dsl_parser, component_name=task.f_component_name, task_version=task.f_task_version)
schedule_logger(job_id=job_id).info(f"create task {task.f_task_id} new version {task.f_task_version} successfully")
job_can_rerun = True
if job_can_rerun:
if EndStatus.contains(job.f_status):
job.f_status = JobStatus.WAITING
job.f_end_time = None
job.f_elapsed = None
job.f_progress = 0
schedule_logger(job_id=job_id).info(f"job {job_id} has been finished, set waiting to rerun")
status, response = FederatedScheduler.sync_job_status(job=job)
if status == FederatedSchedulingStatusCode.SUCCESS:
FederatedScheduler.sync_job(job=job, update_fields=["end_time", "elapsed", "progress"])
JobQueue.create_event(job_id=job_id, initiator_role=initiator_role, initiator_party_id=initiator_party_id)
schedule_logger(job_id=job_id).info(f"job {job_id} set waiting to rerun successfully")
else:
schedule_logger(job_id=job_id).info(f"job {job_id} set waiting to rerun failed")
else:
# status updates may be delayed, and in a very small probability they will be executed after the rerun command
schedule_logger(job_id=job_id).info(f"job {job_id} status is {job.f_status}, will be run new version waiting task")
else:
schedule_logger(job_id=job_id).info(f"job {job_id} no task to rerun")
def rerun_job(cls, job_id, initiator_role, initiator_party_id, component_name):
schedule_logger(job_id=job_id).info(f"try to rerun job {job_id} on initiator {initiator_role} {initiator_party_id}")
jobs = JobSaver.query_job(job_id=job_id, role=initiator_role, party_id=initiator_party_id)
if jobs:
job = jobs[0]
else:
raise RuntimeError(f"can not found job {job_id} on initiator {initiator_role} {initiator_party_id}")
if component_name != job_utils.job_virtual_component_name():
tasks = JobSaver.query_task(job_id=job_id, role=initiator_role, party_id=initiator_party_id, component_name=component_name)
else:
tasks = JobSaver.query_task(job_id=job_id, role=initiator_role, party_id=initiator_party_id)
job_can_rerun = False
dsl_parser = schedule_utils.get_job_dsl_parser(dsl=job.f_dsl,
runtime_conf=job.f_runtime_conf_on_party,
train_runtime_conf=job.f_train_runtime_conf)
for task in tasks:
if task.f_status in {TaskStatus.WAITING, TaskStatus.SUCCESS}:
if task.f_status == TaskStatus.WAITING:
job_can_rerun = True
schedule_logger(job_id=job_id).info(f"task {task.f_task_id} {task.f_task_version} on {task.f_role} {task.f_party_id} is {task.f_status}, pass rerun")
else:
# stop old version task
FederatedScheduler.stop_task(job=job, task=task, stop_status=TaskStatus.CANCELED)
FederatedScheduler.clean_task(job=job, task=task, content_type="metrics")
# create new version task
task.f_task_version = task.f_task_version + 1
task.f_run_pid = None
task.f_run_ip = None
FederatedScheduler.create_task(job=job, task=task)
# Save the status information of all participants in the initiator for scheduling
schedule_logger(job_id=job_id).info(f"create task {task.f_task_id} new version {task.f_task_version}")
for _role, _party_ids in job.f_runtime_conf_on_party["role"].items():
for _party_id in _party_ids:
if _role == initiator_role and _party_id == initiator_party_id:
continue
JobController.initialize_tasks(job_id, _role, _party_id, False, job.f_initiator_role, job.f_initiator_party_id, RunParameters(**job.f_runtime_conf_on_party["job_parameters"]), dsl_parser, component_name=task.f_component_name, task_version=task.f_task_version)
schedule_logger(job_id=job_id).info(f"create task {task.f_task_id} new version {task.f_task_version} successfully")
job_can_rerun = True
if job_can_rerun:
schedule_logger(job_id=job_id).info(f"job {job_id} set rerun signal")
status = cls.rerun_signal(job_id=job_id, set_or_reset=True)
if status:
schedule_logger(job_id=job_id).info(f"job {job_id} set rerun signal successfully")
else:
schedule_logger(job_id=job_id).info(f"job {job_id} set rerun signal failed")
else:
FederatedScheduler.sync_job_status(job=job)
schedule_logger(job_id=job_id).info(f"job {job_id} no task to rerun")
