DATAUP-544 API Defined

execution_engine2.spec

@@ -16,6 +16,9 @@
         /* A job id. */
         typedef string job_id;
 
+        /* A job state's job status. */
+        typedef string job_status;
+
         /*
             A structure representing the Execution Engine status
             git_commit - the Git hash of the version of the module.
@@ -229,6 +232,20 @@
             boolean as_admin;
         } BulkRetryParams;
 
+
+        /*
+            batch_job_id: BATCH_ID to retry
+            status_list: job states in ['terminated', 'error'] (valid retry states)
+            as_admin: retry someone else's job in your namespace
+            #TODO: Possibly Add list<JobRequirements> job_requirements;
+        */
+        typedef structure {
+            job_id batch_job_id;
+            list<job_status> status_list;
+            boolean as_admin;
+        } BatchRetryParams;
+
+
         /*
             #TODO write retry parent tests to ensure BOTH the parent_job_id is present, and retry_job_id is present
             #TODO Add retry child that checks the status of the child? to prevent multiple retries
@@ -246,14 +263,17 @@
         */
         funcdef retry_jobs(BulkRetryParams params) returns (list<RetryResult> retry_result) authentication required;
 
-
+        /*
+            Retry a job based on a batch id with a job_state status list ['error', 'terminated']
+            Requires the user to keep track of the job states of the Status enum in the ee2 models file
+            If no status_list is provided, an exception is thrown.
+        */
+        funcdef retry_batch_jobs(BatchRetryParams params) returns (list<RetryResult> retry_result) authentication required;
 
 
         funcdef abandon_children(AbandonChildren params)
             returns (BatchSubmission parent_and_child_ids) authentication required;
 
-
-
         /* EE2Constants Concierge Params are
             request_cpus: int
             request_memory: int in MB
@@ -585,6 +605,29 @@
         */
         funcdef cancel_job(CancelJobParams params) returns () authentication required;
 
+
+        /*
+            batch_job_id: BATCH_ID to cancel
+            status_list: optional filter of either 'terminated' or 'error'. Not setting this results in cancel of both
+            terminated_code: optional terminated code, default to terminated by user
+            as_admin: retry someone else's job in your namespace
+            @optional terminated_code
+        */
+        typedef structure {
+            job_id batch_job_id;
+            list<job_status> status_list;
+            int terminated_code;
+            boolean as_admin;
+        } BatchCancelParams;
+
+        /*
+            Cancels children of a batch job. This results in the status becoming "terminated" with termination_code 0.
+            Valid statuses are ['created', 'estimating', 'queued', 'running']
+            (Requires the user to keep track of the job states of the Status enum in the ee2 models file)
+            If no status_list is provided, an exception is thrown.
+        */
+        funcdef cancel_batch_job(BatchCancelParams params) returns (list<job_id> job_ids) authentication required;
+
         /*
             job_id - id of job running method
             finished - indicates whether job is done (including error/cancel cases) or not

lib/execution_engine2/db/MongoUtil.py

@@ -4,7 +4,8 @@
 import traceback
 from contextlib import contextmanager
 from datetime import datetime
-from typing import Dict, List
+from typing import Dict, List, Union
+
 from bson.objectid import ObjectId
 from mongoengine import connect, connection
 from pymongo import MongoClient, UpdateOne
@@ -15,9 +16,8 @@
     RecordNotFoundException,
     InvalidStatusTransitionException,
 )
-
-from lib.execution_engine2.utils.arg_processing import parse_bool
 from execution_engine2.sdk.EE2Runjob import JobIdPair
+from lib.execution_engine2.utils.arg_processing import parse_bool
 
 
 class MongoUtil:
@@ -223,6 +223,48 @@ def get_job(self, job_id=None, exclude_fields=None) -> Job:
 
         return job
 
+    def get_jobs_with_status(
+        self,
+        job_ids: List[str],
+        status_list: List[str],
+        only_job_ids: bool = False,
+        retried_jobs_allowed=True,
+    ) -> Union[List[Job], List[str]]:
+        if not (job_ids and isinstance(job_ids, list)):
+            raise ValueError("Please provide a non empty list of job ids")
+
+        if not (status_list and isinstance(job_ids, list)):
+            raise ValueError("Please provide a non empty list of job statuses")
+
+        with self.mongo_engine_connection():
+            # TODO: Only seems to be returning other fields as well. Is .only() broken?
+            if retried_jobs_allowed:
+                jobs = Job.objects(id__in=job_ids, status__in=status_list)
+            else:
+                jobs = Job.objects(
+                    id__in=job_ids, status__in=status_list, retry_parent__exists=False
+                )
+
+            if only_job_ids:
+                return [str(job.id) for job in jobs]
+            return jobs
+
+    def eligible_for_retry(self, job: Job):
+        """
+        Checks the job record to see if it has any retry_ids,
+        and if those retry_ids do not contain an ineligble job state
+        :param job:
+        :return:
+        """
+        if job.retry_ids == []:
+            return True
+        eligble_states = [Status.completed.value, Status.error.value]
+        jobs = self.get_jobs(job_ids=job.retry_ids)
+        for job in jobs:
+            if job.status not in eligble_states:
+                return False
+        return True
+
     def get_jobs(
         self, job_ids=None, exclude_fields=None, sort_id_ascending=None
     ) -> List[Job]:

lib/execution_engine2/db/models/models.py

@@ -227,6 +227,10 @@ class Status(Enum):
     A job begins at created, then can either be estimating
     """
 
+    @classmethod
+    def get_status_names(cls):
+        return list(map(lambda x: x.value, cls._member_map_.values()))
+
     created = "created"
     estimating = "estimating"
     queued = "queued"

lib/execution_engine2/exceptions.py

@@ -18,10 +18,22 @@ def __init__(self, msg=None, *args, **kwargs):
         super().__init__(msg or self.__doc__, *args, **kwargs)
 
 
+class InvalidStatusListException(ExecutionEngineValueError):
+    """Invalid job state status provided"""
+
+
+class BatchTerminationException(ExecutionEngineException):
+    """No jobs to terminate"""
+
+
 class IncorrectParamsException(ExecutionEngineValueError):
     """Wrong parameters were provided"""
 
 
+class NotBatchJobException(ExecutionEngineValueError):
+    """Requested job is not a batch job"""
+
+
 class InvalidParameterForBatch(ExecutionEngineValueError):
     """Workspace ids are not allowed in RunJobParams in Batch Mode"""