10498: --create-workflow and --create-template do the same thing, depending on -...
[arvados.git] / sdk / cwl / arvados_cwl / __init__.py
1 #!/usr/bin/env python
2
3 # Implement cwl-runner interface for submitting and running work on Arvados, using
4 # either the Crunch jobs API or Crunch containers API.
5
6 import argparse
7 import logging
8 import os
9 import sys
10 import threading
11 import hashlib
12 import copy
13 import json
14 from functools import partial
15 import pkg_resources  # part of setuptools
16
17 from cwltool.errors import WorkflowException
18 import cwltool.main
19 import cwltool.workflow
20 import schema_salad
21
22 import arvados
23 import arvados.config
24
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__
34
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
40
41 logger = logging.getLogger('arvados.cwl-runner')
42 metrics = logging.getLogger('arvados.cwl-runner.metrics')
43 logger.setLevel(logging.INFO)
44
45
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.
49
50     """
51
52     def __init__(self, api_client, work_api=None, keep_client=None, output_name=None, output_tags=None):
53         self.api = api_client
54         self.processes = {}
55         self.lock = threading.Lock()
56         self.cond = threading.Condition(self.lock)
57         self.final_output = None
58         self.final_status = None
59         self.uploaded = {}
60         self.num_retries = 4
61         self.uuid = None
62         self.stop_polling = threading.Event()
63         self.poll_api = None
64         self.pipeline = None
65         self.final_output_collection = None
66         self.output_name = output_name
67         self.output_tags = output_tags
68         self.project_uuid = None
69
70         if keep_client is not None:
71             self.keep_client = keep_client
72         else:
73             self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
74
75         for api in ["jobs", "containers"]:
76             try:
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)):
80                     self.work_api = api
81                     break
82             except KeyError:
83                 pass
84         if not self.work_api:
85             if work_api is None:
86                 raise Exception("No supported APIs")
87             else:
88                 raise Exception("Unsupported API '%s'" % work_api)
89
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)
96         else:
97             return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
98
99     def output_callback(self, out, processStatus):
100         if processStatus == "success":
101             logger.info("Overall process status is %s", processStatus)
102             if self.pipeline:
103                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
104                                                      body={"state": "Complete"}).execute(num_retries=self.num_retries)
105         else:
106             logger.warn("Overall process status is %s", processStatus)
107             if self.pipeline:
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
112
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"]
118                     with self.lock:
119                         j = self.processes[uuid]
120                         logger.info("Job %s (%s) is Running", j.name, uuid)
121                         j.running = True
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"]
125                     try:
126                         self.cond.acquire()
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"])
132                         self.cond.notify()
133                     finally:
134                         self.cond.release()
135
136     def poll_states(self):
137         """Poll status of jobs or containers listed in the processes dict.
138
139         Runs in a separate thread.
140         """
141
142         while True:
143             self.stop_polling.wait(15)
144             if self.stop_polling.is_set():
145                 break
146             with self.lock:
147                 keys = self.processes.keys()
148             if not keys:
149                 continue
150
151             if self.work_api == "containers":
152                 table = self.poll_api.containers()
153             elif self.work_api == "jobs":
154                 table = self.poll_api.jobs()
155
156             try:
157                 proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries)
158             except Exception as e:
159                 logger.warn("Error checking states on API server: %s", e)
160                 continue
161
162             for p in proc_states["items"]:
163                 self.on_message({
164                     "object_uuid": p["uuid"],
165                     "event_type": "update",
166                     "properties": {
167                         "new_attributes": p
168                     }
169                 })
170
171     def get_uploaded(self):
172         return self.uploaded.copy()
173
174     def add_uploaded(self, src, pair):
175         self.uploaded[src] = pair
176
177     def check_writable(self, obj):
178         if isinstance(obj, dict):
179             if obj.get("writable"):
180                 raise UnsupportedRequirement("InitialWorkDir feature 'writable: true' not supported")
181             for v in obj.itervalues():
182                 self.check_writable(v)
183         if isinstance(obj, list):
184             for v in obj:
185                 self.check_writable(v)
186
187     def make_output_collection(self, name, tagsString, outputObj):
188         outputObj = copy.deepcopy(outputObj)
189
190         files = []
191         def capture(fileobj):
192             files.append(fileobj)
193
194         adjustDirObjs(outputObj, capture)
195         adjustFileObjs(outputObj, capture)
196
197         generatemapper = FinalOutputPathMapper(files, "", "", separateDirs=False)
198
199         final = arvados.collection.Collection(api_client=self.api,
200                                               keep_client=self.keep_client,
201                                               num_retries=self.num_retries)
202
203         srccollections = {}
204         for k,v in generatemapper.items():
205             if k.startswith("_:"):
206                 if v.type == "Directory":
207                     continue
208                 if v.type == "CreateFile":
209                     with final.open(v.target, "wb") as f:
210                         f.write(v.resolved.encode("utf-8"))
211                     continue
212
213             if not k.startswith("keep:"):
214                 raise Exception("Output source is not in keep or a literal")
215             sp = k.split("/")
216             srccollection = sp[0][5:]
217             if srccollection not in srccollections:
218                 try:
219                     srccollections[srccollection] = arvados.collection.CollectionReader(
220                         srccollection,
221                         api_client=self.api,
222                         keep_client=self.keep_client,
223                         num_retries=self.num_retries)
224                 except arvados.errors.ArgumentError as e:
225                     logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e)
226                     raise
227             reader = srccollections[srccollection]
228             try:
229                 srcpath = "/".join(sp[1:]) if len(sp) > 1 else "."
230                 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
231             except IOError as e:
232                 logger.warn("While preparing output collection: %s", e)
233
234         def rewrite(fileobj):
235             fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
236             for k in ("basename", "listing", "contents"):
237                 if k in fileobj:
238                     del fileobj[k]
239
240         adjustDirObjs(outputObj, rewrite)
241         adjustFileObjs(outputObj, rewrite)
242
243         with final.open("cwl.output.json", "w") as f:
244             json.dump(outputObj, f, sort_keys=True, indent=4, separators=(',',': '))
245
246         final.save_new(name=name, owner_uuid=self.project_uuid, ensure_unique_name=True)
247
248         logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
249                     final.api_response()["name"],
250                     final.manifest_locator())
251
252         final_uuid = final.manifest_locator()
253         tags = tagsString.split(',')
254         for tag in tags:
255              self.api.links().create(body={
256                 "head_uuid": final_uuid, "link_class": "tag", "name": tag
257                 }).execute(num_retries=self.num_retries)
258
259         def finalcollection(fileobj):
260             fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"])
261
262         adjustDirObjs(outputObj, finalcollection)
263         adjustFileObjs(outputObj, finalcollection)
264
265         return (outputObj, final)
266
267     def set_crunch_output(self):
268         if self.work_api == "containers":
269             try:
270                 current = self.api.containers().current().execute(num_retries=self.num_retries)
271                 self.api.containers().update(uuid=current['uuid'],
272                                              body={
273                                                  'output': self.final_output_collection.portable_data_hash(),
274                                              }).execute(num_retries=self.num_retries)
275             except Exception as e:
276                 logger.info("Setting container output: %s", e)
277         elif self.work_api == "jobs" and "TASK_UUID" in os.environ:
278             self.api.job_tasks().update(uuid=os.environ["TASK_UUID"],
279                                    body={
280                                        'output': self.final_output_collection.portable_data_hash(),
281                                        'success': self.final_status == "success",
282                                        'progress':1.0
283                                    }).execute(num_retries=self.num_retries)
284
285     def arv_executor(self, tool, job_order, **kwargs):
286         self.debug = kwargs.get("debug")
287
288         tool.visit(self.check_writable)
289
290         self.project_uuid = kwargs.get("project_uuid")
291         self.pipeline = None
292         make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess,
293                                                                  api_client=self.api,
294                                                                  keep_client=self.keep_client)
295         self.fs_access = make_fs_access(kwargs["basedir"])
296
297         update_uuid = kwargs.get("update_workflow")
298         if update_uuid or kwargs.get("create_workflow"):
299             if self.work_api == "jobs":
300                 tmpl = RunnerTemplate(self, tool, job_order,
301                                       kwargs.get("enable_reuse"),
302                                       uuid=update_uuid)
303                 tmpl.save()
304                 # cwltool.main will write our return value to stdout.
305                 return tmpl.uuid
306             else:
307                 return upload_workflow(self, tool, job_order,
308                                        self.project_uuid,
309                                        uuid=update_uuid)
310
311         self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
312
313         kwargs["make_fs_access"] = make_fs_access
314         kwargs["enable_reuse"] = kwargs.get("enable_reuse")
315         kwargs["use_container"] = True
316         kwargs["tmpdir_prefix"] = "tmp"
317         kwargs["on_error"] = "continue"
318         kwargs["compute_checksum"] = kwargs.get("compute_checksum")
319
320         if self.work_api == "containers":
321             kwargs["outdir"] = "/var/spool/cwl"
322             kwargs["docker_outdir"] = "/var/spool/cwl"
323             kwargs["tmpdir"] = "/tmp"
324             kwargs["docker_tmpdir"] = "/tmp"
325         elif self.work_api == "jobs":
326             kwargs["outdir"] = "$(task.outdir)"
327             kwargs["docker_outdir"] = "$(task.outdir)"
328             kwargs["tmpdir"] = "$(task.tmpdir)"
329
330         upload_instance(self, shortname(tool.tool["id"]), tool, job_order)
331
332         runnerjob = None
333         if kwargs.get("submit"):
334             if self.work_api == "containers":
335                 if tool.tool["class"] == "CommandLineTool":
336                     runnerjob = tool.job(job_order,
337                                          self.output_callback,
338                                          **kwargs).next()
339                 else:
340                     runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name, self.output_tags)
341             else:
342                 runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name, self.output_tags)
343
344         if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers":
345             # Create pipeline for local run
346             self.pipeline = self.api.pipeline_instances().create(
347                 body={
348                     "owner_uuid": self.project_uuid,
349                     "name": shortname(tool.tool["id"]),
350                     "components": {},
351                     "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
352             logger.info("Pipeline instance %s", self.pipeline["uuid"])
353
354         if runnerjob and not kwargs.get("wait"):
355             runnerjob.run(wait=kwargs.get("wait"))
356             return runnerjob.uuid
357
358         self.poll_api = arvados.api('v1')
359         self.polling_thread = threading.Thread(target=self.poll_states)
360         self.polling_thread.start()
361
362         if runnerjob:
363             jobiter = iter((runnerjob,))
364         else:
365             if "cwl_runner_job" in kwargs:
366                 self.uuid = kwargs.get("cwl_runner_job").get('uuid')
367             jobiter = tool.job(job_order,
368                                self.output_callback,
369                                **kwargs)
370
371         try:
372             self.cond.acquire()
373             # Will continue to hold the lock for the duration of this code
374             # except when in cond.wait(), at which point on_message can update
375             # job state and process output callbacks.
376
377             loopperf = Perf(metrics, "jobiter")
378             loopperf.__enter__()
379             for runnable in jobiter:
380                 loopperf.__exit__()
381                 if runnable:
382                     with Perf(metrics, "run"):
383                         runnable.run(**kwargs)
384                 else:
385                     if self.processes:
386                         self.cond.wait(1)
387                     else:
388                         logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
389                         break
390                 loopperf.__enter__()
391             loopperf.__exit__()
392
393             while self.processes:
394                 self.cond.wait(1)
395
396         except UnsupportedRequirement:
397             raise
398         except:
399             if sys.exc_info()[0] is KeyboardInterrupt:
400                 logger.error("Interrupted, marking pipeline as failed")
401             else:
402                 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))
403             if self.pipeline:
404                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
405                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
406             if runnerjob and runnerjob.uuid and self.work_api == "containers":
407                 self.api.container_requests().update(uuid=runnerjob.uuid,
408                                                      body={"priority": "0"}).execute(num_retries=self.num_retries)
409         finally:
410             self.cond.release()
411             self.stop_polling.set()
412             self.polling_thread.join()
413
414         if self.final_status == "UnsupportedRequirement":
415             raise UnsupportedRequirement("Check log for details.")
416
417         if self.final_output is None:
418             raise WorkflowException("Workflow did not return a result.")
419
420         if kwargs.get("submit") and isinstance(runnerjob, Runner):
421             logger.info("Final output collection %s", runnerjob.final_output)
422         else:
423             if self.output_name is None:
424                 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
425             if self.output_tags is None:
426                 self.output_tags = ""
427             self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, self.output_tags, self.final_output)
428             self.set_crunch_output()
429
430         if self.final_status != "success":
431             raise WorkflowException("Workflow failed.")
432
433         if kwargs.get("compute_checksum"):
434             adjustDirObjs(self.final_output, partial(getListing, self.fs_access))
435             adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
436
437         return self.final_output
438
439
440 def versionstring():
441     """Print version string of key packages for provenance and debugging."""
442
443     arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
444     arvpkg = pkg_resources.require("arvados-python-client")
445     cwlpkg = pkg_resources.require("cwltool")
446
447     return "%s %s %s, %s %s, %s %s" % (sys.argv[0], __version__, arvcwlpkg[0].version,
448                                     "arvados-python-client", arvpkg[0].version,
449                                     "cwltool", cwlpkg[0].version)
450
451
452 def arg_parser():  # type: () -> argparse.ArgumentParser
453     parser = argparse.ArgumentParser(description='Arvados executor for Common Workflow Language')
454
455     parser.add_argument("--basedir", type=str,
456                         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).")
457     parser.add_argument("--outdir", type=str, default=os.path.abspath('.'),
458                         help="Output directory, default current directory")
459
460     parser.add_argument("--eval-timeout",
461                         help="Time to wait for a Javascript expression to evaluate before giving an error, default 20s.",
462                         type=float,
463                         default=20)
464     parser.add_argument("--version", action="store_true", help="Print version and exit")
465
466     exgroup = parser.add_mutually_exclusive_group()
467     exgroup.add_argument("--verbose", action="store_true", help="Default logging")
468     exgroup.add_argument("--quiet", action="store_true", help="Only print warnings and errors.")
469     exgroup.add_argument("--debug", action="store_true", help="Print even more logging")
470
471     parser.add_argument("--metrics", action="store_true", help="Print timing metrics")
472
473     parser.add_argument("--tool-help", action="store_true", help="Print command line help for tool")
474
475     exgroup = parser.add_mutually_exclusive_group()
476     exgroup.add_argument("--enable-reuse", action="store_true",
477                         default=True, dest="enable_reuse",
478                         help="")
479     exgroup.add_argument("--disable-reuse", action="store_false",
480                         default=True, dest="enable_reuse",
481                         help="")
482
483     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.")
484     parser.add_argument("--output-name", type=str, help="Name to use for collection that stores the final output.", default=None)
485     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)
486     parser.add_argument("--ignore-docker-for-reuse", action="store_true",
487                         help="Ignore Docker image version when deciding whether to reuse past jobs.",
488                         default=False)
489
490     exgroup = parser.add_mutually_exclusive_group()
491     exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
492                         default=True, dest="submit")
493     exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
494                         default=True, dest="submit")
495     exgroup.add_argument("--create-template", action="store_true", help="(Deprecated) synonym for --create-workflow.",
496                          dest="create_workflow")
497     exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow or pipeline template (depending on selected API, see --api).")
498     exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update an existing Arvados workflow or pipeline template with the given UUID.")
499
500     exgroup = parser.add_mutually_exclusive_group()
501     exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
502                         default=True, dest="wait")
503     exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
504                         default=True, dest="wait")
505
506     parser.add_argument("--api", type=str,
507                         default=None, dest="work_api",
508                         help="Select work submission API, one of 'jobs' or 'containers'. Default is 'jobs' if that API is available, otherwise 'containers'.")
509
510     parser.add_argument("--compute-checksum", action="store_true", default=False,
511                         help="Compute checksum of contents while collecting outputs",
512                         dest="compute_checksum")
513
514     parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
515     parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
516
517     return parser
518
519 def add_arv_hints():
520     cache = {}
521     res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml')
522     cache["http://arvados.org/cwl"] = res.read()
523     res.close()
524     document_loader, cwlnames, _, _ = cwltool.process.get_schema("v1.0")
525     _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache)
526     for n in extnames.names:
527         if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""):
528             cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, ""))
529         document_loader.idx["http://arvados.org/cwl#"+n] = {}
530
531 def main(args, stdout, stderr, api_client=None, keep_client=None):
532     parser = arg_parser()
533
534     job_order_object = None
535     arvargs = parser.parse_args(args)
536
537     if arvargs.update_workflow:
538         if arvargs.update_workflow.find('-7fd4e-') == 5:
539             want_api = 'containers'
540         elif arvargs.update_workflow.find('-p5p6p-') == 5:
541             want_api = 'jobs'
542         else:
543             want_api = None
544         if want_api and arvargs.work_api and want_api != arvargs.work_api:
545             logger.error('--update-workflow arg {!r} uses {!r} API, but --api={!r} specified'.format(
546                 arvargs.update_workflow, want_api, arvargs.work_api))
547             return 1
548         arvargs.work_api = want_api
549
550     if (arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order:
551         job_order_object = ({}, "")
552
553     add_arv_hints()
554
555     try:
556         if api_client is None:
557             api_client=arvados.api('v1', model=OrderedJsonModel())
558         runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client, output_name=arvargs.output_name, output_tags=arvargs.output_tags)
559     except Exception as e:
560         logger.error(e)
561         return 1
562
563     if arvargs.debug:
564         logger.setLevel(logging.DEBUG)
565
566     if arvargs.quiet:
567         logger.setLevel(logging.WARN)
568         logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
569
570     if arvargs.metrics:
571         metrics.setLevel(logging.DEBUG)
572         logging.getLogger("cwltool.metrics").setLevel(logging.DEBUG)
573
574     arvargs.conformance_test = None
575     arvargs.use_container = True
576
577     return cwltool.main.main(args=arvargs,
578                              stdout=stdout,
579                              stderr=stderr,
580                              executor=runner.arv_executor,
581                              makeTool=runner.arv_make_tool,
582                              versionfunc=versionstring,
583                              job_order_object=job_order_object,
584                              make_fs_access=partial(CollectionFsAccess, api_client=api_client))