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         useruuid = self.api.users().current().execute()["uuid"]
 291         self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid
 292         self.pipeline = None
 293         make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess,
 294                                                                  api_client=self.api,
 295                                                                  keep_client=self.keep_client)
 296         self.fs_access = make_fs_access(kwargs["basedir"])
 297
 298         if kwargs.get("create_template"):
 299             tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse"))
 300             tmpl.save()
 301             # cwltool.main will write our return value to stdout.
 302             return tmpl.uuid
 303
 304         if kwargs.get("create_workflow") or kwargs.get("update_workflow"):
 305             return upload_workflow(self, tool, job_order, self.project_uuid, kwargs.get("update_workflow"))
 306
 307         self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
 308
 309         kwargs["make_fs_access"] = make_fs_access
 310         kwargs["enable_reuse"] = kwargs.get("enable_reuse")
 311         kwargs["use_container"] = True
 312         kwargs["tmpdir_prefix"] = "tmp"
 313         kwargs["on_error"] = "continue"
 314         kwargs["compute_checksum"] = kwargs.get("compute_checksum")
 315
 316         if self.work_api == "containers":
 317             kwargs["outdir"] = "/var/spool/cwl"
 318             kwargs["docker_outdir"] = "/var/spool/cwl"
 319             kwargs["tmpdir"] = "/tmp"
 320             kwargs["docker_tmpdir"] = "/tmp"
 321         elif self.work_api == "jobs":
 322             kwargs["outdir"] = "$(task.outdir)"
 323             kwargs["docker_outdir"] = "$(task.outdir)"
 324             kwargs["tmpdir"] = "$(task.tmpdir)"
 325
 326         upload_instance(self, shortname(tool.tool["id"]), tool, job_order)
 327
 328         runnerjob = None
 329         if kwargs.get("submit"):
 330             if self.work_api == "containers":
 331                 if tool.tool["class"] == "CommandLineTool":
 332                     runnerjob = tool.job(job_order,
 333                                          self.output_callback,
 334                                          **kwargs).next()
 335                 else:
 336                     runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name, self.output_tags)
 337             else:
 338                 runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name, self.output_tags)
 339
 340         if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers":
 341             # Create pipeline for local run
 342             self.pipeline = self.api.pipeline_instances().create(
 343                 body={
 344                     "owner_uuid": self.project_uuid,
 345                     "name": shortname(tool.tool["id"]),
 346                     "components": {},
 347                     "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
 348             logger.info("Pipeline instance %s", self.pipeline["uuid"])
 349
 350         if runnerjob and not kwargs.get("wait"):
 351             runnerjob.run(wait=kwargs.get("wait"))
 352             return runnerjob.uuid
 353
 354         self.poll_api = arvados.api('v1')
 355         self.polling_thread = threading.Thread(target=self.poll_states)
 356         self.polling_thread.start()
 357
 358         if runnerjob:
 359             jobiter = iter((runnerjob,))
 360         else:
 361             if "cwl_runner_job" in kwargs:
 362                 self.uuid = kwargs.get("cwl_runner_job").get('uuid')
 363             jobiter = tool.job(job_order,
 364                                self.output_callback,
 365                                **kwargs)
 366
 367         try:
 368             self.cond.acquire()
 369             # Will continue to hold the lock for the duration of this code
 370             # except when in cond.wait(), at which point on_message can update
 371             # job state and process output callbacks.
 372
 373             loopperf = Perf(metrics, "jobiter")
 374             loopperf.__enter__()
 375             for runnable in jobiter:
 376                 loopperf.__exit__()
 377                 if runnable:
 378                     with Perf(metrics, "run"):
 379                         runnable.run(**kwargs)
 380                 else:
 381                     if self.processes:
 382                         self.cond.wait(1)
 383                     else:
 384                         logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
 385                         break
 386                 loopperf.__enter__()
 387             loopperf.__exit__()
 388
 389             while self.processes:
 390                 self.cond.wait(1)
 391
 392         except UnsupportedRequirement:
 393             raise
 394         except:
 395             if sys.exc_info()[0] is KeyboardInterrupt:
 396                 logger.error("Interrupted, marking pipeline as failed")
 397             else:
 398                 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))
 399             if self.pipeline:
 400                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 401                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
 402             if runnerjob and runnerjob.uuid and self.work_api == "containers":
 403                 self.api.container_requests().update(uuid=runnerjob.uuid,
 404                                                      body={"priority": "0"}).execute(num_retries=self.num_retries)
 405         finally:
 406             self.cond.release()
 407             self.stop_polling.set()
 408             self.polling_thread.join()
 409
 410         if self.final_status == "UnsupportedRequirement":
 411             raise UnsupportedRequirement("Check log for details.")
 412
 413         if self.final_output is None:
 414             raise WorkflowException("Workflow did not return a result.")
 415
 416         if kwargs.get("submit") and isinstance(runnerjob, Runner):
 417             logger.info("Final output collection %s", runnerjob.final_output)
 418         else:
 419             if self.output_name is None:
 420                 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
 421             if self.output_tags is None:
 422                 self.output_tags = ""
 423             self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, self.output_tags, self.final_output)
 424             self.set_crunch_output()
 425
 426         if self.final_status != "success":
 427             raise WorkflowException("Workflow failed.")
 428
 429         if kwargs.get("compute_checksum"):
 430             adjustDirObjs(self.final_output, partial(getListing, self.fs_access))
 431             adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
 432
 433         return self.final_output
 434
 435
 436 def versionstring():
 437     """Print version string of key packages for provenance and debugging."""
 438
 439     arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
 440     arvpkg = pkg_resources.require("arvados-python-client")
 441     cwlpkg = pkg_resources.require("cwltool")
 442
 443     return "%s %s %s, %s %s, %s %s" % (sys.argv[0], __version__, arvcwlpkg[0].version,
 444                                     "arvados-python-client", arvpkg[0].version,
 445                                     "cwltool", cwlpkg[0].version)
 446
 447
 448 def arg_parser():  # type: () -> argparse.ArgumentParser
 449     parser = argparse.ArgumentParser(description='Arvados executor for Common Workflow Language')
 450
 451     parser.add_argument("--basedir", type=str,
 452                         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).")
 453     parser.add_argument("--outdir", type=str, default=os.path.abspath('.'),
 454                         help="Output directory, default current directory")
 455
 456     parser.add_argument("--eval-timeout",
 457                         help="Time to wait for a Javascript expression to evaluate before giving an error, default 20s.",
 458                         type=float,
 459                         default=20)
 460     parser.add_argument("--version", action="store_true", help="Print version and exit")
 461
 462     exgroup = parser.add_mutually_exclusive_group()
 463     exgroup.add_argument("--verbose", action="store_true", help="Default logging")
 464     exgroup.add_argument("--quiet", action="store_true", help="Only print warnings and errors.")
 465     exgroup.add_argument("--debug", action="store_true", help="Print even more logging")
 466
 467     parser.add_argument("--metrics", action="store_true", help="Print timing metrics")
 468
 469     parser.add_argument("--tool-help", action="store_true", help="Print command line help for tool")
 470
 471     exgroup = parser.add_mutually_exclusive_group()
 472     exgroup.add_argument("--enable-reuse", action="store_true",
 473                         default=True, dest="enable_reuse",
 474                         help="")
 475     exgroup.add_argument("--disable-reuse", action="store_false",
 476                         default=True, dest="enable_reuse",
 477                         help="")
 478
 479     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.")
 480     parser.add_argument("--output-name", type=str, help="Name to use for collection that stores the final output.", default=None)
 481     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)
 482     parser.add_argument("--ignore-docker-for-reuse", action="store_true",
 483                         help="Ignore Docker image version when deciding whether to reuse past jobs.",
 484                         default=False)
 485
 486     exgroup = parser.add_mutually_exclusive_group()
 487     exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
 488                         default=True, dest="submit")
 489     exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
 490                         default=True, dest="submit")
 491     exgroup.add_argument("--create-template", action="store_true", help="Create an Arvados pipeline template.")
 492     exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow.")
 493     exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update existing Arvados workflow with uuid.")
 494
 495     exgroup = parser.add_mutually_exclusive_group()
 496     exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
 497                         default=True, dest="wait")
 498     exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
 499                         default=True, dest="wait")
 500
 501     parser.add_argument("--api", type=str,
 502                         default=None, dest="work_api",
 503                         help="Select work submission API, one of 'jobs' or 'containers'. Default is 'jobs' if that API is available, otherwise 'containers'.")
 504
 505     parser.add_argument("--compute-checksum", action="store_true", default=False,
 506                         help="Compute checksum of contents while collecting outputs",
 507                         dest="compute_checksum")
 508
 509     parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
 510     parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
 511
 512     return parser
 513
 514 def add_arv_hints():
 515     cache = {}
 516     res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml')
 517     cache["http://arvados.org/cwl"] = res.read()
 518     res.close()
 519     document_loader, cwlnames, _, _ = cwltool.process.get_schema("v1.0")
 520     _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache)
 521     for n in extnames.names:
 522         if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""):
 523             cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, ""))
 524         document_loader.idx["http://arvados.org/cwl#"+n] = {}
 525
 526 def main(args, stdout, stderr, api_client=None, keep_client=None):
 527     parser = arg_parser()
 528
 529     job_order_object = None
 530     arvargs = parser.parse_args(args)
 531     if (arvargs.create_template or arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order:
 532         job_order_object = ({}, "")
 533
 534     add_arv_hints()
 535
 536     try:
 537         if api_client is None:
 538             api_client=arvados.api('v1', model=OrderedJsonModel())
 539         runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client, output_name=arvargs.output_name, output_tags=arvargs.output_tags)
 540     except Exception as e:
 541         logger.error(e)
 542         return 1
 543
 544     if arvargs.debug:
 545         logger.setLevel(logging.DEBUG)
 546
 547     if arvargs.quiet:
 548         logger.setLevel(logging.WARN)
 549         logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
 550
 551     if arvargs.metrics:
 552         metrics.setLevel(logging.DEBUG)
 553         logging.getLogger("cwltool.metrics").setLevel(logging.DEBUG)
 554
 555     arvargs.conformance_test = None
 556     arvargs.use_container = True
 557
 558     return cwltool.main.main(args=arvargs,
 559                              stdout=stdout,
 560                              stderr=stderr,
 561                              executor=runner.arv_executor,
 562                              makeTool=runner.arv_make_tool,
 563                              versionfunc=versionstring,
 564                              job_order_object=job_order_object,
 565                              make_fs_access=partial(CollectionFsAccess, api_client=api_client))