3 # Implement cwl-runner interface for submitting and running work on Arvados, using
4 # either the Crunch jobs API or Crunch containers API.
14 from functools import partial
15 import pkg_resources # part of setuptools
17 from cwltool.errors import WorkflowException
19 import cwltool.workflow
25 from .arvcontainer import ArvadosContainer, RunnerContainer
26 from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate
27 from. runner import Runner, upload_instance
28 from .arvtool import ArvadosCommandTool
29 from .arvworkflow import ArvadosWorkflow, upload_workflow
30 from .fsaccess import CollectionFsAccess
31 from .perf import Perf
32 from .pathmapper import FinalOutputPathMapper
33 from ._version import __version__
35 from cwltool.pack import pack
36 from cwltool.process import shortname, UnsupportedRequirement, getListing
37 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
38 from cwltool.draft2tool import compute_checksums
39 from arvados.api import OrderedJsonModel
41 logger = logging.getLogger('arvados.cwl-runner')
42 metrics = logging.getLogger('arvados.cwl-runner.metrics')
43 logger.setLevel(logging.INFO)
46 class ArvCwlRunner(object):
47 """Execute a CWL tool or workflow, submit work (using either jobs or
48 containers API), wait for them to complete, and report output.
52 def __init__(self, api_client, work_api=None, keep_client=None, output_name=None, output_tags=None):
55 self.lock = threading.Lock()
56 self.cond = threading.Condition(self.lock)
57 self.final_output = None
58 self.final_status = None
62 self.stop_polling = threading.Event()
65 self.final_output_collection = None
66 self.output_name = output_name
67 self.output_tags = output_tags
68 self.project_uuid = None
70 if keep_client is not None:
71 self.keep_client = keep_client
73 self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
75 for api in ["jobs", "containers"]:
77 methods = self.api._rootDesc.get('resources')[api]['methods']
78 if ('httpMethod' in methods['create'] and
79 (work_api == api or work_api is None)):
86 raise Exception("No supported APIs")
88 raise Exception("Unsupported API '%s'" % work_api)
90 def arv_make_tool(self, toolpath_object, **kwargs):
91 kwargs["work_api"] = self.work_api
92 if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
93 return ArvadosCommandTool(self, toolpath_object, **kwargs)
94 elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
95 return ArvadosWorkflow(self, toolpath_object, **kwargs)
97 return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
99 def output_callback(self, out, processStatus):
100 if processStatus == "success":
101 logger.info("Overall process status is %s", processStatus)
103 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
104 body={"state": "Complete"}).execute(num_retries=self.num_retries)
106 logger.warn("Overall process status is %s", processStatus)
108 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
109 body={"state": "Failed"}).execute(num_retries=self.num_retries)
110 self.final_status = processStatus
111 self.final_output = out
113 def on_message(self, event):
114 if "object_uuid" in event:
115 if event["object_uuid"] in self.processes and event["event_type"] == "update":
116 if event["properties"]["new_attributes"]["state"] == "Running" and self.processes[event["object_uuid"]].running is False:
117 uuid = event["object_uuid"]
119 j = self.processes[uuid]
120 logger.info("Job %s (%s) is Running", j.name, uuid)
122 j.update_pipeline_component(event["properties"]["new_attributes"])
123 elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
124 uuid = event["object_uuid"]
127 j = self.processes[uuid]
128 txt = self.work_api[0].upper() + self.work_api[1:-1]
129 logger.info("%s %s (%s) is %s", txt, j.name, uuid, event["properties"]["new_attributes"]["state"])
130 with Perf(metrics, "done %s" % j.name):
131 j.done(event["properties"]["new_attributes"])
136 def poll_states(self):
137 """Poll status of jobs or containers listed in the processes dict.
139 Runs in a separate thread.
144 self.stop_polling.wait(15)
145 if self.stop_polling.is_set():
148 keys = self.processes.keys()
152 if self.work_api == "containers":
153 table = self.poll_api.containers()
154 elif self.work_api == "jobs":
155 table = self.poll_api.jobs()
158 proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries)
159 except Exception as e:
160 logger.warn("Error checking states on API server: %s", e)
163 for p in proc_states["items"]:
165 "object_uuid": p["uuid"],
166 "event_type": "update",
172 logger.error("Fatal error in state polling thread.", exc_info=(sys.exc_info()[1] if self.debug else False))
174 self.processes.clear()
178 self.stop_polling.set()
180 def get_uploaded(self):
181 return self.uploaded.copy()
183 def add_uploaded(self, src, pair):
184 self.uploaded[src] = pair
186 def check_writable(self, obj):
187 if isinstance(obj, dict):
188 if obj.get("writable"):
189 raise UnsupportedRequirement("InitialWorkDir feature 'writable: true' not supported")
190 for v in obj.itervalues():
191 self.check_writable(v)
192 if isinstance(obj, list):
194 self.check_writable(v)
196 def make_output_collection(self, name, tagsString, outputObj):
197 outputObj = copy.deepcopy(outputObj)
200 def capture(fileobj):
201 files.append(fileobj)
203 adjustDirObjs(outputObj, capture)
204 adjustFileObjs(outputObj, capture)
206 generatemapper = FinalOutputPathMapper(files, "", "", separateDirs=False)
208 final = arvados.collection.Collection(api_client=self.api,
209 keep_client=self.keep_client,
210 num_retries=self.num_retries)
213 for k,v in generatemapper.items():
214 if k.startswith("_:"):
215 if v.type == "Directory":
217 if v.type == "CreateFile":
218 with final.open(v.target, "wb") as f:
219 f.write(v.resolved.encode("utf-8"))
222 if not k.startswith("keep:"):
223 raise Exception("Output source is not in keep or a literal")
225 srccollection = sp[0][5:]
226 if srccollection not in srccollections:
228 srccollections[srccollection] = arvados.collection.CollectionReader(
231 keep_client=self.keep_client,
232 num_retries=self.num_retries)
233 except arvados.errors.ArgumentError as e:
234 logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e)
236 reader = srccollections[srccollection]
238 srcpath = "/".join(sp[1:]) if len(sp) > 1 else "."
239 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
241 logger.warn("While preparing output collection: %s", e)
243 def rewrite(fileobj):
244 fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
245 for k in ("basename", "listing", "contents"):
249 adjustDirObjs(outputObj, rewrite)
250 adjustFileObjs(outputObj, rewrite)
252 with final.open("cwl.output.json", "w") as f:
253 json.dump(outputObj, f, sort_keys=True, indent=4, separators=(',',': '))
255 final.save_new(name=name, owner_uuid=self.project_uuid, ensure_unique_name=True)
257 logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
258 final.api_response()["name"],
259 final.manifest_locator())
261 final_uuid = final.manifest_locator()
262 tags = tagsString.split(',')
264 self.api.links().create(body={
265 "head_uuid": final_uuid, "link_class": "tag", "name": tag
266 }).execute(num_retries=self.num_retries)
268 def finalcollection(fileobj):
269 fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"])
271 adjustDirObjs(outputObj, finalcollection)
272 adjustFileObjs(outputObj, finalcollection)
274 return (outputObj, final)
276 def set_crunch_output(self):
277 if self.work_api == "containers":
279 current = self.api.containers().current().execute(num_retries=self.num_retries)
280 self.api.containers().update(uuid=current['uuid'],
282 'output': self.final_output_collection.portable_data_hash(),
283 }).execute(num_retries=self.num_retries)
284 except Exception as e:
285 logger.info("Setting container output: %s", e)
286 elif self.work_api == "jobs" and "TASK_UUID" in os.environ:
287 self.api.job_tasks().update(uuid=os.environ["TASK_UUID"],
289 'output': self.final_output_collection.portable_data_hash(),
290 'success': self.final_status == "success",
292 }).execute(num_retries=self.num_retries)
294 def arv_executor(self, tool, job_order, **kwargs):
295 self.debug = kwargs.get("debug")
297 tool.visit(self.check_writable)
299 self.project_uuid = kwargs.get("project_uuid")
301 make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess,
303 keep_client=self.keep_client)
304 self.fs_access = make_fs_access(kwargs["basedir"])
306 existing_uuid = kwargs.get("update_workflow")
307 if existing_uuid or kwargs.get("create_workflow"):
308 if self.work_api == "jobs":
309 tmpl = RunnerTemplate(self, tool, job_order,
310 kwargs.get("enable_reuse"),
312 submit_runner_ram=kwargs.get("submit_runner_ram"))
314 # cwltool.main will write our return value to stdout.
317 return upload_workflow(self, tool, job_order,
320 submit_runner_ram=kwargs.get("submit_runner_ram"))
322 self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
324 kwargs["make_fs_access"] = make_fs_access
325 kwargs["enable_reuse"] = kwargs.get("enable_reuse")
326 kwargs["use_container"] = True
327 kwargs["tmpdir_prefix"] = "tmp"
328 kwargs["on_error"] = "continue"
329 kwargs["compute_checksum"] = kwargs.get("compute_checksum")
331 if self.work_api == "containers":
332 kwargs["outdir"] = "/var/spool/cwl"
333 kwargs["docker_outdir"] = "/var/spool/cwl"
334 kwargs["tmpdir"] = "/tmp"
335 kwargs["docker_tmpdir"] = "/tmp"
336 elif self.work_api == "jobs":
337 kwargs["outdir"] = "$(task.outdir)"
338 kwargs["docker_outdir"] = "$(task.outdir)"
339 kwargs["tmpdir"] = "$(task.tmpdir)"
341 upload_instance(self, shortname(tool.tool["id"]), tool, job_order)
344 if kwargs.get("submit"):
345 if self.work_api == "containers":
346 if tool.tool["class"] == "CommandLineTool":
347 runnerjob = tool.job(job_order,
348 self.output_callback,
351 runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name,
352 self.output_tags, submit_runner_ram=kwargs.get("submit_runner_ram"))
354 runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name,
355 self.output_tags, submit_runner_ram=kwargs.get("submit_runner_ram"))
357 if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers":
358 # Create pipeline for local run
359 self.pipeline = self.api.pipeline_instances().create(
361 "owner_uuid": self.project_uuid,
362 "name": shortname(tool.tool["id"]),
364 "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
365 logger.info("Pipeline instance %s", self.pipeline["uuid"])
367 if runnerjob and not kwargs.get("wait"):
368 runnerjob.run(wait=kwargs.get("wait"))
369 return runnerjob.uuid
371 self.poll_api = arvados.api('v1')
372 self.polling_thread = threading.Thread(target=self.poll_states)
373 self.polling_thread.start()
376 jobiter = iter((runnerjob,))
378 if "cwl_runner_job" in kwargs:
379 self.uuid = kwargs.get("cwl_runner_job").get('uuid')
380 jobiter = tool.job(job_order,
381 self.output_callback,
386 # Will continue to hold the lock for the duration of this code
387 # except when in cond.wait(), at which point on_message can update
388 # job state and process output callbacks.
390 loopperf = Perf(metrics, "jobiter")
392 for runnable in jobiter:
395 if self.stop_polling.is_set():
399 with Perf(metrics, "run"):
400 runnable.run(**kwargs)
405 logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
410 while self.processes:
413 except UnsupportedRequirement:
416 if sys.exc_info()[0] is KeyboardInterrupt:
417 logger.error("Interrupted, marking pipeline as failed")
419 logger.error("Execution failed: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
421 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
422 body={"state": "Failed"}).execute(num_retries=self.num_retries)
423 if runnerjob and runnerjob.uuid and self.work_api == "containers":
424 self.api.container_requests().update(uuid=runnerjob.uuid,
425 body={"priority": "0"}).execute(num_retries=self.num_retries)
428 self.stop_polling.set()
429 self.polling_thread.join()
431 if self.final_status == "UnsupportedRequirement":
432 raise UnsupportedRequirement("Check log for details.")
434 if self.final_output is None:
435 raise WorkflowException("Workflow did not return a result.")
437 if kwargs.get("submit") and isinstance(runnerjob, Runner):
438 logger.info("Final output collection %s", runnerjob.final_output)
440 if self.output_name is None:
441 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
442 if self.output_tags is None:
443 self.output_tags = ""
444 self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, self.output_tags, self.final_output)
445 self.set_crunch_output()
447 if self.final_status != "success":
448 raise WorkflowException("Workflow failed.")
450 if kwargs.get("compute_checksum"):
451 adjustDirObjs(self.final_output, partial(getListing, self.fs_access))
452 adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
454 return self.final_output
458 """Print version string of key packages for provenance and debugging."""
460 arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
461 arvpkg = pkg_resources.require("arvados-python-client")
462 cwlpkg = pkg_resources.require("cwltool")
464 return "%s %s %s, %s %s, %s %s" % (sys.argv[0], __version__, arvcwlpkg[0].version,
465 "arvados-python-client", arvpkg[0].version,
466 "cwltool", cwlpkg[0].version)
469 def arg_parser(): # type: () -> argparse.ArgumentParser
470 parser = argparse.ArgumentParser(description='Arvados executor for Common Workflow Language')
472 parser.add_argument("--basedir", type=str,
473 help="Base directory used to resolve relative references in the input, default to directory of input object file or current directory (if inputs piped/provided on command line).")
474 parser.add_argument("--outdir", type=str, default=os.path.abspath('.'),
475 help="Output directory, default current directory")
477 parser.add_argument("--eval-timeout",
478 help="Time to wait for a Javascript expression to evaluate before giving an error, default 20s.",
481 parser.add_argument("--version", action="store_true", help="Print version and exit")
483 exgroup = parser.add_mutually_exclusive_group()
484 exgroup.add_argument("--verbose", action="store_true", help="Default logging")
485 exgroup.add_argument("--quiet", action="store_true", help="Only print warnings and errors.")
486 exgroup.add_argument("--debug", action="store_true", help="Print even more logging")
488 parser.add_argument("--metrics", action="store_true", help="Print timing metrics")
490 parser.add_argument("--tool-help", action="store_true", help="Print command line help for tool")
492 exgroup = parser.add_mutually_exclusive_group()
493 exgroup.add_argument("--enable-reuse", action="store_true",
494 default=True, dest="enable_reuse",
496 exgroup.add_argument("--disable-reuse", action="store_false",
497 default=True, dest="enable_reuse",
500 parser.add_argument("--project-uuid", type=str, metavar="UUID", help="Project that will own the workflow jobs, if not provided, will go to home project.")
501 parser.add_argument("--output-name", type=str, help="Name to use for collection that stores the final output.", default=None)
502 parser.add_argument("--output-tags", type=str, help="Tags for the final output collection separated by commas, e.g., '--output-tags tag0,tag1,tag2'.", default=None)
503 parser.add_argument("--ignore-docker-for-reuse", action="store_true",
504 help="Ignore Docker image version when deciding whether to reuse past jobs.",
507 exgroup = parser.add_mutually_exclusive_group()
508 exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
509 default=True, dest="submit")
510 exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
511 default=True, dest="submit")
512 exgroup.add_argument("--create-template", action="store_true", help="(Deprecated) synonym for --create-workflow.",
513 dest="create_workflow")
514 exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow (if using the 'containers' API) or pipeline template (if using the 'jobs' API). See --api.")
515 exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update an existing Arvados workflow or pipeline template with the given UUID.")
517 exgroup = parser.add_mutually_exclusive_group()
518 exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
519 default=True, dest="wait")
520 exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
521 default=True, dest="wait")
523 parser.add_argument("--api", type=str,
524 default=None, dest="work_api",
525 help="Select work submission API, one of 'jobs' or 'containers'. Default is 'jobs' if that API is available, otherwise 'containers'.")
527 parser.add_argument("--compute-checksum", action="store_true", default=False,
528 help="Compute checksum of contents while collecting outputs",
529 dest="compute_checksum")
531 parser.add_argument("--submit-runner-ram", type=int,
532 help="RAM (in MiB) required for the workflow runner job.",
535 parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
536 parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
542 res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml')
543 cache["http://arvados.org/cwl"] = res.read()
545 document_loader, cwlnames, _, _ = cwltool.process.get_schema("v1.0")
546 _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache)
547 for n in extnames.names:
548 if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""):
549 cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, ""))
550 document_loader.idx["http://arvados.org/cwl#"+n] = {}
552 def main(args, stdout, stderr, api_client=None, keep_client=None):
553 parser = arg_parser()
555 job_order_object = None
556 arvargs = parser.parse_args(args)
558 if arvargs.update_workflow:
559 if arvargs.update_workflow.find('-7fd4e-') == 5:
560 want_api = 'containers'
561 elif arvargs.update_workflow.find('-p5p6p-') == 5:
565 if want_api and arvargs.work_api and want_api != arvargs.work_api:
566 logger.error('--update-workflow arg {!r} uses {!r} API, but --api={!r} specified'.format(
567 arvargs.update_workflow, want_api, arvargs.work_api))
569 arvargs.work_api = want_api
571 if (arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order:
572 job_order_object = ({}, "")
577 if api_client is None:
578 api_client=arvados.api('v1', model=OrderedJsonModel())
579 runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client, output_name=arvargs.output_name, output_tags=arvargs.output_tags)
580 except Exception as e:
585 logger.setLevel(logging.DEBUG)
588 logger.setLevel(logging.WARN)
589 logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
592 metrics.setLevel(logging.DEBUG)
593 logging.getLogger("cwltool.metrics").setLevel(logging.DEBUG)
595 arvargs.conformance_test = None
596 arvargs.use_container = True
597 arvargs.relax_path_checks = True
599 return cwltool.main.main(args=arvargs,
602 executor=runner.arv_executor,
603 makeTool=runner.arv_make_tool,
604 versionfunc=versionstring,
605 job_order_object=job_order_object,
606 make_fs_access=partial(CollectionFsAccess, api_client=api_client))