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):
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.project_uuid = None
69 if keep_client is not None:
70 self.keep_client = keep_client
72 self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
74 for api in ["jobs", "containers"]:
76 methods = self.api._rootDesc.get('resources')[api]['methods']
77 if ('httpMethod' in methods['create'] and
78 (work_api == api or work_api is None)):
85 raise Exception("No supported APIs")
87 raise Exception("Unsupported API '%s'" % work_api)
89 def arv_make_tool(self, toolpath_object, **kwargs):
90 kwargs["work_api"] = self.work_api
91 if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
92 return ArvadosCommandTool(self, toolpath_object, **kwargs)
93 elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
94 return ArvadosWorkflow(self, toolpath_object, **kwargs)
96 return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
98 def output_callback(self, out, processStatus):
99 if processStatus == "success":
100 logger.info("Overall process status is %s", processStatus)
102 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
103 body={"state": "Complete"}).execute(num_retries=self.num_retries)
105 logger.warn("Overall process status is %s", processStatus)
107 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
108 body={"state": "Failed"}).execute(num_retries=self.num_retries)
109 self.final_status = processStatus
110 self.final_output = out
112 def on_message(self, event):
113 if "object_uuid" in event:
114 if event["object_uuid"] in self.processes and event["event_type"] == "update":
115 if event["properties"]["new_attributes"]["state"] == "Running" and self.processes[event["object_uuid"]].running is False:
116 uuid = event["object_uuid"]
118 j = self.processes[uuid]
119 logger.info("Job %s (%s) is Running", j.name, uuid)
121 j.update_pipeline_component(event["properties"]["new_attributes"])
122 elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
123 uuid = event["object_uuid"]
126 j = self.processes[uuid]
127 txt = self.work_api[0].upper() + self.work_api[1:-1]
128 logger.info("%s %s (%s) is %s", txt, j.name, uuid, event["properties"]["new_attributes"]["state"])
129 with Perf(metrics, "done %s" % j.name):
130 j.done(event["properties"]["new_attributes"])
135 def poll_states(self):
136 """Poll status of jobs or containers listed in the processes dict.
138 Runs in a separate thread.
142 self.stop_polling.wait(15)
143 if self.stop_polling.is_set():
146 keys = self.processes.keys()
150 if self.work_api == "containers":
151 table = self.poll_api.containers()
152 elif self.work_api == "jobs":
153 table = self.poll_api.jobs()
156 proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries)
157 except Exception as e:
158 logger.warn("Error checking states on API server: %s", e)
161 for p in proc_states["items"]:
163 "object_uuid": p["uuid"],
164 "event_type": "update",
170 def get_uploaded(self):
171 return self.uploaded.copy()
173 def add_uploaded(self, src, pair):
174 self.uploaded[src] = pair
176 def check_writable(self, obj):
177 if isinstance(obj, dict):
178 if obj.get("writable"):
179 raise UnsupportedRequirement("InitialWorkDir feature 'writable: true' not supported")
180 for v in obj.itervalues():
181 self.check_writable(v)
182 if isinstance(obj, list):
184 self.check_writable(v)
186 def make_output_collection(self, name, outputObj):
187 outputObj = copy.deepcopy(outputObj)
190 def capture(fileobj):
191 files.append(fileobj)
193 adjustDirObjs(outputObj, capture)
194 adjustFileObjs(outputObj, capture)
196 generatemapper = FinalOutputPathMapper(files, "", "", separateDirs=False)
198 final = arvados.collection.Collection(api_client=self.api,
199 keep_client=self.keep_client,
200 num_retries=self.num_retries)
203 for k,v in generatemapper.items():
205 srccollection = sp[0][5:]
206 if srccollection not in srccollections:
207 srccollections[srccollection] = arvados.collection.CollectionReader(
210 keep_client=self.keep_client,
211 num_retries=self.num_retries)
212 reader = srccollections[srccollection]
214 srcpath = "/".join(sp[1:]) if len(sp) > 1 else "."
215 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
217 logger.warn("While preparing output collection: %s", e)
219 def rewrite(fileobj):
220 fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
221 for k in ("basename", "size", "listing"):
225 adjustDirObjs(outputObj, rewrite)
226 adjustFileObjs(outputObj, rewrite)
228 with final.open("cwl.output.json", "w") as f:
229 json.dump(outputObj, f, sort_keys=True, indent=4, separators=(',',': '))
231 final.save_new(name=name, owner_uuid=self.project_uuid, ensure_unique_name=True)
233 logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
234 final.api_response()["name"],
235 final.manifest_locator())
237 self.final_output_collection = final
239 def set_crunch_output(self):
240 if self.work_api == "containers":
242 current = self.api.containers().current().execute(num_retries=self.num_retries)
243 self.api.containers().update(uuid=current['uuid'],
245 'output': self.final_output_collection.portable_data_hash(),
246 }).execute(num_retries=self.num_retries)
247 except Exception as e:
248 logger.info("Setting container output: %s", e)
249 elif self.work_api == "jobs" and "TASK_UUID" in os.environ:
250 self.api.job_tasks().update(uuid=os.environ["TASK_UUID"],
252 'output': self.final_output_collection.portable_data_hash(),
253 'success': self.final_status == "success",
255 }).execute(num_retries=self.num_retries)
257 def arv_executor(self, tool, job_order, **kwargs):
258 self.debug = kwargs.get("debug")
260 tool.visit(self.check_writable)
262 useruuid = self.api.users().current().execute()["uuid"]
263 self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid
265 make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess,
267 keep_client=self.keep_client)
268 self.fs_access = make_fs_access(kwargs["basedir"])
270 if kwargs.get("create_template"):
271 tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse"))
273 # cwltool.main will write our return value to stdout.
276 if kwargs.get("create_workflow") or kwargs.get("update_workflow"):
277 return upload_workflow(self, tool, job_order, self.project_uuid, kwargs.get("update_workflow"))
279 self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
281 kwargs["make_fs_access"] = make_fs_access
282 kwargs["enable_reuse"] = kwargs.get("enable_reuse")
283 kwargs["use_container"] = True
284 kwargs["tmpdir_prefix"] = "tmp"
285 kwargs["on_error"] = "continue"
286 kwargs["compute_checksum"] = kwargs.get("compute_checksum")
288 if self.work_api == "containers":
289 kwargs["outdir"] = "/var/spool/cwl"
290 kwargs["docker_outdir"] = "/var/spool/cwl"
291 kwargs["tmpdir"] = "/tmp"
292 kwargs["docker_tmpdir"] = "/tmp"
293 elif self.work_api == "jobs":
294 kwargs["outdir"] = "$(task.outdir)"
295 kwargs["docker_outdir"] = "$(task.outdir)"
296 kwargs["tmpdir"] = "$(task.tmpdir)"
298 upload_instance(self, shortname(tool.tool["id"]), tool, job_order)
301 if kwargs.get("submit"):
302 if self.work_api == "containers":
303 if tool.tool["class"] == "CommandLineTool":
304 runnerjob = tool.job(job_order,
305 self.output_callback,
308 runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name)
310 runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name)
312 if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers":
313 # Create pipeline for local run
314 self.pipeline = self.api.pipeline_instances().create(
316 "owner_uuid": self.project_uuid,
317 "name": shortname(tool.tool["id"]),
319 "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
320 logger.info("Pipeline instance %s", self.pipeline["uuid"])
322 if runnerjob and not kwargs.get("wait"):
323 runnerjob.run(wait=kwargs.get("wait"))
324 return runnerjob.uuid
326 self.poll_api = arvados.api('v1')
327 self.polling_thread = threading.Thread(target=self.poll_states)
328 self.polling_thread.start()
331 jobiter = iter((runnerjob,))
333 if "cwl_runner_job" in kwargs:
334 self.uuid = kwargs.get("cwl_runner_job").get('uuid')
335 jobiter = tool.job(job_order,
336 self.output_callback,
341 # Will continue to hold the lock for the duration of this code
342 # except when in cond.wait(), at which point on_message can update
343 # job state and process output callbacks.
345 loopperf = Perf(metrics, "jobiter")
347 for runnable in jobiter:
350 with Perf(metrics, "run"):
351 runnable.run(**kwargs)
356 logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
361 while self.processes:
364 except UnsupportedRequirement:
367 if sys.exc_info()[0] is KeyboardInterrupt:
368 logger.error("Interrupted, marking pipeline as failed")
370 logger.error("Caught unhandled exception, marking pipeline as failed. Error was: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
372 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
373 body={"state": "Failed"}).execute(num_retries=self.num_retries)
374 if runnerjob and runnerjob.uuid and self.work_api == "containers":
375 self.api.container_requests().update(uuid=runnerjob.uuid,
376 body={"priority": "0"}).execute(num_retries=self.num_retries)
379 self.stop_polling.set()
380 self.polling_thread.join()
382 if self.final_status == "UnsupportedRequirement":
383 raise UnsupportedRequirement("Check log for details.")
385 if self.final_output is None:
386 raise WorkflowException("Workflow did not return a result.")
388 if kwargs.get("submit") and isinstance(runnerjob, Runner):
389 logger.info("Final output collection %s", runnerjob.final_output)
391 if self.output_name is None:
392 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
393 self.make_output_collection(self.output_name, self.final_output)
394 self.set_crunch_output()
396 if self.final_status != "success":
397 raise WorkflowException("Workflow failed.")
399 if kwargs.get("compute_checksum"):
400 adjustDirObjs(self.final_output, partial(getListing, self.fs_access))
401 adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
403 return self.final_output
407 """Print version string of key packages for provenance and debugging."""
409 arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
410 arvpkg = pkg_resources.require("arvados-python-client")
411 cwlpkg = pkg_resources.require("cwltool")
413 return "%s %s %s, %s %s, %s %s" % (sys.argv[0], __version__, arvcwlpkg[0].version,
414 "arvados-python-client", arvpkg[0].version,
415 "cwltool", cwlpkg[0].version)
418 def arg_parser(): # type: () -> argparse.ArgumentParser
419 parser = argparse.ArgumentParser(description='Arvados executor for Common Workflow Language')
421 parser.add_argument("--basedir", type=str,
422 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).")
423 parser.add_argument("--outdir", type=str, default=os.path.abspath('.'),
424 help="Output directory, default current directory")
426 parser.add_argument("--eval-timeout",
427 help="Time to wait for a Javascript expression to evaluate before giving an error, default 20s.",
430 parser.add_argument("--version", action="store_true", help="Print version and exit")
432 exgroup = parser.add_mutually_exclusive_group()
433 exgroup.add_argument("--verbose", action="store_true", help="Default logging")
434 exgroup.add_argument("--quiet", action="store_true", help="Only print warnings and errors.")
435 exgroup.add_argument("--debug", action="store_true", help="Print even more logging")
437 parser.add_argument("--metrics", action="store_true", help="Print timing metrics")
439 parser.add_argument("--tool-help", action="store_true", help="Print command line help for tool")
441 exgroup = parser.add_mutually_exclusive_group()
442 exgroup.add_argument("--enable-reuse", action="store_true",
443 default=True, dest="enable_reuse",
445 exgroup.add_argument("--disable-reuse", action="store_false",
446 default=True, dest="enable_reuse",
449 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.")
450 parser.add_argument("--output-name", type=str, help="Name to use for collection that stores the final output.", default=None)
451 parser.add_argument("--ignore-docker-for-reuse", action="store_true",
452 help="Ignore Docker image version when deciding whether to reuse past jobs.",
455 exgroup = parser.add_mutually_exclusive_group()
456 exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
457 default=True, dest="submit")
458 exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
459 default=True, dest="submit")
460 exgroup.add_argument("--create-template", action="store_true", help="Create an Arvados pipeline template.")
461 exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow.")
462 exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update existing Arvados workflow with uuid.")
464 exgroup = parser.add_mutually_exclusive_group()
465 exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
466 default=True, dest="wait")
467 exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
468 default=True, dest="wait")
470 parser.add_argument("--api", type=str,
471 default=None, dest="work_api",
472 help="Select work submission API, one of 'jobs' or 'containers'. Default is 'jobs' if that API is available, otherwise 'containers'.")
474 parser.add_argument("--compute-checksum", action="store_true", default=False,
475 help="Compute checksum of contents while collecting outputs",
476 dest="compute_checksum")
478 parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
479 parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
485 res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml')
486 cache["http://arvados.org/cwl"] = res.read()
488 document_loader, cwlnames, _, _ = cwltool.process.get_schema("v1.0")
489 _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache)
490 for n in extnames.names:
491 if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""):
492 cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, ""))
493 document_loader.idx["http://arvados.org/cwl#"+n] = {}
495 def main(args, stdout, stderr, api_client=None, keep_client=None):
496 parser = arg_parser()
498 job_order_object = None
499 arvargs = parser.parse_args(args)
500 if (arvargs.create_template or arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order:
501 job_order_object = ({}, "")
506 if api_client is None:
507 api_client=arvados.api('v1', model=OrderedJsonModel())
508 runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client, output_name=arvargs.output_name)
509 except Exception as e:
514 logger.setLevel(logging.DEBUG)
517 logger.setLevel(logging.WARN)
518 logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
521 metrics.setLevel(logging.DEBUG)
522 logging.getLogger("cwltool.metrics").setLevel(logging.DEBUG)
524 arvargs.conformance_test = None
525 arvargs.use_container = True
527 return cwltool.main.main(args=arvargs,
530 executor=runner.arv_executor,
531 makeTool=runner.arv_make_tool,
532 versionfunc=versionstring,
533 job_order_object=job_order_object,
534 make_fs_access=partial(CollectionFsAccess, api_client=api_client))