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, runtimeContext):
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 with Perf(metrics, "generatefiles %s" % self.name):
150 if self.generatefiles["listing"]:
151 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
152 keep_client=self.arvrunner.keep_client,
153 num_retries=self.arvrunner.num_retries)
154 generatemapper = NoFollowPathMapper(self.generatefiles["listing"], "", "",
157 sorteditems = sorted(generatemapper.items(), key=lambda n: n[1].target)
159 logger.debug("generatemapper is %s", sorteditems)
161 with Perf(metrics, "createfiles %s" % self.name):
162 for f, p in sorteditems:
166 if p.target.startswith("/"):
167 dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:]
171 if p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
172 if p.resolved.startswith("_:"):
175 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
176 vwd.copy(path or ".", dst, source_collection=source)
177 elif p.type == "CreateFile":
178 if self.arvrunner.secret_store.has_secret(p.resolved):
179 mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target)
180 secret_mounts[mountpoint] = {
182 "content": self.arvrunner.secret_store.retrieve(p.resolved)
185 with vwd.open(dst, "w") as n:
188 def keepemptydirs(p):
189 if isinstance(p, arvados.collection.RichCollectionBase):
191 p.open(".keep", "w").close()
198 if not runtimeContext.current_container:
199 runtimeContext.current_container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
200 info = arvados_cwl.util.get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl)
201 vwd.save_new(name=info["name"],
202 owner_uuid=runtimeContext.project_uuid,
203 ensure_unique_name=True,
204 trash_at=info["trash_at"],
205 properties=info["properties"])
208 for f, p in sorteditems:
209 if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
210 (prev is not None and p.target.startswith(prev))):
212 if p.target.startswith("/"):
213 dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:]
216 mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target)
217 mounts[mountpoint] = {"kind": "collection",
218 "portable_data_hash": vwd.portable_data_hash(),
220 if p.type.startswith("Writable"):
221 mounts[mountpoint]["writable"] = True
222 prev = p.target + "/"
224 container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
226 container_request["environment"].update(self.environment)
229 sp = self.stdin[6:].split("/", 1)
230 mounts["stdin"] = {"kind": "collection",
231 "portable_data_hash": sp[0],
235 mounts["stderr"] = {"kind": "file",
236 "path": "%s/%s" % (self.outdir, self.stderr)}
239 mounts["stdout"] = {"kind": "file",
240 "path": "%s/%s" % (self.outdir, self.stdout)}
242 (docker_req, docker_is_req) = self.get_requirement("DockerRequirement")
244 container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
246 runtimeContext.pull_image,
247 runtimeContext.project_uuid,
248 runtimeContext.force_docker_pull,
249 runtimeContext.tmp_outdir_prefix)
251 network_req, _ = self.get_requirement("NetworkAccess")
253 runtime_constraints["API"] = network_req["networkAccess"]
255 api_req, _ = self.get_requirement("http://arvados.org/cwl#APIRequirement")
257 runtime_constraints["API"] = True
259 runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints")
261 if "keep_cache" in runtime_req:
262 runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20)
263 if "outputDirType" in runtime_req:
264 if runtime_req["outputDirType"] == "local_output_dir":
265 # Currently the default behavior.
267 elif runtime_req["outputDirType"] == "keep_output_dir":
268 mounts[self.outdir]= {
269 "kind": "collection",
273 partition_req, _ = self.get_requirement("http://arvados.org/cwl#PartitionRequirement")
275 scheduling_parameters["partitions"] = aslist(partition_req["partition"])
277 intermediate_output_req, _ = self.get_requirement("http://arvados.org/cwl#IntermediateOutput")
278 if intermediate_output_req:
279 self.output_ttl = intermediate_output_req["outputTTL"]
281 self.output_ttl = self.arvrunner.intermediate_output_ttl
283 if self.output_ttl < 0:
284 raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
286 storage_class_req, _ = self.get_requirement("http://arvados.org/cwl#OutputStorageClass")
287 if storage_class_req and storage_class_req.get("intermediateStorageClass"):
288 container_request["output_storage_classes"] = aslist(storage_class_req["intermediateStorageClass"])
290 container_request["output_storage_classes"] = runtimeContext.intermediate_storage_classes.strip().split(",")
292 if self.timelimit is not None and self.timelimit > 0:
293 scheduling_parameters["max_run_time"] = self.timelimit
295 extra_submit_params = {}
296 if runtimeContext.submit_runner_cluster:
297 extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
299 container_request["output_name"] = "Output for step %s" % (self.name)
300 container_request["output_ttl"] = self.output_ttl
301 container_request["mounts"] = mounts
302 container_request["secret_mounts"] = secret_mounts
303 container_request["runtime_constraints"] = runtime_constraints
304 container_request["scheduling_parameters"] = scheduling_parameters
306 enable_reuse = runtimeContext.enable_reuse
308 reuse_req, _ = self.get_requirement("WorkReuse")
310 enable_reuse = reuse_req["enableReuse"]
311 reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
313 enable_reuse = reuse_req["enableReuse"]
314 container_request["use_existing"] = enable_reuse
316 properties_req, _ = self.get_requirement("http://arvados.org/cwl#ProcessProperties")
318 for pr in properties_req["processProperties"]:
319 container_request["properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
321 if runtimeContext.runnerjob.startswith("arvwf:"):
322 wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")]
323 wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
324 if container_request["name"] == "main":
325 container_request["name"] = wfrecord["name"]
326 container_request["properties"]["template_uuid"] = wfuuid
328 self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
331 if runtimeContext.submit_request_uuid:
332 response = self.arvrunner.api.container_requests().update(
333 uuid=runtimeContext.submit_request_uuid,
334 body=container_request,
335 **extra_submit_params
336 ).execute(num_retries=self.arvrunner.num_retries)
338 response = self.arvrunner.api.container_requests().create(
339 body=container_request,
340 **extra_submit_params
341 ).execute(num_retries=self.arvrunner.num_retries)
343 self.uuid = response["uuid"]
344 self.arvrunner.process_submitted(self)
346 if response["state"] == "Final":
347 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
349 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
350 except Exception as e:
351 logger.exception("%s error submitting container\n%s", self.arvrunner.label(self), e)
352 logger.debug("Container request was %s", container_request)
353 self.output_callback({}, "permanentFail")
355 def done(self, record):
358 container = self.arvrunner.api.containers().get(
359 uuid=record["container_uuid"]
360 ).execute(num_retries=self.arvrunner.num_retries)
361 if container["state"] == "Complete":
362 rcode = container["exit_code"]
363 if self.successCodes and rcode in self.successCodes:
364 processStatus = "success"
365 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
366 processStatus = "temporaryFail"
367 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
368 processStatus = "permanentFail"
370 processStatus = "success"
372 processStatus = "permanentFail"
374 processStatus = "permanentFail"
376 if processStatus == "permanentFail" and record["log_uuid"]:
377 logc = arvados.collection.CollectionReader(record["log_uuid"],
378 api_client=self.arvrunner.api,
379 keep_client=self.arvrunner.keep_client,
380 num_retries=self.arvrunner.num_retries)
381 label = self.arvrunner.label(self)
384 "%s (%s) error log:" % (label, record["uuid"]), maxlen=40)
386 if record["output_uuid"]:
387 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
388 # Compute the trash time to avoid requesting the collection record.
389 trash_at = ciso8601.parse_datetime_as_naive(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
390 aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
391 orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
392 oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
393 logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
394 self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
395 self.arvrunner.add_intermediate_output(record["output_uuid"])
397 if container["output"]:
398 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
399 except WorkflowException as e:
400 # Only include a stack trace if in debug mode.
401 # A stack trace may obfuscate more useful output about the workflow.
402 logger.error("%s unable to collect output from %s:\n%s",
403 self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
404 processStatus = "permanentFail"
406 logger.exception("%s while getting output object:", self.arvrunner.label(self))
407 processStatus = "permanentFail"
409 self.output_callback(outputs, processStatus)
412 class RunnerContainer(Runner):
413 """Submit and manage a container that runs arvados-cwl-runner."""
415 def arvados_job_spec(self, runtimeContext):
416 """Create an Arvados container request for this workflow.
418 The returned dict can be used to create a container passed as
419 the +body+ argument to container_requests().create().
422 adjustDirObjs(self.job_order, trim_listing)
423 visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
424 visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
427 for param in sorted(self.job_order.keys()):
428 if self.secret_store.has_secret(self.job_order[param]):
429 mnt = "/secrets/s%d" % len(secret_mounts)
430 secret_mounts[mnt] = {
432 "content": self.secret_store.retrieve(self.job_order[param])
434 self.job_order[param] = {"$include": mnt}
438 "output_path": "/var/spool/cwl",
439 "cwd": "/var/spool/cwl",
440 "priority": self.priority,
441 "state": "Committed",
442 "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
444 "/var/lib/cwl/cwl.input.json": {
446 "content": self.job_order
450 "path": "/var/spool/cwl/cwl.output.json"
453 "kind": "collection",
457 "secret_mounts": secret_mounts,
458 "runtime_constraints": {
459 "vcpus": math.ceil(self.submit_runner_cores),
460 "ram": 1024*1024 * (math.ceil(self.submit_runner_ram) + math.ceil(self.collection_cache_size)),
463 "use_existing": False, # Never reuse the runner container - see #15497.
467 if self.embedded_tool.tool.get("id", "").startswith("keep:"):
468 sp = self.embedded_tool.tool["id"].split('/')
469 workflowcollection = sp[0][5:]
470 workflowname = "/".join(sp[1:])
471 workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
472 container_req["mounts"]["/var/lib/cwl/workflow"] = {
473 "kind": "collection",
474 "portable_data_hash": "%s" % workflowcollection
477 packed = packed_workflow(self.arvrunner, self.embedded_tool, self.merged_map)
478 workflowpath = "/var/lib/cwl/workflow.json#main"
479 container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
483 if self.embedded_tool.tool.get("id", "").startswith("arvwf:"):
484 container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33]
486 properties_req, _ = self.embedded_tool.get_requirement("http://arvados.org/cwl#ProcessProperties")
488 builder = make_builder(self.job_order, self.embedded_tool.hints, self.embedded_tool.requirements, runtimeContext, self.embedded_tool.metadata)
489 for pr in properties_req["processProperties"]:
490 container_req["properties"][pr["propertyName"]] = builder.do_eval(pr["propertyValue"])
492 # --local means execute the workflow instead of submitting a container request
493 # --api=containers means use the containers API
494 # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
495 # --disable-validate because we already validated so don't need to do it again
496 # --eval-timeout is the timeout for javascript invocation
497 # --parallel-task-count is the number of threads to use for job submission
498 # --enable/disable-reuse sets desired job reuse
499 # --collection-cache-size sets aside memory to store collections
500 command = ["arvados-cwl-runner",
503 "--no-log-timestamps",
504 "--disable-validate",
506 "--eval-timeout=%s" % self.arvrunner.eval_timeout,
507 "--thread-count=%s" % self.arvrunner.thread_count,
508 "--enable-reuse" if self.enable_reuse else "--disable-reuse",
509 "--collection-cache-size=%s" % self.collection_cache_size]
512 command.append("--output-name=" + self.output_name)
513 container_req["output_name"] = self.output_name
516 command.append("--output-tags=" + self.output_tags)
518 if runtimeContext.debug:
519 command.append("--debug")
521 if runtimeContext.storage_classes != "default":
522 command.append("--storage-classes=" + runtimeContext.storage_classes)
524 if runtimeContext.intermediate_storage_classes != "default":
525 command.append("--intermediate-storage-classes=" + runtimeContext.intermediate_storage_classes)
528 command.append("--on-error=" + self.on_error)
530 if self.intermediate_output_ttl:
531 command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
533 if self.arvrunner.trash_intermediate:
534 command.append("--trash-intermediate")
536 if self.arvrunner.project_uuid:
537 command.append("--project-uuid="+self.arvrunner.project_uuid)
540 command.append("--enable-dev")
542 command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
544 container_req["command"] = command
549 def run(self, runtimeContext):
550 runtimeContext.keepprefix = "keep:"
551 job_spec = self.arvados_job_spec(runtimeContext)
552 if self.arvrunner.project_uuid:
553 job_spec["owner_uuid"] = self.arvrunner.project_uuid
555 extra_submit_params = {}
556 if runtimeContext.submit_runner_cluster:
557 extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
559 if runtimeContext.submit_request_uuid:
560 if "cluster_id" in extra_submit_params:
561 # Doesn't make sense for "update" and actually fails
562 del extra_submit_params["cluster_id"]
563 response = self.arvrunner.api.container_requests().update(
564 uuid=runtimeContext.submit_request_uuid,
566 **extra_submit_params
567 ).execute(num_retries=self.arvrunner.num_retries)
569 response = self.arvrunner.api.container_requests().create(
571 **extra_submit_params
572 ).execute(num_retries=self.arvrunner.num_retries)
574 self.uuid = response["uuid"]
575 self.arvrunner.process_submitted(self)
577 logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"])
579 workbench1 = self.arvrunner.api.config()["Services"]["Workbench1"]["ExternalURL"]
580 workbench2 = self.arvrunner.api.config()["Services"]["Workbench2"]["ExternalURL"]
583 url = "{}processes/{}".format(workbench2, response["uuid"])
585 url = "{}container_requests/{}".format(workbench1, response["uuid"])
587 logger.info("Monitor workflow progress at %s", url)
590 def done(self, record):
592 container = self.arvrunner.api.containers().get(
593 uuid=record["container_uuid"]
594 ).execute(num_retries=self.arvrunner.num_retries)
595 container["log"] = record["log_uuid"]
597 logger.exception("%s while getting runner container", self.arvrunner.label(self))
598 self.arvrunner.output_callback({}, "permanentFail")
600 super(RunnerContainer, self).done(container)