3 # Implement cwl-runner interface for submitting and running jobs on Arvados.
8 import arvados.commands.keepdocker
9 import arvados.commands.run
10 import arvados.collection
12 import cwltool.draft2tool
13 import cwltool.workflow
15 from cwltool.process import shortname
16 from cwltool.errors import WorkflowException
26 import pkg_resources # part of setuptools
28 from cwltool.process import get_feature, adjustFiles, scandeps
29 from arvados.api import OrderedJsonModel
31 logger = logging.getLogger('arvados.cwl-runner')
32 logger.setLevel(logging.INFO)
34 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
35 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
36 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
39 def arv_docker_get_image(api_client, dockerRequirement, pull_image, project_uuid):
40 """Check if a Docker image is available in Keep, if not, upload it using arv-keepdocker."""
42 if "dockerImageId" not in dockerRequirement and "dockerPull" in dockerRequirement:
43 dockerRequirement["dockerImageId"] = dockerRequirement["dockerPull"]
45 sp = dockerRequirement["dockerImageId"].split(":")
47 image_tag = sp[1] if len(sp) > 1 else None
49 images = arvados.commands.keepdocker.list_images_in_arv(api_client, 3,
50 image_name=image_name,
54 imageId = cwltool.docker.get_image(dockerRequirement, pull_image)
55 args = ["--project-uuid="+project_uuid, image_name]
57 args.append(image_tag)
58 logger.info("Uploading Docker image %s", ":".join(args[1:]))
59 arvados.commands.keepdocker.main(args, stdout=sys.stderr)
61 return dockerRequirement["dockerImageId"]
64 class CollectionFsAccess(cwltool.process.StdFsAccess):
65 """Implement the cwltool FsAccess interface for Arvados Collections."""
67 def __init__(self, basedir):
69 self.basedir = basedir
71 def get_collection(self, path):
73 if p[0].startswith("keep:") and arvados.util.keep_locator_pattern.match(p[0][5:]):
75 if pdh not in self.collections:
76 self.collections[pdh] = arvados.collection.CollectionReader(pdh)
77 return (self.collections[pdh], "/".join(p[1:]))
81 def _match(self, collection, patternsegments, parent):
82 if not patternsegments:
85 if not isinstance(collection, arvados.collection.RichCollectionBase):
89 # iterate over the files and subcollections in 'collection'
90 for filename in collection:
91 if patternsegments[0] == '.':
92 # Pattern contains something like "./foo" so just shift
94 ret.extend(self._match(collection, patternsegments[1:], parent))
95 elif fnmatch.fnmatch(filename, patternsegments[0]):
96 cur = os.path.join(parent, filename)
97 if len(patternsegments) == 1:
100 ret.extend(self._match(collection[filename], patternsegments[1:], cur))
103 def glob(self, pattern):
104 collection, rest = self.get_collection(pattern)
105 patternsegments = rest.split("/")
106 return self._match(collection, patternsegments, "keep:" + collection.manifest_locator())
108 def open(self, fn, mode):
109 collection, rest = self.get_collection(fn)
111 return collection.open(rest, mode)
113 return open(self._abs(fn), mode)
115 def exists(self, fn):
116 collection, rest = self.get_collection(fn)
118 return collection.exists(rest)
120 return os.path.exists(self._abs(fn))
122 class ArvadosJob(object):
123 """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
125 def __init__(self, runner):
126 self.arvrunner = runner
129 def run(self, dry_run=False, pull_image=True, **kwargs):
130 script_parameters = {
131 "command": self.command_line
133 runtime_constraints = {}
135 if self.generatefiles:
136 vwd = arvados.collection.Collection()
137 script_parameters["task.vwd"] = {}
138 for t in self.generatefiles:
139 if isinstance(self.generatefiles[t], dict):
140 src, rest = self.arvrunner.fs_access.get_collection(self.generatefiles[t]["path"].replace("$(task.keep)/", "keep:"))
141 vwd.copy(rest, t, source_collection=src)
143 with vwd.open(t, "w") as f:
144 f.write(self.generatefiles[t])
146 for t in self.generatefiles:
147 script_parameters["task.vwd"][t] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), t)
149 script_parameters["task.env"] = {"TMPDIR": "$(task.tmpdir)"}
151 script_parameters["task.env"].update(self.environment)
154 script_parameters["task.stdin"] = self.pathmapper.mapper(self.stdin)[1]
157 script_parameters["task.stdout"] = self.stdout
159 (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
160 if docker_req and kwargs.get("use_container") is not False:
161 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
163 runtime_constraints["docker_image"] = "arvados/jobs"
165 resources = self.builder.resources
166 if resources is not None:
167 runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
168 runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
169 runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
171 filters = [["repository", "=", "arvados"],
172 ["script", "=", "crunchrunner"],
173 ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
174 if not self.arvrunner.ignore_docker_for_reuse:
175 filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
178 response = self.arvrunner.api.jobs().create(
180 "owner_uuid": self.arvrunner.project_uuid,
181 "script": "crunchrunner",
182 "repository": "arvados",
183 "script_version": "master",
184 "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
185 "script_parameters": {"tasks": [script_parameters]},
186 "runtime_constraints": runtime_constraints
189 find_or_create=kwargs.get("enable_reuse", True)
190 ).execute(num_retries=self.arvrunner.num_retries)
192 self.arvrunner.jobs[response["uuid"]] = self
194 self.arvrunner.pipeline["components"][self.name] = {"job": response}
195 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
197 "components": self.arvrunner.pipeline["components"]
198 }).execute(num_retries=self.arvrunner.num_retries)
200 logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
202 if response["state"] in ("Complete", "Failed", "Cancelled"):
204 except Exception as e:
205 logger.error("Got error %s" % str(e))
206 self.output_callback({}, "permanentFail")
208 def update_pipeline_component(self, record):
209 self.arvrunner.pipeline["components"][self.name] = {"job": record}
210 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
212 "components": self.arvrunner.pipeline["components"]
213 }).execute(num_retries=self.arvrunner.num_retries)
215 def done(self, record):
217 self.update_pipeline_component(record)
222 if record["state"] == "Complete":
223 processStatus = "success"
225 processStatus = "permanentFail"
230 logc = arvados.collection.Collection(record["log"])
231 log = logc.open(logc.keys()[0])
236 # Determine the tmpdir, outdir and keepdir paths from
237 # the job run. Unfortunately, we can't take the first
238 # values we find (which are expected to be near the
239 # top) and stop scanning because if the node fails and
240 # the job restarts on a different node these values
241 # will different runs, and we need to know about the
242 # final run that actually produced output.
244 g = tmpdirre.match(l)
247 g = outdirre.match(l)
254 colname = "Output %s of %s" % (record["output"][0:7], self.name)
256 # check if collection already exists with same owner, name and content
257 collection_exists = self.arvrunner.api.collections().list(
258 filters=[["owner_uuid", "=", self.arvrunner.project_uuid],
259 ['portable_data_hash', '=', record["output"]],
260 ["name", "=", colname]]
261 ).execute(num_retries=self.arvrunner.num_retries)
263 if not collection_exists["items"]:
264 # Create a collection located in the same project as the
265 # pipeline with the contents of the output.
266 # First, get output record.
267 collections = self.arvrunner.api.collections().list(
269 filters=[['portable_data_hash', '=', record["output"]]],
270 select=["manifest_text"]
271 ).execute(num_retries=self.arvrunner.num_retries)
273 if not collections["items"]:
274 raise WorkflowException(
275 "Job output '%s' cannot be found on API server" % (
278 # Create new collection in the parent project
279 # with the output contents.
280 self.arvrunner.api.collections().create(body={
281 "owner_uuid": self.arvrunner.project_uuid,
283 "portable_data_hash": record["output"],
284 "manifest_text": collections["items"][0]["manifest_text"]
285 }, ensure_unique_name=True).execute(
286 num_retries=self.arvrunner.num_retries)
288 self.builder.outdir = outdir
289 self.builder.pathmapper.keepdir = keepdir
290 outputs = self.collect_outputs("keep:" + record["output"])
291 except WorkflowException as e:
292 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
293 processStatus = "permanentFail"
294 except Exception as e:
295 logger.exception("Got unknown exception while collecting job outputs:")
296 processStatus = "permanentFail"
298 self.output_callback(outputs, processStatus)
300 del self.arvrunner.jobs[record["uuid"]]
303 class RunnerJob(object):
304 """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
306 def __init__(self, runner, tool, job_order, enable_reuse):
307 self.arvrunner = runner
309 self.job_order = job_order
311 self.enable_reuse = enable_reuse
313 def update_pipeline_component(self, record):
316 def upload_docker(self, tool):
317 if isinstance(tool, cwltool.draft2tool.CommandLineTool):
318 (docker_req, docker_is_req) = get_feature(tool, "DockerRequirement")
320 arv_docker_get_image(self.arvrunner.api, docker_req, True, self.arvrunner.project_uuid)
321 elif isinstance(tool, cwltool.workflow.Workflow):
323 self.upload_docker(s.embedded_tool)
325 def run(self, dry_run=False, pull_image=True, **kwargs):
326 self.upload_docker(self.tool)
328 workflowfiles = set()
330 workflowfiles.add(self.tool.tool["id"])
332 self.name = os.path.basename(self.tool.tool["id"])
334 def visitFiles(files, path):
338 document_loader, _, _ = cwltool.process.get_schema()
340 return document_loader.resolve_ref(u, base_url=b)[0]
342 sc = scandeps("", self.tool.tool,
343 set(("$import", "run")),
344 set(("$include", "$schemas", "path")),
346 adjustFiles(sc, functools.partial(visitFiles, workflowfiles))
347 adjustFiles(self.job_order, functools.partial(visitFiles, jobfiles))
349 workflowmapper = ArvPathMapper(self.arvrunner, workflowfiles, "",
355 jobmapper = ArvPathMapper(self.arvrunner, jobfiles, "",
358 name=os.path.basename(self.job_order.get("id", "#")),
361 adjustFiles(self.job_order, lambda p: jobmapper.mapper(p)[1])
363 if "id" in self.job_order:
364 del self.job_order["id"]
366 self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"])[1]
368 response = self.arvrunner.api.jobs().create(body={
369 "owner_uuid": self.arvrunner.project_uuid,
370 "script": "cwl-runner",
371 "script_version": "master",
372 "repository": "arvados",
373 "script_parameters": self.job_order,
374 "runtime_constraints": {
375 "docker_image": "arvados/jobs"
377 }, find_or_create=self.enable_reuse).execute(num_retries=self.arvrunner.num_retries)
379 self.arvrunner.jobs[response["uuid"]] = self
381 logger.info("Submitted job %s", response["uuid"])
383 if response["state"] in ("Complete", "Failed", "Cancelled"):
386 def done(self, record):
387 if record["state"] == "Complete":
388 processStatus = "success"
390 processStatus = "permanentFail"
395 outc = arvados.collection.Collection(record["output"])
396 with outc.open("cwl.output.json") as f:
397 outputs = json.load(f)
398 except Exception as e:
399 logger.error("While getting final output object: %s", e)
400 self.arvrunner.output_callback(outputs, processStatus)
402 del self.arvrunner.jobs[record["uuid"]]
404 class ArvPathMapper(cwltool.pathmapper.PathMapper):
405 """Convert container-local paths to and from Keep collection ids."""
407 def __init__(self, arvrunner, referenced_files, basedir,
408 collection_pattern, file_pattern, name=None, **kwargs):
409 self._pathmap = arvrunner.get_uploaded()
412 pdh_path = re.compile(r'^keep:[0-9a-f]{32}\+\d+/.+')
414 for src in referenced_files:
415 if isinstance(src, basestring) and pdh_path.match(src):
416 self._pathmap[src] = (src, collection_pattern % src[5:])
418 src = src[:src.index("#")]
419 if src not in self._pathmap:
420 ab = cwltool.pathmapper.abspath(src, basedir)
421 st = arvados.commands.run.statfile("", ab, fnPattern=file_pattern)
422 if kwargs.get("conformance_test"):
423 self._pathmap[src] = (src, ab)
424 elif isinstance(st, arvados.commands.run.UploadFile):
425 uploadfiles.add((src, ab, st))
426 elif isinstance(st, arvados.commands.run.ArvFile):
427 self._pathmap[src] = (ab, st.fn)
429 raise cwltool.workflow.WorkflowException("Input file path '%s' is invalid" % st)
432 arvados.commands.run.uploadfiles([u[2] for u in uploadfiles],
434 dry_run=kwargs.get("dry_run"),
436 fnPattern=file_pattern,
438 project=arvrunner.project_uuid)
440 for src, ab, st in uploadfiles:
441 arvrunner.add_uploaded(src, (ab, st.fn))
442 self._pathmap[src] = (ab, st.fn)
446 def reversemap(self, target):
447 if target.startswith("keep:"):
448 return (target, target)
449 elif self.keepdir and target.startswith(self.keepdir):
450 return (target, "keep:" + target[len(self.keepdir)+1:])
452 return super(ArvPathMapper, self).reversemap(target)
455 class ArvadosCommandTool(cwltool.draft2tool.CommandLineTool):
456 """Wrap cwltool CommandLineTool to override selected methods."""
458 def __init__(self, arvrunner, toolpath_object, **kwargs):
459 super(ArvadosCommandTool, self).__init__(toolpath_object, **kwargs)
460 self.arvrunner = arvrunner
462 def makeJobRunner(self):
463 return ArvadosJob(self.arvrunner)
465 def makePathMapper(self, reffiles, input_basedir, **kwargs):
466 return ArvPathMapper(self.arvrunner, reffiles, input_basedir,
468 "$(task.keep)/%s/%s",
472 class ArvCwlRunner(object):
473 """Execute a CWL tool or workflow, submit crunch jobs, wait for them to
474 complete, and report output."""
476 def __init__(self, api_client):
477 self.api = api_client
479 self.lock = threading.Lock()
480 self.cond = threading.Condition(self.lock)
481 self.final_output = None
485 def arvMakeTool(self, toolpath_object, **kwargs):
486 if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
487 return ArvadosCommandTool(self, toolpath_object, **kwargs)
489 return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
491 def output_callback(self, out, processStatus):
492 if processStatus == "success":
493 logger.info("Overall job status is %s", processStatus)
495 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
496 body={"state": "Complete"}).execute(num_retries=self.num_retries)
499 logger.warn("Overall job status is %s", processStatus)
501 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
502 body={"state": "Failed"}).execute(num_retries=self.num_retries)
503 self.final_output = out
506 def on_message(self, event):
507 if "object_uuid" in event:
508 if event["object_uuid"] in self.jobs and event["event_type"] == "update":
509 if event["properties"]["new_attributes"]["state"] == "Running" and self.jobs[event["object_uuid"]].running is False:
510 uuid = event["object_uuid"]
513 logger.info("Job %s (%s) is Running", j.name, uuid)
515 j.update_pipeline_component(event["properties"]["new_attributes"])
516 elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
517 uuid = event["object_uuid"]
521 logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"])
522 j.done(event["properties"]["new_attributes"])
527 def get_uploaded(self):
528 return self.uploaded.copy()
530 def add_uploaded(self, src, pair):
531 self.uploaded[src] = pair
533 def arvExecutor(self, tool, job_order, input_basedir, args, **kwargs):
534 self.debug = args.debug
537 logger.setLevel(logging.WARN)
538 logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
540 useruuid = self.api.users().current().execute()["uuid"]
541 self.project_uuid = args.project_uuid if args.project_uuid else useruuid
545 runnerjob = RunnerJob(self, tool, job_order, args.enable_reuse)
550 events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message)
552 self.debug = args.debug
553 self.ignore_docker_for_reuse = args.ignore_docker_for_reuse
554 self.fs_access = CollectionFsAccess(input_basedir)
556 kwargs["fs_access"] = self.fs_access
557 kwargs["enable_reuse"] = args.enable_reuse
559 kwargs["outdir"] = "$(task.outdir)"
560 kwargs["tmpdir"] = "$(task.tmpdir)"
562 if kwargs.get("conformance_test"):
563 return cwltool.main.single_job_executor(tool, job_order, input_basedir, args, **kwargs)
566 jobiter = iter((runnerjob,))
569 if "cwl_runner_job" in kwargs:
570 components[os.path.basename(tool.tool["id"])] = {"job": kwargs["cwl_runner_job"]}
572 self.pipeline = self.api.pipeline_instances().create(
574 "owner_uuid": self.project_uuid,
575 "name": shortname(tool.tool["id"]),
576 "components": components,
577 "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
579 logger.info("Pipeline instance %s", self.pipeline["uuid"])
581 jobiter = tool.job(job_order,
583 self.output_callback,
584 docker_outdir="$(task.outdir)",
589 # Will continue to hold the lock for the duration of this code
590 # except when in cond.wait(), at which point on_message can update
591 # job state and process output callbacks.
593 for runnable in jobiter:
595 runnable.run(**kwargs)
600 logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
608 if self.final_output is None:
609 raise cwltool.workflow.WorkflowException("Workflow did not return a result.")
611 # create final output collection
613 if sys.exc_info()[0] is KeyboardInterrupt:
614 logger.error("Interrupted, marking pipeline as failed")
616 logger.error("Caught unhandled exception, marking pipeline as failed. Error was: %s", sys.exc_info()[0], exc_info=(sys.exc_info()[1] if self.debug else False))
618 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
619 body={"state": "Failed"}).execute(num_retries=self.num_retries)
623 return self.final_output
626 """Print version string of key packages for provenance and debugging."""
628 arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
629 arvpkg = pkg_resources.require("arvados-python-client")
630 cwlpkg = pkg_resources.require("cwltool")
632 return "%s %s, %s %s, %s %s" % (sys.argv[0], arvcwlpkg[0].version,
633 "arvados-python-client", arvpkg[0].version,
634 "cwltool", cwlpkg[0].version)
636 def main(args, stdout, stderr, api_client=None):
637 args.insert(0, "--leave-outputs")
638 parser = cwltool.main.arg_parser()
640 exgroup = parser.add_mutually_exclusive_group()
641 exgroup.add_argument("--enable-reuse", action="store_true",
642 default=True, dest="enable_reuse",
644 exgroup.add_argument("--disable-reuse", action="store_false",
645 default=True, dest="enable_reuse",
648 parser.add_argument("--project-uuid", type=str, help="Project that will own the workflow jobs")
649 parser.add_argument("--ignore-docker-for-reuse", action="store_true",
650 help="Ignore Docker image version when deciding whether to reuse past jobs.",
653 exgroup = parser.add_mutually_exclusive_group()
654 exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
655 default=True, dest="submit")
656 exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
657 default=True, dest="submit")
659 exgroup = parser.add_mutually_exclusive_group()
660 exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
661 default=True, dest="wait")
662 exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
663 default=True, dest="wait")
666 if api_client is None:
667 api_client=arvados.api('v1', model=OrderedJsonModel())
668 runner = ArvCwlRunner(api_client)
669 except Exception as e:
673 return cwltool.main.main(args,
676 executor=runner.arvExecutor,
677 makeTool=runner.arvMakeTool,
679 versionfunc=versionstring)