1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: Apache-2.0
5 from __future__ import division
6 from builtins import next
7 from builtins import object
8 from builtins import str
9 from future.utils import viewvalues
19 from functools import partial
22 from cwltool.errors import WorkflowException
23 import cwltool.workflow
24 from schema_salad.sourceline import SourceLine
25 import schema_salad.validate as validate
29 from arvados.keep import KeepClient
30 from arvados.errors import ApiError
32 import arvados_cwl.util
33 from .arvcontainer import RunnerContainer
34 from .arvjob import RunnerJob, RunnerTemplate
35 from .runner import Runner, upload_docker, upload_job_order, upload_workflow_deps
36 from .arvtool import ArvadosCommandTool, validate_cluster_target, ArvadosExpressionTool
37 from .arvworkflow import ArvadosWorkflow, upload_workflow
38 from .fsaccess import CollectionFsAccess, CollectionFetcher, collectionResolver, CollectionCache, pdh_size
39 from .perf import Perf
40 from .pathmapper import NoFollowPathMapper
41 from .task_queue import TaskQueue
42 from .context import ArvLoadingContext, ArvRuntimeContext
43 from ._version import __version__
45 from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema
46 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, get_listing, visit_class
47 from cwltool.command_line_tool import compute_checksums
48 from cwltool.load_tool import load_tool
50 logger = logging.getLogger('arvados.cwl-runner')
51 metrics = logging.getLogger('arvados.cwl-runner.metrics')
53 DEFAULT_PRIORITY = 500
55 class RuntimeStatusLoggingHandler(logging.Handler):
57 Intercepts logging calls and report them as runtime statuses on runner
60 def __init__(self, runtime_status_update_func):
61 super(RuntimeStatusLoggingHandler, self).__init__()
62 self.runtime_status_update = runtime_status_update_func
63 self.updatingRuntimeStatus = False
65 def emit(self, record):
67 if record.levelno >= logging.ERROR:
69 elif record.levelno >= logging.WARNING:
71 if kind is not None and self.updatingRuntimeStatus is not True:
72 self.updatingRuntimeStatus = True
74 log_msg = record.getMessage()
76 # If the logged message is multi-line, use its first line as status
77 # and the rest as detail.
78 status, detail = log_msg.split('\n', 1)
79 self.runtime_status_update(
81 "%s: %s" % (record.name, status),
85 self.runtime_status_update(
87 "%s: %s" % (record.name, record.getMessage())
90 self.updatingRuntimeStatus = False
93 class ArvCwlExecutor(object):
94 """Execute a CWL tool or workflow, submit work (using either jobs or
95 containers API), wait for them to complete, and report output.
99 def __init__(self, api_client,
106 arvargs = argparse.Namespace()
107 arvargs.work_api = None
108 arvargs.output_name = None
109 arvargs.output_tags = None
110 arvargs.thread_count = 1
111 arvargs.collection_cache_size = None
113 self.api = api_client
115 self.workflow_eval_lock = threading.Condition(threading.RLock())
116 self.final_output = None
117 self.final_status = None
118 self.num_retries = num_retries
120 self.stop_polling = threading.Event()
123 self.final_output_collection = None
124 self.output_name = arvargs.output_name
125 self.output_tags = arvargs.output_tags
126 self.project_uuid = None
127 self.intermediate_output_ttl = 0
128 self.intermediate_output_collections = []
129 self.trash_intermediate = False
130 self.thread_count = arvargs.thread_count
131 self.poll_interval = 12
132 self.loadingContext = None
133 self.should_estimate_cache_size = True
135 if keep_client is not None:
136 self.keep_client = keep_client
138 self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
140 if arvargs.collection_cache_size:
141 collection_cache_size = arvargs.collection_cache_size*1024*1024
142 self.should_estimate_cache_size = False
144 collection_cache_size = 256*1024*1024
146 self.collection_cache = CollectionCache(self.api, self.keep_client, self.num_retries,
147 cap=collection_cache_size)
149 self.fetcher_constructor = partial(CollectionFetcher,
151 fs_access=CollectionFsAccess("", collection_cache=self.collection_cache),
152 num_retries=self.num_retries)
155 expected_api = ["jobs", "containers"]
156 for api in expected_api:
158 methods = self.api._rootDesc.get('resources')[api]['methods']
159 if ('httpMethod' in methods['create'] and
160 (arvargs.work_api == api or arvargs.work_api is None)):
166 if not self.work_api:
167 if arvargs.work_api is None:
168 raise Exception("No supported APIs")
170 raise Exception("Unsupported API '%s', expected one of %s" % (arvargs.work_api, expected_api))
172 if self.work_api == "jobs":
174 *******************************
175 Using the deprecated 'jobs' API.
177 To get rid of this warning:
179 Users: read about migrating at
180 http://doc.arvados.org/user/cwl/cwl-style.html#migrate
181 and use the option --api=containers
183 Admins: configure the cluster to disable the 'jobs' API as described at:
184 http://doc.arvados.org/install/install-api-server.html#disable_api_methods
185 *******************************""")
187 self.loadingContext = ArvLoadingContext(vars(arvargs))
188 self.loadingContext.fetcher_constructor = self.fetcher_constructor
189 self.loadingContext.resolver = partial(collectionResolver, self.api, num_retries=self.num_retries)
190 self.loadingContext.construct_tool_object = self.arv_make_tool
192 # Add a custom logging handler to the root logger for runtime status reporting
193 # if running inside a container
194 if arvados_cwl.util.get_current_container(self.api, self.num_retries, logger):
195 root_logger = logging.getLogger('')
197 # Remove existing RuntimeStatusLoggingHandlers if they exist
198 handlers = [h for h in root_logger.handlers if not isinstance(h, RuntimeStatusLoggingHandler)]
199 root_logger.handlers = handlers
201 handler = RuntimeStatusLoggingHandler(self.runtime_status_update)
202 root_logger.addHandler(handler)
204 self.runtimeContext = ArvRuntimeContext(vars(arvargs))
205 self.runtimeContext.make_fs_access = partial(CollectionFsAccess,
206 collection_cache=self.collection_cache)
208 validate_cluster_target(self, self.runtimeContext)
211 def arv_make_tool(self, toolpath_object, loadingContext):
212 if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
213 return ArvadosCommandTool(self, toolpath_object, loadingContext)
214 elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
215 return ArvadosWorkflow(self, toolpath_object, loadingContext)
216 elif "class" in toolpath_object and toolpath_object["class"] == "ExpressionTool":
217 return ArvadosExpressionTool(self, toolpath_object, loadingContext)
219 raise Exception("Unknown tool %s" % toolpath_object.get("class"))
221 def output_callback(self, out, processStatus):
222 with self.workflow_eval_lock:
223 if processStatus == "success":
224 logger.info("Overall process status is %s", processStatus)
227 logger.error("Overall process status is %s", processStatus)
230 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
231 body={"state": state}).execute(num_retries=self.num_retries)
232 self.final_status = processStatus
233 self.final_output = out
234 self.workflow_eval_lock.notifyAll()
237 def start_run(self, runnable, runtimeContext):
238 self.task_queue.add(partial(runnable.run, runtimeContext),
239 self.workflow_eval_lock, self.stop_polling)
241 def process_submitted(self, container):
242 with self.workflow_eval_lock:
243 self.processes[container.uuid] = container
245 def process_done(self, uuid, record):
246 with self.workflow_eval_lock:
247 j = self.processes[uuid]
248 logger.info("%s %s is %s", self.label(j), uuid, record["state"])
249 self.task_queue.add(partial(j.done, record),
250 self.workflow_eval_lock, self.stop_polling)
251 del self.processes[uuid]
253 def runtime_status_update(self, kind, message, detail=None):
255 Updates the runtime_status field on the runner container.
256 Called when there's a need to report errors, warnings or just
257 activity statuses, for example in the RuntimeStatusLoggingHandler.
259 with self.workflow_eval_lock:
260 current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
263 runtime_status = current.get('runtime_status', {})
264 # In case of status being an error, only report the first one.
266 if not runtime_status.get('error'):
267 runtime_status.update({
270 if detail is not None:
271 runtime_status.update({
272 'errorDetail': detail
274 # Further errors are only mentioned as a count.
276 # Get anything before an optional 'and N more' string.
278 error_msg = re.match(
279 r'^(.*?)(?=\s*\(and \d+ more\)|$)', runtime_status.get('error')).groups()[0]
280 more_failures = re.match(
281 r'.*\(and (\d+) more\)', runtime_status.get('error'))
283 # Ignore tests stubbing errors
286 failure_qty = int(more_failures.groups()[0])
287 runtime_status.update({
288 'error': "%s (and %d more)" % (error_msg, failure_qty+1)
291 runtime_status.update({
292 'error': "%s (and 1 more)" % error_msg
294 elif kind in ['warning', 'activity']:
295 # Record the last warning/activity status without regard of
296 # previous occurences.
297 runtime_status.update({
300 if detail is not None:
301 runtime_status.update({
302 kind+"Detail": detail
305 # Ignore any other status kind
308 self.api.containers().update(uuid=current['uuid'],
310 'runtime_status': runtime_status,
311 }).execute(num_retries=self.num_retries)
312 except Exception as e:
313 logger.info("Couldn't update runtime_status: %s", e)
315 def wrapped_callback(self, cb, obj, st):
316 with self.workflow_eval_lock:
318 self.workflow_eval_lock.notifyAll()
320 def get_wrapped_callback(self, cb):
321 return partial(self.wrapped_callback, cb)
323 def on_message(self, event):
324 if event.get("object_uuid") in self.processes and event["event_type"] == "update":
325 uuid = event["object_uuid"]
326 if event["properties"]["new_attributes"]["state"] == "Running":
327 with self.workflow_eval_lock:
328 j = self.processes[uuid]
329 if j.running is False:
331 j.update_pipeline_component(event["properties"]["new_attributes"])
332 logger.info("%s %s is Running", self.label(j), uuid)
333 elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled", "Final"):
334 self.process_done(uuid, event["properties"]["new_attributes"])
336 def label(self, obj):
337 return "[%s %s]" % (self.work_api[0:-1], obj.name)
339 def poll_states(self):
340 """Poll status of jobs or containers listed in the processes dict.
342 Runs in a separate thread.
346 remain_wait = self.poll_interval
349 self.stop_polling.wait(remain_wait)
350 if self.stop_polling.is_set():
352 with self.workflow_eval_lock:
353 keys = list(self.processes)
355 remain_wait = self.poll_interval
358 begin_poll = time.time()
359 if self.work_api == "containers":
360 table = self.poll_api.container_requests()
361 elif self.work_api == "jobs":
362 table = self.poll_api.jobs()
364 pageSize = self.poll_api._rootDesc.get('maxItemsPerResponse', 1000)
367 page = keys[:pageSize]
368 keys = keys[pageSize:]
370 proc_states = table.list(filters=[["uuid", "in", page]]).execute(num_retries=self.num_retries)
372 logger.exception("Error checking states on API server: %s")
373 remain_wait = self.poll_interval
376 for p in proc_states["items"]:
378 "object_uuid": p["uuid"],
379 "event_type": "update",
384 finish_poll = time.time()
385 remain_wait = self.poll_interval - (finish_poll - begin_poll)
387 logger.exception("Fatal error in state polling thread.")
388 with self.workflow_eval_lock:
389 self.processes.clear()
390 self.workflow_eval_lock.notifyAll()
392 self.stop_polling.set()
394 def add_intermediate_output(self, uuid):
396 self.intermediate_output_collections.append(uuid)
398 def trash_intermediate_output(self):
399 logger.info("Cleaning up intermediate output collections")
400 for i in self.intermediate_output_collections:
402 self.api.collections().delete(uuid=i).execute(num_retries=self.num_retries)
404 logger.warning("Failed to delete intermediate output: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
405 except (KeyboardInterrupt, SystemExit):
408 def check_features(self, obj):
409 if isinstance(obj, dict):
410 if obj.get("writable") and self.work_api != "containers":
411 raise SourceLine(obj, "writable", UnsupportedRequirement).makeError("InitialWorkDir feature 'writable: true' not supported with --api=jobs")
412 if obj.get("class") == "DockerRequirement":
413 if obj.get("dockerOutputDirectory"):
414 if self.work_api != "containers":
415 raise SourceLine(obj, "dockerOutputDirectory", UnsupportedRequirement).makeError(
416 "Option 'dockerOutputDirectory' of DockerRequirement not supported with --api=jobs.")
417 if not obj.get("dockerOutputDirectory").startswith('/'):
418 raise SourceLine(obj, "dockerOutputDirectory", validate.ValidationException).makeError(
419 "Option 'dockerOutputDirectory' must be an absolute path.")
420 if obj.get("class") == "http://commonwl.org/cwltool#Secrets" and self.work_api != "containers":
421 raise SourceLine(obj, "class", UnsupportedRequirement).makeError("Secrets not supported with --api=jobs")
422 for v in viewvalues(obj):
423 self.check_features(v)
424 elif isinstance(obj, list):
425 for i,v in enumerate(obj):
426 with SourceLine(obj, i, UnsupportedRequirement, logger.isEnabledFor(logging.DEBUG)):
427 self.check_features(v)
429 def make_output_collection(self, name, storage_classes, tagsString, outputObj):
430 outputObj = copy.deepcopy(outputObj)
433 def capture(fileobj):
434 files.append(fileobj)
436 adjustDirObjs(outputObj, capture)
437 adjustFileObjs(outputObj, capture)
439 generatemapper = NoFollowPathMapper(files, "", "", separateDirs=False)
441 final = arvados.collection.Collection(api_client=self.api,
442 keep_client=self.keep_client,
443 num_retries=self.num_retries)
445 for k,v in generatemapper.items():
446 if v.type == "Directory" and v.resolved.startswith("_:"):
448 if v.type == "CreateFile" and (k.startswith("_:") or v.resolved.startswith("_:")):
449 with final.open(v.target, "wb") as f:
450 f.write(v.resolved.encode("utf-8"))
453 if not v.resolved.startswith("keep:"):
454 raise Exception("Output source is not in keep or a literal")
455 sp = v.resolved.split("/")
456 srccollection = sp[0][5:]
458 reader = self.collection_cache.get(srccollection)
459 srcpath = "/".join(sp[1:]) if len(sp) > 1 else "."
460 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
461 except arvados.errors.ArgumentError as e:
462 logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e)
465 logger.error("While preparing output collection: %s", e)
468 def rewrite(fileobj):
469 fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
470 for k in ("listing", "contents", "nameext", "nameroot", "dirname"):
474 adjustDirObjs(outputObj, rewrite)
475 adjustFileObjs(outputObj, rewrite)
477 with final.open("cwl.output.json", "w") as f:
478 res = str(json.dumps(outputObj, sort_keys=True, indent=4, separators=(',',': '), ensure_ascii=False))
481 final.save_new(name=name, owner_uuid=self.project_uuid, storage_classes=storage_classes, ensure_unique_name=True)
483 logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
484 final.api_response()["name"],
485 final.manifest_locator())
487 final_uuid = final.manifest_locator()
488 tags = tagsString.split(',')
490 self.api.links().create(body={
491 "head_uuid": final_uuid, "link_class": "tag", "name": tag
492 }).execute(num_retries=self.num_retries)
494 def finalcollection(fileobj):
495 fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"])
497 adjustDirObjs(outputObj, finalcollection)
498 adjustFileObjs(outputObj, finalcollection)
500 return (outputObj, final)
502 def set_crunch_output(self):
503 if self.work_api == "containers":
504 current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
508 self.api.containers().update(uuid=current['uuid'],
510 'output': self.final_output_collection.portable_data_hash(),
511 }).execute(num_retries=self.num_retries)
512 self.api.collections().update(uuid=self.final_output_collection.manifest_locator(),
515 }).execute(num_retries=self.num_retries)
517 logger.exception("Setting container output")
519 elif self.work_api == "jobs" and "TASK_UUID" in os.environ:
520 self.api.job_tasks().update(uuid=os.environ["TASK_UUID"],
522 'output': self.final_output_collection.portable_data_hash(),
523 'success': self.final_status == "success",
525 }).execute(num_retries=self.num_retries)
527 def arv_executor(self, tool, job_order, runtimeContext, logger=None):
528 self.debug = runtimeContext.debug
530 tool.visit(self.check_features)
532 self.project_uuid = runtimeContext.project_uuid
534 self.fs_access = runtimeContext.make_fs_access(runtimeContext.basedir)
535 self.secret_store = runtimeContext.secret_store
537 self.trash_intermediate = runtimeContext.trash_intermediate
538 if self.trash_intermediate and self.work_api != "containers":
539 raise Exception("--trash-intermediate is only supported with --api=containers.")
541 self.intermediate_output_ttl = runtimeContext.intermediate_output_ttl
542 if self.intermediate_output_ttl and self.work_api != "containers":
543 raise Exception("--intermediate-output-ttl is only supported with --api=containers.")
544 if self.intermediate_output_ttl < 0:
545 raise Exception("Invalid value %d for --intermediate-output-ttl, cannot be less than zero" % self.intermediate_output_ttl)
547 if runtimeContext.submit_request_uuid and self.work_api != "containers":
548 raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api))
550 if not runtimeContext.name:
551 runtimeContext.name = self.name = tool.tool.get("label") or tool.metadata.get("label") or os.path.basename(tool.tool["id"])
553 submitting = (runtimeContext.update_workflow or
554 runtimeContext.create_workflow or
555 (runtimeContext.submit and not
556 (tool.tool["class"] == "CommandLineTool" and
557 runtimeContext.wait and
558 not runtimeContext.always_submit_runner)))
560 loadingContext = self.loadingContext.copy()
561 loadingContext.do_validate = False
562 loadingContext.do_update = False
564 # Document may have been auto-updated. Reload the original
565 # document with updating disabled because we want to
566 # submit the original document, not the auto-updated one.
567 tool = load_tool(tool.tool["id"], loadingContext)
569 # Upload direct dependencies of workflow steps, get back mapping of files to keep references.
570 # Also uploads docker images.
571 merged_map = upload_workflow_deps(self, tool)
573 # Recreate process object (ArvadosWorkflow or
574 # ArvadosCommandTool) because tool document may have been
575 # updated by upload_workflow_deps in ways that modify
576 # inheritance of hints or requirements.
577 tool = load_tool(tool.tool, loadingContext)
579 # Upload local file references in the job order.
580 job_order = upload_job_order(self, "%s input" % runtimeContext.name,
583 existing_uuid = runtimeContext.update_workflow
584 if existing_uuid or runtimeContext.create_workflow:
585 # Create a pipeline template or workflow record and exit.
586 if self.work_api == "jobs":
587 tmpl = RunnerTemplate(self, tool, job_order,
588 runtimeContext.enable_reuse,
590 submit_runner_ram=runtimeContext.submit_runner_ram,
591 name=runtimeContext.name,
592 merged_map=merged_map,
593 loadingContext=loadingContext)
595 # cwltool.main will write our return value to stdout.
596 return (tmpl.uuid, "success")
597 elif self.work_api == "containers":
598 return (upload_workflow(self, tool, job_order,
601 submit_runner_ram=runtimeContext.submit_runner_ram,
602 name=runtimeContext.name,
603 merged_map=merged_map),
606 self.ignore_docker_for_reuse = runtimeContext.ignore_docker_for_reuse
607 self.eval_timeout = runtimeContext.eval_timeout
609 runtimeContext = runtimeContext.copy()
610 runtimeContext.use_container = True
611 runtimeContext.tmpdir_prefix = "tmp"
612 runtimeContext.work_api = self.work_api
614 if self.work_api == "containers":
615 if self.ignore_docker_for_reuse:
616 raise Exception("--ignore-docker-for-reuse not supported with containers API.")
617 runtimeContext.outdir = "/var/spool/cwl"
618 runtimeContext.docker_outdir = "/var/spool/cwl"
619 runtimeContext.tmpdir = "/tmp"
620 runtimeContext.docker_tmpdir = "/tmp"
621 elif self.work_api == "jobs":
622 if runtimeContext.priority != DEFAULT_PRIORITY:
623 raise Exception("--priority not implemented for jobs API.")
624 runtimeContext.outdir = "$(task.outdir)"
625 runtimeContext.docker_outdir = "$(task.outdir)"
626 runtimeContext.tmpdir = "$(task.tmpdir)"
628 if runtimeContext.priority < 1 or runtimeContext.priority > 1000:
629 raise Exception("--priority must be in the range 1..1000.")
631 if self.should_estimate_cache_size:
634 def estimate_collection_cache(obj):
635 if obj.get("location", "").startswith("keep:"):
636 m = pdh_size.match(obj["location"][5:])
637 if m and m.group(1) not in visited:
638 visited.add(m.group(1))
639 estimated_size[0] += int(m.group(2))
640 visit_class(job_order, ("File", "Directory"), estimate_collection_cache)
641 runtimeContext.collection_cache_size = max(((estimated_size[0]*192) // (1024*1024))+1, 256)
642 self.collection_cache.set_cap(runtimeContext.collection_cache_size*1024*1024)
644 logger.info("Using collection cache size %s MiB", runtimeContext.collection_cache_size)
647 if runtimeContext.submit:
648 # Submit a runner job to run the workflow for us.
649 if self.work_api == "containers":
650 loadingContext.loader = tool.doc_loader
651 loadingContext.avsc_names = tool.doc_schema
652 if tool.tool["class"] == "CommandLineTool" and runtimeContext.wait and (not runtimeContext.always_submit_runner):
653 runtimeContext.runnerjob = tool.tool["id"]
655 tool = RunnerContainer(self, tool, loadingContext, runtimeContext.enable_reuse,
658 submit_runner_ram=runtimeContext.submit_runner_ram,
659 name=runtimeContext.name,
660 on_error=runtimeContext.on_error,
661 submit_runner_image=runtimeContext.submit_runner_image,
662 intermediate_output_ttl=runtimeContext.intermediate_output_ttl,
663 merged_map=merged_map,
664 priority=runtimeContext.priority,
665 secret_store=self.secret_store,
666 collection_cache_size=runtimeContext.collection_cache_size,
667 collection_cache_is_default=self.should_estimate_cache_size)
668 elif self.work_api == "jobs":
669 tool = RunnerJob(self, tool, loadingContext, runtimeContext.enable_reuse,
672 submit_runner_ram=runtimeContext.submit_runner_ram,
673 name=runtimeContext.name,
674 on_error=runtimeContext.on_error,
675 submit_runner_image=runtimeContext.submit_runner_image,
676 merged_map=merged_map)
677 elif runtimeContext.cwl_runner_job is None and self.work_api == "jobs":
678 # Create pipeline for local run
679 self.pipeline = self.api.pipeline_instances().create(
681 "owner_uuid": self.project_uuid,
682 "name": runtimeContext.name if runtimeContext.name else shortname(tool.tool["id"]),
684 "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
685 logger.info("Pipeline instance %s", self.pipeline["uuid"])
687 if runtimeContext.cwl_runner_job is not None:
688 self.uuid = runtimeContext.cwl_runner_job.get('uuid')
690 jobiter = tool.job(job_order,
691 self.output_callback,
694 if runtimeContext.submit and not runtimeContext.wait:
695 runnerjob = next(jobiter)
696 runnerjob.run(runtimeContext)
697 return (runnerjob.uuid, "success")
699 current_container = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
700 if current_container:
701 logger.info("Running inside container %s", current_container.get("uuid"))
703 self.poll_api = arvados.api('v1', timeout=runtimeContext.http_timeout)
704 self.polling_thread = threading.Thread(target=self.poll_states)
705 self.polling_thread.start()
707 self.task_queue = TaskQueue(self.workflow_eval_lock, self.thread_count)
710 self.workflow_eval_lock.acquire()
712 # Holds the lock while this code runs and releases it when
713 # it is safe to do so in self.workflow_eval_lock.wait(),
714 # at which point on_message can update job state and
715 # process output callbacks.
717 loopperf = Perf(metrics, "jobiter")
719 for runnable in jobiter:
722 if self.stop_polling.is_set():
725 if self.task_queue.error is not None:
726 raise self.task_queue.error
729 with Perf(metrics, "run"):
730 self.start_run(runnable, runtimeContext)
732 if (self.task_queue.in_flight + len(self.processes)) > 0:
733 self.workflow_eval_lock.wait(3)
735 logger.error("Workflow is deadlocked, no runnable processes and not waiting on any pending processes.")
738 if self.stop_polling.is_set():
744 while (self.task_queue.in_flight + len(self.processes)) > 0:
745 if self.task_queue.error is not None:
746 raise self.task_queue.error
747 self.workflow_eval_lock.wait(3)
749 except UnsupportedRequirement:
752 if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit:
753 logger.error("Interrupted, workflow will be cancelled")
754 elif isinstance(sys.exc_info()[1], WorkflowException):
755 logger.error("Workflow execution failed:\n%s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
757 logger.exception("Workflow execution failed")
760 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
761 body={"state": "Failed"}).execute(num_retries=self.num_retries)
762 if runtimeContext.submit and isinstance(tool, Runner):
764 if runnerjob.uuid and self.work_api == "containers":
765 self.api.container_requests().update(uuid=runnerjob.uuid,
766 body={"priority": "0"}).execute(num_retries=self.num_retries)
768 self.workflow_eval_lock.release()
769 self.task_queue.drain()
770 self.stop_polling.set()
771 self.polling_thread.join()
772 self.task_queue.join()
774 if self.final_status == "UnsupportedRequirement":
775 raise UnsupportedRequirement("Check log for details.")
777 if self.final_output is None:
778 raise WorkflowException("Workflow did not return a result.")
780 if runtimeContext.submit and isinstance(tool, Runner):
781 logger.info("Final output collection %s", tool.final_output)
783 if self.output_name is None:
784 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
785 if self.output_tags is None:
786 self.output_tags = ""
788 storage_classes = runtimeContext.storage_classes.strip().split(",")
789 self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, storage_classes, self.output_tags, self.final_output)
790 self.set_crunch_output()
792 if runtimeContext.compute_checksum:
793 adjustDirObjs(self.final_output, partial(get_listing, self.fs_access))
794 adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
796 if self.trash_intermediate and self.final_status == "success":
797 self.trash_intermediate_output()
799 return (self.final_output, self.final_status)