def collect_task_of_all_party(cls, job, task):
status, federated_response = FederatedScheduler.collect_task(job=job, task=task)
if status != FederatedSchedulingStatusCode.SUCCESS:
schedule_logger(job_id=job.f_job_id).warning(f"collect task {task.f_task_id} {task.f_task_version} on {task.f_role} {task.f_party_id} failed")
return
for _role in federated_response.keys():
for _party_id, party_response in federated_response[_role].items():
JobSaver.update_task_status(task_info=party_response["data"])
JobSaver.update_task(task_info=party_response["data"])
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 create_job(cls, job_id, role, party_id, job_info):
# parse job configuration
dsl = job_info['dsl']
runtime_conf = job_info['runtime_conf']
train_runtime_conf = job_info['train_runtime_conf']
if USE_AUTHENTICATION:
authentication_check(src_role=job_info.get('src_role', None),
src_party_id=job_info.get(
'src_party_id', None),
dsl=dsl,
runtime_conf=runtime_conf,
role=role,
party_id=party_id)
job_parameters = RunParameters(**runtime_conf['job_parameters'])
job_initiator = runtime_conf['initiator']
dsl_parser = schedule_utils.get_job_dsl_parser(
dsl=dsl,
runtime_conf=runtime_conf,
train_runtime_conf=train_runtime_conf)
# save new job into db
if role == job_initiator['role'] and party_id == job_initiator[
'party_id']:
is_initiator = True
else:
is_initiator = False
job_info["status"] = JobStatus.WAITING
roles = job_info['roles']
# this party configuration
job_info["role"] = role
job_info["party_id"] = party_id
job_info["is_initiator"] = is_initiator
job_info["progress"] = 0
engines_info = cls.get_job_engines_address(
job_parameters=job_parameters)
cls.special_role_parameters(role=role, job_parameters=job_parameters)
cls.check_parameters(job_parameters=job_parameters,
engines_info=engines_info)
runtime_conf["job_parameters"] = job_parameters.to_dict()
JobSaver.create_job(job_info=job_info)
job_utils.save_job_conf(job_id=job_id,
job_dsl=dsl,
job_runtime_conf=runtime_conf,
train_runtime_conf=train_runtime_conf,
pipeline_dsl=None)
cls.initialize_tasks(job_id, role, party_id, True, job_initiator,
job_parameters, dsl_parser)
cls.initialize_job_tracker(job_id=job_id,
role=role,
party_id=party_id,
job_info=job_info,
is_initiator=is_initiator,
dsl_parser=dsl_parser)
def collect_task_of_all_party(cls, job, initiator_task, set_status=None):
tasks_on_all_party = JobSaver.query_task(task_id=initiator_task.f_task_id, task_version=initiator_task.f_task_version)
tasks_status_on_all = set([task.f_status for task in tasks_on_all_party])
if not len(tasks_status_on_all) > 1 and not TaskStatus.RUNNING in tasks_status_on_all:
return
status, federated_response = FederatedScheduler.collect_task(job=job, task=initiator_task)
if status != FederatedSchedulingStatusCode.SUCCESS:
schedule_logger(job_id=job.f_job_id).warning(f"collect task {initiator_task.f_task_id} {initiator_task.f_task_version} on {initiator_task.f_role} {initiator_task.f_party_id} failed")
for _role in federated_response.keys():
for _party_id, party_response in federated_response[_role].items():
if party_response["retcode"] == RetCode.SUCCESS:
JobSaver.update_task_status(task_info=party_response["data"])
JobSaver.update_task(task_info=party_response["data"])
elif party_response["retcode"] == RetCode.FEDERATED_ERROR and set_status:
tmp_task_info = {
"job_id": initiator_task.f_job_id,
"task_id": initiator_task.f_task_id,
"task_version": initiator_task.f_task_version,
"role": _role,
"party_id": _party_id,
"party_status": TaskStatus.RUNNING
}
JobSaver.update_task_status(task_info=tmp_task_info)
tmp_task_info["party_status"] = set_status
JobSaver.update_task_status(task_info=tmp_task_info)
def update_job_on_initiator(cls, initiator_job: Job, update_fields: list):
jobs = JobSaver.query_job(job_id=initiator_job.f_job_id)
if not jobs:
raise Exception("Failed to update job status on initiator")
job_info = initiator_job.to_human_model_dict(only_primary_with=update_fields)
for field in update_fields:
job_info[field] = getattr(initiator_job, "f_%s" % field)
for job in jobs:
job_info["role"] = job.f_role
job_info["party_id"] = job.f_party_id
JobSaver.update_job_status(job_info=job_info)
JobSaver.update_job(job_info=job_info)
def create_task(cls, role, party_id, run_on_this_party, task_info):
task_info["role"] = role
task_info["party_id"] = party_id
task_info["status"] = TaskStatus.WAITING
task_info["party_status"] = TaskStatus.WAITING
task_info["create_time"] = base_utils.current_timestamp()
task_info["run_on_this_party"] = run_on_this_party
if "task_id" not in task_info:
task_info["task_id"] = job_utils.generate_task_id(job_id=task_info["job_id"], component_name=task_info["component_name"])
if "task_version" not in task_info:
task_info["task_version"] = 0
JobSaver.create_task(task_info=task_info)
def stop_jobs(cls, job_id, stop_status, role=None, party_id=None):
if role and party_id:
jobs = JobSaver.query_job(job_id=job_id, role=role, party_id=party_id)
else:
jobs = JobSaver.query_job(job_id=job_id)
kill_status = True
kill_details = {}
for job in jobs:
kill_job_status, kill_job_details = cls.stop_job(job=job, stop_status=stop_status)
kill_status = kill_status & kill_job_status
kill_details[job_id] = kill_job_details
return kill_status, kill_details
def schedule(cls, job, dsl_parser, canceled=False):
schedule_logger(job_id=job.f_job_id).info("scheduling job {} tasks".format(job.f_job_id))
initiator_tasks_group = JobSaver.get_tasks_asc(job_id=job.f_job_id, role=job.f_role, party_id=job.f_party_id)
waiting_tasks = []
for initiator_task in initiator_tasks_group.values():
# collect all party task party status
if job.f_runtime_conf_on_party["job_parameters"]["federated_status_collect_type"] == FederatedCommunicationType.PULL:
tasks_on_all_party = JobSaver.query_task(task_id=initiator_task.f_task_id, task_version=initiator_task.f_task_version)
tasks_status_on_all = set([task.f_status for task in tasks_on_all_party])
if len(tasks_status_on_all) > 1 or TaskStatus.RUNNING in tasks_status_on_all:
cls.collect_task_of_all_party(job=job, task=initiator_task)
new_task_status = cls.federated_task_status(job_id=initiator_task.f_job_id, task_id=initiator_task.f_task_id, task_version=initiator_task.f_task_version)
task_status_have_update = False
if new_task_status != initiator_task.f_status:
task_status_have_update = True
initiator_task.f_status = new_task_status
FederatedScheduler.sync_task_status(job=job, task=initiator_task)
if initiator_task.f_status == TaskStatus.WAITING:
waiting_tasks.append(initiator_task)
elif task_status_have_update and EndStatus.contains(initiator_task.f_status):
FederatedScheduler.stop_task(job=job, task=initiator_task, stop_status=initiator_task.f_status)
scheduling_status_code = SchedulingStatusCode.NO_NEXT
if not canceled:
for waiting_task in waiting_tasks:
for component in dsl_parser.get_upstream_dependent_components(component_name=waiting_task.f_component_name):
dependent_task = initiator_tasks_group[
JobSaver.task_key(task_id=job_utils.generate_task_id(job_id=job.f_job_id, component_name=component.get_name()),
role=job.f_role,
party_id=job.f_party_id
)
]
if dependent_task.f_status != TaskStatus.SUCCESS:
# can not start task
break
else:
# all upstream dependent tasks have been successful, can start this task
scheduling_status_code = SchedulingStatusCode.HAVE_NEXT
status_code = cls.start_task(job=job, task=waiting_task)
if status_code == SchedulingStatusCode.NO_RESOURCE:
# wait for the next round of scheduling
schedule_logger(job_id=job.f_job_id).info(f"job {waiting_task.f_job_id} task {waiting_task.f_task_id} can not apply resource, wait for the next round of scheduling")
break
elif status_code == SchedulingStatusCode.FAILED:
scheduling_status_code = SchedulingStatusCode.FAILED
break
else:
schedule_logger(job_id=job.f_job_id).info("have cancel signal, pass start job {} tasks".format(job.f_job_id))
schedule_logger(job_id=job.f_job_id).info("finish scheduling job {} tasks".format(job.f_job_id))
return scheduling_status_code, initiator_tasks_group.values()
def report_task(job_id, component_name, task_id, task_version, role, party_id):
task_info = {}
task_info.update(request.json)
task_info.update({
"job_id": job_id,
"task_id": task_id,
"task_version": task_version,
"role": role,
"party_id": party_id,
})
JobSaver.update_task(task_info=task_info)
if task_info.get("party_status"):
JobSaver.update_status(Task, task_info)
return get_json_result(retcode=0, retmsg='success')
def upload_history():
request_data = request.json
if request_data.get('job_id'):
tasks = JobSaver.query_task(component_name='upload_0', status=StatusSet.SUCCESS, job_id=request_data.get('job_id'), run_on_this_party=True)
else:
tasks = JobSaver.query_task(component_name='upload_0', status=StatusSet.SUCCESS, run_on_this_party=True)
limit = request_data.get('limit')
if not limit:
tasks = tasks[-1::-1]
else:
tasks = tasks[-1:-limit - 1:-1]
jobs_run_conf = job_utils.get_job_configuration(None, None, None, tasks)
data = get_upload_info(jobs_run_conf=jobs_run_conf)
return get_json_result(retcode=0, retmsg='success', data=data)
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")
def clean_task(cls, job_id, task_id, task_version, role, party_id,
content_type):
status = set()
if content_type == "metrics":
tracker = Tracker(job_id=job_id,
role=role,
party_id=party_id,
task_id=task_id,
task_version=task_version)
status.add(tracker.clean_metrics())
elif content_type == "table":
jobs = JobSaver.query_job(job_id=job_id,
role=role,
party_id=party_id)
if jobs:
job = jobs[0]
job_parameters = RunParameters(
**job.f_runtime_conf_on_party["job_parameters"])
tracker = Tracker(job_id=job_id,
role=role,
party_id=party_id,
task_id=task_id,
task_version=task_version,
job_parameters=job_parameters)
status.add(tracker.clean_task(job.f_runtime_conf_on_party))
if len(status) == 1 and True in status:
return True
else:
return False
def update_job_status(cls, job_info):
update_status = JobSaver.update_job_status(job_info=job_info)
if update_status and EndStatus.contains(job_info.get("status")):
ResourceManager.return_job_resource(job_id=job_info["job_id"],
role=job_info["role"],
party_id=job_info["party_id"])
return update_status
def pipeline_dag_dependency(job_info):
try:
detect_utils.check_config(job_info,
required_arguments=["party_id", "role"])
if job_info.get('job_id'):
jobs = JobSaver.query_job(job_id=job_info["job_id"],
party_id=job_info["party_id"],
role=job_info["role"])
if not jobs:
raise Exception('query job {} failed'.format(
job_info.get('job_id', '')))
job = jobs[0]
job_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)
else:
job_dsl_parser = schedule_utils.get_job_dsl_parser(
dsl=job_info.get('job_dsl', {}),
runtime_conf=job_info.get('job_runtime_conf', {}),
train_runtime_conf=job_info.get('job_train_runtime_conf', {}))
return job_dsl_parser.get_dependency(role=job_info["role"],
party_id=int(
job_info["party_id"]))
except Exception as e:
stat_logger.exception(e)
raise e
def query_task():
tasks = JobSaver.query_task(**request.json)
if not tasks:
return get_json_result(retcode=101, retmsg='find task failed')
return get_json_result(retcode=0,
retmsg='success',
data=[task.to_json() for task in tasks])
def cancel_job(cls, job_id, role, party_id):
schedule_logger(job_id).info(
'{} {} get cancel waiting job {} command'.format(
role, party_id, job_id))
jobs = JobSaver.query_job(job_id=job_id)
if jobs:
job = jobs[0]
try:
# You cannot delete an event directly, otherwise the status might not be updated
status = JobQueue.update_event(
job_id=job.f_job_id,
initiator_role=job.f_initiator_role,
initiator_party_id=job.f_initiator_party_id,
job_status=JobStatus.CANCELED)
if not status:
return False
except:
return False
schedule_logger(job_id).info(
'cancel {} job successfully, job id is {}'.format(
job.f_status, job.f_job_id))
return True
else:
schedule_logger(job_id).warning(
'role {} party id {} cancel job failed, no find jod {}'.format(
role, party_id, job_id))
raise Exception(
'role {} party id {} cancel job failed, no find jod {}'.format(
role, party_id, job_id))
def stop_job():
job_id = request.json.get('job_id')
stop_status = request.json.get("stop_status", "canceled")
jobs = JobSaver.query_job(job_id=job_id)
if jobs:
schedule_logger(job_id).info(f"stop job on this party")
kill_status, kill_details = JobController.stop_jobs(
job_id=job_id, stop_status=stop_status)
schedule_logger(job_id).info(
f"stop job on this party status {kill_status}")
schedule_logger(job_id).info(
f"request stop job {jobs[0]} to {stop_status}")
status_code, response = FederatedScheduler.request_stop_job(
job=jobs[0],
stop_status=stop_status,
command_body=jobs[0].to_json())
if status_code == FederatedSchedulingStatusCode.SUCCESS:
return get_json_result(
retcode=RetCode.SUCCESS,
retmsg=f"stop job on this party {kill_status};\n"
f"stop job on all party success")
else:
return get_json_result(retcode=RetCode.OPERATING_ERROR,
retmsg="stop job on this party {};\n"
"stop job failed:\n{}".format(
kill_status,
json_dumps(response, indent=4)))
else:
schedule_logger(job_id).info(f"can not found job {job_id} to stop")
return get_json_result(retcode=RetCode.DATA_ERROR,
retmsg="can not found job")
def update_job(cls, job_info):
"""
Save to local database
:param job_info:
:return:
"""
return JobSaver.update_job(job_info=job_info)
def stop_job(cls, job_id, role, party_id, stop_status):
schedule_logger(job_id=job_id).info(
f"request stop job {job_id} with {stop_status}")
jobs = JobSaver.query_job(job_id=job_id,
role=role,
party_id=party_id,
is_initiator=True)
if len(jobs) > 0:
if stop_status == JobStatus.CANCELED:
schedule_logger(job_id=job_id).info(f"cancel job {job_id}")
set_cancel_status = cls.cancel_signal(job_id=job_id,
set_or_reset=True)
schedule_logger(job_id=job_id).info(
f"set job {job_id} cancel signal {set_cancel_status}")
job = jobs[0]
job.f_status = stop_status
schedule_logger(job_id=job_id).info(
f"request stop job {job_id} with {stop_status} to all party")
status_code, response = FederatedScheduler.stop_job(
job=jobs[0], stop_status=stop_status)
if status_code == FederatedSchedulingStatusCode.SUCCESS:
schedule_logger(job_id=job_id).info(
f"stop job {job_id} with {stop_status} successfully")
return RetCode.SUCCESS, "success"
else:
schedule_logger(job_id=job_id).info(
f"stop job {job_id} with {stop_status} failed, {response}")
return RetCode.FEDERATED_ERROR, json_dumps(response)
else:
return RetCode.SUCCESS, "can not found job"
def start_job(cls, job_id, initiator_role, initiator_party_id):
schedule_logger(job_id=job_id).info(
"Try to start job {} on initiator {} {}".format(
job_id, initiator_role, initiator_party_id))
job_info = {}
job_info["job_id"] = job_id
job_info["role"] = initiator_role
job_info["party_id"] = initiator_party_id
job_info["status"] = JobStatus.RUNNING
job_info["party_status"] = JobStatus.RUNNING
job_info["start_time"] = current_timestamp()
job_info["tag"] = 'end_waiting'
jobs = JobSaver.query_job(job_id=job_id,
role=initiator_role,
party_id=initiator_party_id)
if jobs:
job = jobs[0]
FederatedScheduler.start_job(job=job)
schedule_logger(job_id=job_id).info(
"start job {} on initiator {} {}".format(
job_id, initiator_role, initiator_party_id))
else:
schedule_logger(job_id=job_id).error(
"can not found job {} on initiator {} {}".format(
job_id, initiator_role, initiator_party_id))
def update_job():
job_info = request.json
jobs = JobSaver.query_job(job_id=job_info['job_id'],
party_id=job_info['party_id'],
role=job_info['role'])
if not jobs:
return get_json_result(retcode=101, retmsg='find job failed')
else:
JobSaver.update_job(
job_info={
'description': job_info.get('notes', ''),
'job_id': job_info['job_id'],
'role': job_info['role'],
'party_id': job_info['party_id']
})
return get_json_result(retcode=0, retmsg='success')
def report_task_to_initiator(cls, task_info):
tasks = JobSaver.query_task(task_id=task_info["task_id"],
task_version=task_info["task_version"],
role=task_info["role"],
party_id=task_info["party_id"])
if tasks[0].f_federated_status_collect_type == FederatedCommunicationType.PUSH:
FederatedScheduler.report_task_to_initiator(task=tasks[0])
def clean_queue():
jobs = JobSaver.query_job(is_initiator=True, status=JobStatus.WAITING)
clean_status = {}
for job in jobs:
status_code, response = FederatedScheduler.request_stop_job(
job=job, stop_status=JobStatus.CANCELED)
clean_status[job.f_job_id] = status_code
return get_json_result(retcode=0, retmsg='success', data=clean_status)
def query_job():
jobs = JobSaver.query_job(**request.json)
if not jobs:
return get_json_result(retcode=0,
retmsg='no job could be found',
data=[])
return get_json_result(retcode=0,
retmsg='success',
data=[job.to_json() for job in jobs])
def component_output_data_table():
request_data = request.json
detect_utils.check_config(config=request_data, required_arguments=['job_id', 'role', 'party_id', 'component_name'])
jobs = JobSaver.query_job(job_id=request_data.get('job_id'))
if jobs:
job = jobs[0]
return jsonify(FederatedScheduler.tracker_command(job, request_data, 'output/table'))
else:
return get_json_result(retcode=100, retmsg='No found job')
def stop_job(cls, job, stop_status):
tasks = JobSaver.query_task(job_id=job.f_job_id, role=job.f_role, party_id=job.f_party_id, reverse=True)
kill_status = True
kill_details = {}
for task in tasks:
kill_task_status = TaskController.stop_task(task=task, stop_status=stop_status)
kill_status = kill_status & kill_task_status
kill_details[task.f_task_id] = 'success' if kill_task_status else 'failed'
return kill_status, kill_details
def create_job(cls, job_id, role, party_id, job_info):
# parse job configuration
dsl = job_info['dsl']
runtime_conf = job_info['runtime_conf']
train_runtime_conf = job_info['train_runtime_conf']
if USE_AUTHENTICATION:
authentication_check(src_role=job_info.get('src_role', None), src_party_id=job_info.get('src_party_id', None),
dsl=dsl, runtime_conf=runtime_conf, role=role, party_id=party_id)
dsl_parser = schedule_utils.get_job_dsl_parser(dsl=dsl,
runtime_conf=runtime_conf,
train_runtime_conf=train_runtime_conf)
job_parameters = dsl_parser.get_job_parameters().get(role, {}).get(party_id, {})
schedule_logger(job_id).info('job parameters:{}'.format(job_parameters))
job_parameters = RunParameters(**job_parameters)
# save new job into db
if role == job_info["initiator_role"] and party_id == job_info["initiator_party_id"]:
is_initiator = True
else:
is_initiator = False
job_info["status"] = JobStatus.WAITING
# this party configuration
job_info["role"] = role
job_info["party_id"] = party_id
job_info["is_initiator"] = is_initiator
job_info["progress"] = 0
cls.adapt_job_parameters(role=role, job_parameters=job_parameters)
engines_info = cls.get_job_engines_address(job_parameters=job_parameters)
cls.check_parameters(job_parameters=job_parameters, role=role, party_id=party_id, engines_info=engines_info)
job_info["runtime_conf_on_party"]["job_parameters"] = job_parameters.to_dict()
job_utils.save_job_conf(job_id=job_id,
role=role,
job_dsl=dsl,
job_runtime_conf=runtime_conf,
job_runtime_conf_on_party=job_info["runtime_conf_on_party"],
train_runtime_conf=train_runtime_conf,
pipeline_dsl=None)
cls.initialize_tasks(job_id=job_id, role=role, party_id=party_id, run_on_this_party=True, initiator_role=job_info["initiator_role"], initiator_party_id=job_info["initiator_party_id"], job_parameters=job_parameters, dsl_parser=dsl_parser)
job_parameters = job_info['runtime_conf_on_party']['job_parameters']
roles = job_info['roles']
cls.initialize_job_tracker(job_id=job_id, role=role, party_id=party_id, job_parameters=job_parameters, roles=roles, is_initiator=is_initiator, dsl_parser=dsl_parser)
JobSaver.create_job(job_info=job_info)
def get_job_table_list():
detect_utils.check_config(
config=request.json, required_arguments=['job_id', 'role', 'party_id'])
jobs = JobSaver.query_job(**request.json)
if jobs:
job = jobs[0]
tables = get_job_all_table(job)
return get_json_result(data=tables)
else:
return get_json_result(retcode=101, retmsg='no find job')
def schedule_waiting_jobs(cls, event):
job_id, initiator_role, initiator_party_id, = event.f_job_id, event.f_initiator_role, event.f_initiator_party_id,
update_status = JobQueue.update_event(job_id=job_id,
initiator_role=initiator_role,
initiator_party_id=initiator_party_id,
job_status=JobStatus.READY)
if not update_status:
schedule_logger(job_id).info(f"job {job_id} may be handled by another scheduler")
return
# apply resource on all party
jobs = JobSaver.query_job(job_id=job_id, role=initiator_role, party_id=initiator_party_id)
if not jobs:
JobQueue.delete_event(job_id=job_id)
return
job = jobs[0]
apply_status_code, federated_response = FederatedScheduler.resource_for_job(job=job, operation_type=ResourceOperation.APPLY)
if apply_status_code == FederatedSchedulingStatusCode.SUCCESS:
cls.start_job(job_id=job_id, initiator_role=initiator_role, initiator_party_id=initiator_party_id)
JobQueue.update_event(job_id=job_id,
initiator_role=initiator_role,
initiator_party_id=initiator_party_id,
job_status=JobStatus.RUNNING)
else:
# rollback resource
rollback_party = {}
failed_party = {}
for dest_role in federated_response.keys():
for dest_party_id in federated_response[dest_role].keys():
retcode = federated_response[dest_role][dest_party_id]["retcode"]
if retcode == 0:
rollback_party[dest_role] = rollback_party.get(dest_role, [])
rollback_party[dest_role].append(dest_party_id)
else:
failed_party[dest_role] = failed_party.get(dest_role, [])
failed_party[dest_role].append(dest_party_id)
schedule_logger(job_id).info("job {} apply resource failed on {}, rollback {}".format(
job_id,
",".join([",".join([f"{_r}:{_p}" for _p in _ps]) for _r, _ps in failed_party.items()]),
",".join([",".join([f"{_r}:{_p}" for _p in _ps]) for _r, _ps in rollback_party.items()]),
))
if rollback_party:
return_status_code, federated_response = FederatedScheduler.resource_for_job(job=job, operation_type=ResourceOperation.RETURN, specific_dest=rollback_party)
if return_status_code != FederatedSchedulingStatusCode.SUCCESS:
schedule_logger(job_id).info(f"job {job_id} return resource failed:\n{federated_response}")
else:
schedule_logger(job_id).info(f"job {job_id} no party should be rollback resource")
if apply_status_code == FederatedSchedulingStatusCode.ERROR:
cls.stop_job(job_id=job_id, role=initiator_role, party_id=initiator_party_id, stop_status=JobStatus.FAILED)
schedule_logger(job_id).info(f"apply resource error, stop job {job_id}")
else:
update_status = JobQueue.update_event(job_id=job_id,
initiator_role=initiator_role,
initiator_party_id=initiator_party_id,
job_status=JobStatus.WAITING)
schedule_logger(job_id).info(f"update job {job_id} status to waiting {update_status}")
def federated_task_status(cls, job_id, task_id, task_version):
tasks_on_all_party = JobSaver.query_task(task_id=task_id,
task_version=task_version)
tasks_party_status = [
task.f_party_status for task in tasks_on_all_party
]
status = cls.calculate_multi_party_task_status(tasks_party_status)
schedule_logger(job_id=job_id).info(
"job {} task {} {} status is {}, calculate by task party status list: {}"
.format(job_id, task_id, task_version, status, tasks_party_status))
return status
