1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: Apache-2.0
5 from future import standard_library
6 standard_library.install_aliases()
7 from builtins import str
12 import urllib.request, urllib.parse, urllib.error
19 import arvados_cwl.util
20 import ruamel.yaml as yaml
22 from cwltool.errors import WorkflowException
23 from cwltool.process import UnsupportedRequirement, shortname
24 from cwltool.utils import aslist, adjustFileObjs, adjustDirObjs, visit_class
25 from cwltool.job import JobBase
27 import arvados.collection
29 from .arvdocker import arv_docker_get_image
31 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields, make_builder
32 from .fsaccess import CollectionFetcher
33 from .pathmapper import NoFollowPathMapper, trim_listing
34 from .perf import Perf
35 from ._version import __version__
37 logger = logging.getLogger('arvados.cwl-runner')
38 metrics = logging.getLogger('arvados.cwl-runner.metrics')
40 class ArvadosContainer(JobBase):
41 """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
43 def __init__(self, runner, job_runtime,
44 builder, # type: Builder
45 joborder, # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
46 make_path_mapper, # type: Callable[..., PathMapper]
47 requirements, # type: List[Dict[Text, Text]]
48 hints, # type: List[Dict[Text, Text]]
51 super(ArvadosContainer, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
52 self.arvrunner = runner
53 self.job_runtime = job_runtime
57 def update_pipeline_component(self, r):
60 def _required_env(self):
62 env["HOME"] = self.outdir
63 env["TMPDIR"] = self.tmpdir
66 def run(self, toplevelRuntimeContext):
67 # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
68 # which calls makeJobRunner() to get a new ArvadosContainer
69 # object. The fields that define execution such as
70 # command_line, environment, etc are set on the
71 # ArvadosContainer object by CommandLineTool.job() before
74 runtimeContext = self.job_runtime
76 if runtimeContext.submit_request_uuid:
77 container_request = self.arvrunner.api.container_requests().get(
78 uuid=runtimeContext.submit_request_uuid
79 ).execute(num_retries=self.arvrunner.num_retries)
81 container_request = {}
83 container_request["command"] = self.command_line
84 container_request["name"] = self.name
85 container_request["output_path"] = self.outdir
86 container_request["cwd"] = self.outdir
87 container_request["priority"] = runtimeContext.priority
88 container_request["state"] = "Committed"
89 container_request.setdefault("properties", {})
91 runtime_constraints = {}
93 if runtimeContext.project_uuid:
94 container_request["owner_uuid"] = runtimeContext.project_uuid
96 if self.arvrunner.secret_store.has_secret(self.command_line):
97 raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
99 if self.arvrunner.secret_store.has_secret(self.environment):
100 raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
102 resources = self.builder.resources
103 if resources is not None:
104 runtime_constraints["vcpus"] = math.ceil(resources.get("cores", 1))
105 runtime_constraints["ram"] = math.ceil(resources.get("ram") * 2**20)
110 "capacity": math.ceil(resources.get("outdirSize", 0) * 2**20)
114 "capacity": math.ceil(resources.get("tmpdirSize", 0) * 2**20)
118 scheduling_parameters = {}
120 rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
121 rf.sort(key=lambda k: k.resolved)
123 for resolved, target, tp, stg in rf:
126 if prevdir and target.startswith(prevdir):
128 if tp == "Directory":
131 targetdir = os.path.dirname(target)
132 sp = resolved.split("/", 1)
133 pdh = sp[0][5:] # remove "keep:"
134 mounts[targetdir] = {
135 "kind": "collection",
136 "portable_data_hash": pdh
138 if pdh in self.pathmapper.pdh_to_uuid:
139 mounts[targetdir]["uuid"] = self.pathmapper.pdh_to_uuid[pdh]
141 if tp == "Directory":
144 path = os.path.dirname(sp[1])
145 if path and path != "/":
146 mounts[targetdir]["path"] = path
147 prevdir = targetdir + "/"
149 intermediate_collection_info = arvados_cwl.util.get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl)
151 with Perf(metrics, "generatefiles %s" % self.name):
152 if self.generatefiles["listing"]:
153 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
154 keep_client=self.arvrunner.keep_client,
155 num_retries=self.arvrunner.num_retries)
156 generatemapper = NoFollowPathMapper(self.generatefiles["listing"], "", "",
159 sorteditems = sorted(generatemapper.items(), key=lambda n: n[1].target)
161 logger.debug("generatemapper is %s", sorteditems)
163 with Perf(metrics, "createfiles %s" % self.name):
164 for f, p in sorteditems:
168 if p.target.startswith("/"):
169 dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:]
173 if p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
174 if p.resolved.startswith("_:"):
177 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
178 vwd.copy(path or ".", dst, source_collection=source)
179 elif p.type == "CreateFile":
180 if self.arvrunner.secret_store.has_secret(p.resolved):
181 mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target)
182 secret_mounts[mountpoint] = {
184 "content": self.arvrunner.secret_store.retrieve(p.resolved)
187 with vwd.open(dst, "w") as n:
190 def keepemptydirs(p):
191 if isinstance(p, arvados.collection.RichCollectionBase):
193 p.open(".keep", "w").close()
200 if not runtimeContext.current_container:
201 runtimeContext.current_container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
202 vwd.save_new(name=info["name"],
203 owner_uuid=runtimeContext.project_uuid,
204 ensure_unique_name=True,
205 trash_at=intermediate_collection_info["trash_at"],
206 properties=intermediate_collection_info["properties"])
209 for f, p in sorteditems:
210 if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
211 (prev is not None and p.target.startswith(prev))):
213 if p.target.startswith("/"):
214 dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:]
217 mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target)
218 mounts[mountpoint] = {"kind": "collection",
219 "portable_data_hash": vwd.portable_data_hash(),
221 if p.type.startswith("Writable"):
222 mounts[mountpoint]["writable"] = True
223 prev = p.target + "/"
225 container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
227 container_request["environment"].update(self.environment)
230 sp = self.stdin[6:].split("/", 1)
231 mounts["stdin"] = {"kind": "collection",
232 "portable_data_hash": sp[0],
236 mounts["stderr"] = {"kind": "file",
237 "path": "%s/%s" % (self.outdir, self.stderr)}
240 mounts["stdout"] = {"kind": "file",
241 "path": "%s/%s" % (self.outdir, self.stdout)}
243 (docker_req, docker_is_req) = self.get_requirement("DockerRequirement")
245 container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
247 runtimeContext.pull_image,
248 runtimeContext.project_uuid,
249 runtimeContext.force_docker_pull,
250 runtimeContext.tmp_outdir_prefix,
251 runtimeContext.match_local_docker,
252 runtimeContext.copy_deps)
254 network_req, _ = self.get_requirement("NetworkAccess")
256 runtime_constraints["API"] = network_req["networkAccess"]
258 api_req, _ = self.get_requirement("http://arvados.org/cwl#APIRequirement")
260 runtime_constraints["API"] = True
262 runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints")
264 if "keep_cache" in runtime_req:
265 runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20)
266 if "outputDirType" in runtime_req:
267 if runtime_req["outputDirType"] == "local_output_dir":
268 # Currently the default behavior.
270 elif runtime_req["outputDirType"] == "keep_output_dir":
271 mounts[self.outdir]= {
272 "kind": "collection",
276 partition_req, _ = self.get_requirement("http://arvados.org/cwl#PartitionRequirement")
278 scheduling_parameters["partitions"] = aslist(partition_req["partition"])
280 intermediate_output_req, _ = self.get_requirement("http://arvados.org/cwl#IntermediateOutput")
281 if intermediate_output_req:
282 self.output_ttl = intermediate_output_req["outputTTL"]
284 self.output_ttl = self.arvrunner.intermediate_output_ttl
286 if self.output_ttl < 0:
287 raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
290 if self.arvrunner.api._rootDesc["revision"] >= "20210628":
291 storage_class_req, _ = self.get_requirement("http://arvados.org/cwl#OutputStorageClass")
292 if storage_class_req and storage_class_req.get("intermediateStorageClass"):
293 container_request["output_storage_classes"] = aslist(storage_class_req["intermediateStorageClass"])
295 container_request["output_storage_classes"] = runtimeContext.intermediate_storage_classes.strip().split(",")
297 cuda_req, _ = self.get_requirement("http://commonwl.org/cwltool#CUDARequirement")
299 runtime_constraints["cuda"] = {
300 "device_count": resources.get("cudaDeviceCount", 1),
301 "driver_version": cuda_req["cudaVersionMin"],
302 "hardware_capability": aslist(cuda_req["cudaComputeCapability"])[0]
305 if runtimeContext.enable_preemptible is False:
306 scheduling_parameters["preemptible"] = False
308 preemptible_req, _ = self.get_requirement("http://arvados.org/cwl#UsePreemptible")
310 scheduling_parameters["preemptible"] = preemptible_req["usePreemptible"]
311 elif runtimeContext.enable_preemptible is True:
312 scheduling_parameters["preemptible"] = True
313 elif runtimeContext.enable_preemptible is None:
316 if self.timelimit is not None and self.timelimit > 0:
317 scheduling_parameters["max_run_time"] = self.timelimit
319 extra_submit_params = {}
320 if runtimeContext.submit_runner_cluster:
321 extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
323 container_request["output_name"] = "Output for step %s" % (self.name)
324 container_request["output_ttl"] = self.output_ttl
325 container_request["mounts"] = mounts
326 container_request["secret_mounts"] = secret_mounts
327 container_request["runtime_constraints"] = runtime_constraints
328 container_request["scheduling_parameters"] = scheduling_parameters
330 enable_reuse = runtimeContext.enable_reuse
332 reuse_req, _ = self.get_requirement("WorkReuse")
334 enable_reuse = reuse_req["enableReuse"]
335 reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
337 enable_reuse = reuse_req["enableReuse"]
338 container_request["use_existing"] = enable_reuse
340 properties_req, _ = self.get_requirement("http://arvados.org/cwl#ProcessProperties")
342 for pr in properties_req["processProperties"]:
343 container_request["properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
345 container_request["output_properties"] = {}
346 output_properties_req, _ = self.get_requirement("http://arvados.org/cwl#OutputCollectionProperties")
347 if output_properties_req:
348 for pr in output_properties_req["outputProperties"]:
349 container_request["output_properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
351 container_request["output_properties"].update(intermediate_collection_info["properties"])
353 if runtimeContext.runnerjob.startswith("arvwf:"):
354 wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")]
355 wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
356 if container_request["name"] == "main":
357 container_request["name"] = wfrecord["name"]
358 container_request["properties"]["template_uuid"] = wfuuid
360 self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
363 if runtimeContext.submit_request_uuid:
364 response = self.arvrunner.api.container_requests().update(
365 uuid=runtimeContext.submit_request_uuid,
366 body=container_request,
367 **extra_submit_params
368 ).execute(num_retries=self.arvrunner.num_retries)
370 response = self.arvrunner.api.container_requests().create(
371 body=container_request,
372 **extra_submit_params
373 ).execute(num_retries=self.arvrunner.num_retries)
375 self.uuid = response["uuid"]
376 self.arvrunner.process_submitted(self)
378 if response["state"] == "Final":
379 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
381 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
382 except Exception as e:
383 logger.exception("%s error submitting container\n%s", self.arvrunner.label(self), e)
384 logger.debug("Container request was %s", container_request)
385 self.output_callback({}, "permanentFail")
387 def done(self, record):
390 container = self.arvrunner.api.containers().get(
391 uuid=record["container_uuid"]
392 ).execute(num_retries=self.arvrunner.num_retries)
393 if container["state"] == "Complete":
394 rcode = container["exit_code"]
395 if self.successCodes and rcode in self.successCodes:
396 processStatus = "success"
397 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
398 processStatus = "temporaryFail"
399 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
400 processStatus = "permanentFail"
402 processStatus = "success"
404 processStatus = "permanentFail"
407 logger.warning("%s Container may have been killed for using too much RAM. Try resubmitting with a higher 'ramMin'.",
408 self.arvrunner.label(self))
410 processStatus = "permanentFail"
412 if processStatus == "permanentFail" and record["log_uuid"]:
413 logc = arvados.collection.CollectionReader(record["log_uuid"],
414 api_client=self.arvrunner.api,
415 keep_client=self.arvrunner.keep_client,
416 num_retries=self.arvrunner.num_retries)
417 label = self.arvrunner.label(self)
420 "%s (%s) error log:" % (label, record["uuid"]), maxlen=40)
422 if record["output_uuid"]:
423 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
424 # Compute the trash time to avoid requesting the collection record.
425 trash_at = ciso8601.parse_datetime_as_naive(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
426 aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
427 orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
428 oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
429 logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
430 self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
431 self.arvrunner.add_intermediate_output(record["output_uuid"])
433 if container["output"]:
434 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
435 except WorkflowException as e:
436 # Only include a stack trace if in debug mode.
437 # A stack trace may obfuscate more useful output about the workflow.
438 logger.error("%s unable to collect output from %s:\n%s",
439 self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
440 processStatus = "permanentFail"
442 logger.exception("%s while getting output object:", self.arvrunner.label(self))
443 processStatus = "permanentFail"
445 self.output_callback(outputs, processStatus)
448 class RunnerContainer(Runner):
449 """Submit and manage a container that runs arvados-cwl-runner."""
451 def arvados_job_spec(self, runtimeContext):
452 """Create an Arvados container request for this workflow.
454 The returned dict can be used to create a container passed as
455 the +body+ argument to container_requests().create().
458 adjustDirObjs(self.job_order, trim_listing)
459 visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
460 visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
463 for param in sorted(self.job_order.keys()):
464 if self.secret_store.has_secret(self.job_order[param]):
465 mnt = "/secrets/s%d" % len(secret_mounts)
466 secret_mounts[mnt] = {
468 "content": self.secret_store.retrieve(self.job_order[param])
470 self.job_order[param] = {"$include": mnt}
474 "output_path": "/var/spool/cwl",
475 "cwd": "/var/spool/cwl",
476 "priority": self.priority,
477 "state": "Committed",
478 "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image, runtimeContext),
480 "/var/lib/cwl/cwl.input.json": {
482 "content": self.job_order
486 "path": "/var/spool/cwl/cwl.output.json"
489 "kind": "collection",
493 "secret_mounts": secret_mounts,
494 "runtime_constraints": {
495 "vcpus": math.ceil(self.submit_runner_cores),
496 "ram": 1024*1024 * (math.ceil(self.submit_runner_ram) + math.ceil(self.collection_cache_size)),
499 "use_existing": False, # Never reuse the runner container - see #15497.
503 if self.embedded_tool.tool.get("id", "").startswith("keep:"):
504 sp = self.embedded_tool.tool["id"].split('/')
505 workflowcollection = sp[0][5:]
506 workflowname = "/".join(sp[1:])
507 workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
508 container_req["mounts"]["/var/lib/cwl/workflow"] = {
509 "kind": "collection",
510 "portable_data_hash": "%s" % workflowcollection
513 packed = packed_workflow(self.arvrunner, self.embedded_tool, self.merged_map, runtimeContext)
514 workflowpath = "/var/lib/cwl/workflow.json#main"
515 container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
519 if self.embedded_tool.tool.get("id", "").startswith("arvwf:"):
520 container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33]
522 properties_req, _ = self.embedded_tool.get_requirement("http://arvados.org/cwl#ProcessProperties")
524 builder = make_builder(self.job_order, self.embedded_tool.hints, self.embedded_tool.requirements, runtimeContext, self.embedded_tool.metadata)
525 for pr in properties_req["processProperties"]:
526 container_req["properties"][pr["propertyName"]] = builder.do_eval(pr["propertyValue"])
528 # --local means execute the workflow instead of submitting a container request
529 # --api=containers means use the containers API
530 # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
531 # --disable-validate because we already validated so don't need to do it again
532 # --eval-timeout is the timeout for javascript invocation
533 # --parallel-task-count is the number of threads to use for job submission
534 # --enable/disable-reuse sets desired job reuse
535 # --collection-cache-size sets aside memory to store collections
536 command = ["arvados-cwl-runner",
539 "--no-log-timestamps",
540 "--disable-validate",
542 "--eval-timeout=%s" % self.arvrunner.eval_timeout,
543 "--thread-count=%s" % self.arvrunner.thread_count,
544 "--enable-reuse" if self.enable_reuse else "--disable-reuse",
545 "--collection-cache-size=%s" % self.collection_cache_size]
548 command.append("--output-name=" + self.output_name)
549 container_req["output_name"] = self.output_name
552 command.append("--output-tags=" + self.output_tags)
554 if runtimeContext.debug:
555 command.append("--debug")
557 if runtimeContext.storage_classes != "default" and runtimeContext.storage_classes:
558 command.append("--storage-classes=" + runtimeContext.storage_classes)
560 if runtimeContext.intermediate_storage_classes != "default" and runtimeContext.intermediate_storage_classes:
561 command.append("--intermediate-storage-classes=" + runtimeContext.intermediate_storage_classes)
563 if runtimeContext.on_error:
564 command.append("--on-error=" + self.on_error)
566 if runtimeContext.intermediate_output_ttl:
567 command.append("--intermediate-output-ttl=%d" % runtimeContext.intermediate_output_ttl)
569 if runtimeContext.trash_intermediate:
570 command.append("--trash-intermediate")
572 if runtimeContext.project_uuid:
573 command.append("--project-uuid="+runtimeContext.project_uuid)
576 command.append("--enable-dev")
578 if runtimeContext.enable_preemptible is True:
579 command.append("--enable-preemptible")
581 if runtimeContext.enable_preemptible is False:
582 command.append("--disable-preemptible")
584 command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
586 container_req["command"] = command
591 def run(self, runtimeContext):
592 runtimeContext.keepprefix = "keep:"
593 job_spec = self.arvados_job_spec(runtimeContext)
594 if runtimeContext.project_uuid:
595 job_spec["owner_uuid"] = runtimeContext.project_uuid
597 extra_submit_params = {}
598 if runtimeContext.submit_runner_cluster:
599 extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
601 if runtimeContext.submit_request_uuid:
602 if "cluster_id" in extra_submit_params:
603 # Doesn't make sense for "update" and actually fails
604 del extra_submit_params["cluster_id"]
605 response = self.arvrunner.api.container_requests().update(
606 uuid=runtimeContext.submit_request_uuid,
608 **extra_submit_params
609 ).execute(num_retries=self.arvrunner.num_retries)
611 response = self.arvrunner.api.container_requests().create(
613 **extra_submit_params
614 ).execute(num_retries=self.arvrunner.num_retries)
616 self.uuid = response["uuid"]
617 self.arvrunner.process_submitted(self)
619 logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"])
621 workbench1 = self.arvrunner.api.config()["Services"]["Workbench1"]["ExternalURL"]
622 workbench2 = self.arvrunner.api.config()["Services"]["Workbench2"]["ExternalURL"]
625 url = "{}processes/{}".format(workbench2, response["uuid"])
627 url = "{}container_requests/{}".format(workbench1, response["uuid"])
629 logger.info("Monitor workflow progress at %s", url)
632 def done(self, record):
634 container = self.arvrunner.api.containers().get(
635 uuid=record["container_uuid"]
636 ).execute(num_retries=self.arvrunner.num_retries)
637 container["log"] = record["log_uuid"]
639 logger.exception("%s while getting runner container", self.arvrunner.label(self))
640 self.arvrunner.output_callback({}, "permanentFail")
642 super(RunnerContainer, self).done(container)