1 # Copyright (C) The Arvados Authors. All rights reserved.
3 # SPDX-License-Identifier: Apache-2.0
15 from arvados_cwl.util import get_current_container, get_intermediate_collection_info
16 import ruamel.yaml as yaml
18 from cwltool.errors import WorkflowException
19 from cwltool.process import UnsupportedRequirement, shortname
20 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
21 from cwltool.utils import aslist
22 from cwltool.job import JobBase
24 import arvados.collection
26 from .arvdocker import arv_docker_get_image
28 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields
29 from .fsaccess import CollectionFetcher
30 from .pathmapper import NoFollowPathMapper, trim_listing
31 from .perf import Perf
33 logger = logging.getLogger('arvados.cwl-runner')
34 metrics = logging.getLogger('arvados.cwl-runner.metrics')
36 class ArvadosContainer(JobBase):
37 """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
39 def __init__(self, runner,
40 builder, # type: Builder
41 joborder, # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
42 make_path_mapper, # type: Callable[..., PathMapper]
43 requirements, # type: List[Dict[Text, Text]]
44 hints, # type: List[Dict[Text, Text]]
47 super(ArvadosContainer, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
48 self.arvrunner = runner
52 def update_pipeline_component(self, r):
55 def run(self, runtimeContext):
56 # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
57 # which calls makeJobRunner() to get a new ArvadosContainer
58 # object. The fields that define execution such as
59 # command_line, environment, etc are set on the
60 # ArvadosContainer object by CommandLineTool.job() before
64 "command": self.command_line,
66 "output_path": self.outdir,
68 "priority": runtimeContext.priority,
72 runtime_constraints = {}
74 if self.arvrunner.project_uuid:
75 container_request["owner_uuid"] = self.arvrunner.project_uuid
77 if self.arvrunner.secret_store.has_secret(self.command_line):
78 raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
80 if self.arvrunner.secret_store.has_secret(self.environment):
81 raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
83 resources = self.builder.resources
84 if resources is not None:
85 runtime_constraints["vcpus"] = math.ceil(resources.get("cores", 1))
86 runtime_constraints["ram"] = math.ceil(resources.get("ram") * 2**20)
91 "capacity": math.ceil(resources.get("outdirSize", 0) * 2**20)
95 "capacity": math.ceil(resources.get("tmpdirSize", 0) * 2**20)
99 scheduling_parameters = {}
101 rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
102 rf.sort(key=lambda k: k.resolved)
104 for resolved, target, tp, stg in rf:
107 if prevdir and target.startswith(prevdir):
109 if tp == "Directory":
112 targetdir = os.path.dirname(target)
113 sp = resolved.split("/", 1)
114 pdh = sp[0][5:] # remove "keep:"
115 mounts[targetdir] = {
116 "kind": "collection",
117 "portable_data_hash": pdh
120 if tp == "Directory":
123 path = os.path.dirname(sp[1])
124 if path and path != "/":
125 mounts[targetdir]["path"] = path
126 prevdir = targetdir + "/"
128 with Perf(metrics, "generatefiles %s" % self.name):
129 if self.generatefiles["listing"]:
130 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
131 keep_client=self.arvrunner.keep_client,
132 num_retries=self.arvrunner.num_retries)
133 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
136 sorteditems = sorted(generatemapper.items(), None, key=lambda n: n[1].target)
138 logger.debug("generatemapper is %s", sorteditems)
140 with Perf(metrics, "createfiles %s" % self.name):
141 for f, p in sorteditems:
144 elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
145 if p.resolved.startswith("_:"):
148 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
149 vwd.copy(path, p.target, source_collection=source)
150 elif p.type == "CreateFile":
151 if self.arvrunner.secret_store.has_secret(p.resolved):
152 secret_mounts["%s/%s" % (self.outdir, p.target)] = {
154 "content": self.arvrunner.secret_store.retrieve(p.resolved)
157 with vwd.open(p.target, "w") as n:
158 n.write(p.resolved.encode("utf-8"))
160 def keepemptydirs(p):
161 if isinstance(p, arvados.collection.RichCollectionBase):
163 p.open(".keep", "w").close()
170 if not runtimeContext.current_container:
171 runtimeContext.current_container = get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
172 info = get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl)
173 vwd.save_new(name=info["name"],
174 owner_uuid=self.arvrunner.project_uuid,
175 ensure_unique_name=True,
176 trash_at=info["trash_at"],
177 properties=info["properties"])
180 for f, p in sorteditems:
181 if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
182 (prev is not None and p.target.startswith(prev))):
184 mountpoint = "%s/%s" % (self.outdir, p.target)
185 mounts[mountpoint] = {"kind": "collection",
186 "portable_data_hash": vwd.portable_data_hash(),
188 if p.type.startswith("Writable"):
189 mounts[mountpoint]["writable"] = True
190 prev = p.target + "/"
192 container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
194 container_request["environment"].update(self.environment)
197 sp = self.stdin[6:].split("/", 1)
198 mounts["stdin"] = {"kind": "collection",
199 "portable_data_hash": sp[0],
203 mounts["stderr"] = {"kind": "file",
204 "path": "%s/%s" % (self.outdir, self.stderr)}
207 mounts["stdout"] = {"kind": "file",
208 "path": "%s/%s" % (self.outdir, self.stdout)}
210 (docker_req, docker_is_req) = self.get_requirement("DockerRequirement")
212 docker_req = {"dockerImageId": "arvados/jobs"}
214 container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
216 runtimeContext.pull_image,
217 self.arvrunner.project_uuid)
219 api_req, _ = self.get_requirement("http://arvados.org/cwl#APIRequirement")
221 runtime_constraints["API"] = True
223 runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints")
225 if "keep_cache" in runtime_req:
226 runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20)
227 if "outputDirType" in runtime_req:
228 if runtime_req["outputDirType"] == "local_output_dir":
229 # Currently the default behavior.
231 elif runtime_req["outputDirType"] == "keep_output_dir":
232 mounts[self.outdir]= {
233 "kind": "collection",
237 partition_req, _ = self.get_requirement("http://arvados.org/cwl#PartitionRequirement")
239 scheduling_parameters["partitions"] = aslist(partition_req["partition"])
241 intermediate_output_req, _ = self.get_requirement("http://arvados.org/cwl#IntermediateOutput")
242 if intermediate_output_req:
243 self.output_ttl = intermediate_output_req["outputTTL"]
245 self.output_ttl = self.arvrunner.intermediate_output_ttl
247 if self.output_ttl < 0:
248 raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
250 if self.timelimit is not None:
251 scheduling_parameters["max_run_time"] = self.timelimit
253 container_request["output_name"] = "Output for step %s" % (self.name)
254 container_request["output_ttl"] = self.output_ttl
255 container_request["mounts"] = mounts
256 container_request["secret_mounts"] = secret_mounts
257 container_request["runtime_constraints"] = runtime_constraints
258 container_request["scheduling_parameters"] = scheduling_parameters
260 enable_reuse = runtimeContext.enable_reuse
262 reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
264 enable_reuse = reuse_req["enableReuse"]
265 container_request["use_existing"] = enable_reuse
267 if runtimeContext.runnerjob.startswith("arvwf:"):
268 wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")]
269 wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
270 if container_request["name"] == "main":
271 container_request["name"] = wfrecord["name"]
272 container_request["properties"]["template_uuid"] = wfuuid
274 self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
277 if runtimeContext.submit_request_uuid:
278 response = self.arvrunner.api.container_requests().update(
279 uuid=runtimeContext.submit_request_uuid,
280 body=container_request
281 ).execute(num_retries=self.arvrunner.num_retries)
283 response = self.arvrunner.api.container_requests().create(
284 body=container_request
285 ).execute(num_retries=self.arvrunner.num_retries)
287 self.uuid = response["uuid"]
288 self.arvrunner.process_submitted(self)
290 if response["state"] == "Final":
291 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
293 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
294 except Exception as e:
295 logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
296 self.output_callback({}, "permanentFail")
298 def done(self, record):
301 container = self.arvrunner.api.containers().get(
302 uuid=record["container_uuid"]
303 ).execute(num_retries=self.arvrunner.num_retries)
304 if container["state"] == "Complete":
305 rcode = container["exit_code"]
306 if self.successCodes and rcode in self.successCodes:
307 processStatus = "success"
308 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
309 processStatus = "temporaryFail"
310 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
311 processStatus = "permanentFail"
313 processStatus = "success"
315 processStatus = "permanentFail"
317 processStatus = "permanentFail"
319 if processStatus == "permanentFail":
320 logc = arvados.collection.CollectionReader(container["log"],
321 api_client=self.arvrunner.api,
322 keep_client=self.arvrunner.keep_client,
323 num_retries=self.arvrunner.num_retries)
324 done.logtail(logc, logger.error, "%s (%s) error log:" % (self.arvrunner.label(self), record["uuid"]), maxlen=40)
326 if record["output_uuid"]:
327 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
328 # Compute the trash time to avoid requesting the collection record.
329 trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
330 aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
331 orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
332 oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
333 logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
334 self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
335 self.arvrunner.add_intermediate_output(record["output_uuid"])
337 if container["output"]:
338 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
339 except WorkflowException as e:
340 logger.error("%s unable to collect output from %s:\n%s",
341 self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
342 processStatus = "permanentFail"
343 except Exception as e:
344 logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
345 processStatus = "permanentFail"
347 self.output_callback(outputs, processStatus)
350 class RunnerContainer(Runner):
351 """Submit and manage a container that runs arvados-cwl-runner."""
353 def arvados_job_spec(self, runtimeContext):
354 """Create an Arvados container request for this workflow.
356 The returned dict can be used to create a container passed as
357 the +body+ argument to container_requests().create().
360 adjustDirObjs(self.job_order, trim_listing)
361 visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
362 visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
365 for param in sorted(self.job_order.keys()):
366 if self.secret_store.has_secret(self.job_order[param]):
367 mnt = "/secrets/s%d" % len(secret_mounts)
368 secret_mounts[mnt] = {
370 "content": self.secret_store.retrieve(self.job_order[param])
372 self.job_order[param] = {"$include": mnt}
376 "output_path": "/var/spool/cwl",
377 "cwd": "/var/spool/cwl",
378 "priority": self.priority,
379 "state": "Committed",
380 "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
382 "/var/lib/cwl/cwl.input.json": {
384 "content": self.job_order
388 "path": "/var/spool/cwl/cwl.output.json"
391 "kind": "collection",
395 "secret_mounts": secret_mounts,
396 "runtime_constraints": {
397 "vcpus": math.ceil(self.submit_runner_cores),
398 "ram": math.ceil(1024*1024 * self.submit_runner_ram),
401 "use_existing": self.enable_reuse,
405 if self.tool.tool.get("id", "").startswith("keep:"):
406 sp = self.tool.tool["id"].split('/')
407 workflowcollection = sp[0][5:]
408 workflowname = "/".join(sp[1:])
409 workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
410 container_req["mounts"]["/var/lib/cwl/workflow"] = {
411 "kind": "collection",
412 "portable_data_hash": "%s" % workflowcollection
415 packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
416 workflowpath = "/var/lib/cwl/workflow.json#main"
417 container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
421 if self.tool.tool.get("id", "").startswith("arvwf:"):
422 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
425 # --local means execute the workflow instead of submitting a container request
426 # --api=containers means use the containers API
427 # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
428 # --disable-validate because we already validated so don't need to do it again
429 # --eval-timeout is the timeout for javascript invocation
430 # --parallel-task-count is the number of threads to use for job submission
431 # --enable/disable-reuse sets desired job reuse
432 command = ["arvados-cwl-runner",
435 "--no-log-timestamps",
436 "--disable-validate",
437 "--eval-timeout=%s" % self.arvrunner.eval_timeout,
438 "--thread-count=%s" % self.arvrunner.thread_count,
439 "--enable-reuse" if self.enable_reuse else "--disable-reuse"]
442 command.append("--output-name=" + self.output_name)
443 container_req["output_name"] = self.output_name
446 command.append("--output-tags=" + self.output_tags)
448 if runtimeContext.debug:
449 command.append("--debug")
451 if runtimeContext.storage_classes != "default":
452 command.append("--storage-classes=" + runtimeContext.storage_classes)
455 command.append("--on-error=" + self.on_error)
457 if self.intermediate_output_ttl:
458 command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
460 if self.arvrunner.trash_intermediate:
461 command.append("--trash-intermediate")
463 if self.arvrunner.project_uuid:
464 command.append("--project-uuid="+self.arvrunner.project_uuid)
466 command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
468 container_req["command"] = command
473 def run(self, runtimeContext):
474 runtimeContext.keepprefix = "keep:"
475 job_spec = self.arvados_job_spec(runtimeContext)
476 if self.arvrunner.project_uuid:
477 job_spec["owner_uuid"] = self.arvrunner.project_uuid
479 if runtimeContext.submit_request_uuid:
480 response = self.arvrunner.api.container_requests().update(
481 uuid=runtimeContext.submit_request_uuid,
483 ).execute(num_retries=self.arvrunner.num_retries)
485 response = self.arvrunner.api.container_requests().create(
487 ).execute(num_retries=self.arvrunner.num_retries)
489 self.uuid = response["uuid"]
490 self.arvrunner.process_submitted(self)
492 logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"])
494 def done(self, record):
496 container = self.arvrunner.api.containers().get(
497 uuid=record["container_uuid"]
498 ).execute(num_retries=self.arvrunner.num_retries)
499 except Exception as e:
500 logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
501 self.arvrunner.output_callback({}, "permanentFail")
503 super(RunnerContainer, self).done(container)