sdk/cwl/arvados_cwl/executor.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 from __future__ import division
   6 from builtins import next
   7 from builtins import object
   8 from builtins import str
   9 from future.utils import viewvalues, viewitems
  10
  11 import argparse
  12 import logging
  13 import os
  14 import sys
  15 import threading
  16 import copy
  17 import json
  18 import re
  19 from functools import partial
  20 import time
  21 import urllib
  22
  23 from cwltool.errors import WorkflowException
  24 import cwltool.workflow
  25 from schema_salad.sourceline import SourceLine
  26 import schema_salad.validate as validate
  27
  28 import arvados
  29 import arvados.config
  30 from arvados.keep import KeepClient
  31 from arvados.errors import ApiError
  32
  33 import arvados_cwl.util
  34 from .arvcontainer import RunnerContainer
  35 from .runner import Runner, upload_docker, upload_job_order, upload_workflow_deps
  36 from .arvtool import ArvadosCommandTool, validate_cluster_target, ArvadosExpressionTool
  37 from .arvworkflow import ArvadosWorkflow, upload_workflow
  38 from .fsaccess import CollectionFsAccess, CollectionFetcher, collectionResolver, CollectionCache, pdh_size
  39 from .perf import Perf
  40 from .pathmapper import NoFollowPathMapper
  41 from cwltool.task_queue import TaskQueue
  42 from .context import ArvLoadingContext, ArvRuntimeContext
  43 from ._version import __version__
  44
  45 from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema
  46 from cwltool.utils import adjustFileObjs, adjustDirObjs, get_listing, visit_class, aslist
  47 from cwltool.command_line_tool import compute_checksums
  48 from cwltool.load_tool import load_tool
  49
  50 logger = logging.getLogger('arvados.cwl-runner')
  51 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  52
  53 DEFAULT_PRIORITY = 500
  54
  55 class RuntimeStatusLoggingHandler(logging.Handler):
  56     """
  57     Intercepts logging calls and report them as runtime statuses on runner
  58     containers.
  59     """
  60     def __init__(self, runtime_status_update_func):
  61         super(RuntimeStatusLoggingHandler, self).__init__()
  62         self.runtime_status_update = runtime_status_update_func
  63         self.updatingRuntimeStatus = False
  64
  65     def emit(self, record):
  66         kind = None
  67         if record.levelno >= logging.ERROR:
  68             kind = 'error'
  69         elif record.levelno >= logging.WARNING:
  70             kind = 'warning'
  71         if kind is not None and self.updatingRuntimeStatus is not True:
  72             self.updatingRuntimeStatus = True
  73             try:
  74                 log_msg = record.getMessage()
  75                 if '\n' in log_msg:
  76                     # If the logged message is multi-line, use its first line as status
  77                     # and the rest as detail.
  78                     status, detail = log_msg.split('\n', 1)
  79                     self.runtime_status_update(
  80                         kind,
  81                         "%s: %s" % (record.name, status),
  82                         detail
  83                     )
  84                 else:
  85                     self.runtime_status_update(
  86                         kind,
  87                         "%s: %s" % (record.name, record.getMessage())
  88                     )
  89             finally:
  90                 self.updatingRuntimeStatus = False
  91
  92
  93 class ArvCwlExecutor(object):
  94     """Execute a CWL tool or workflow, submit work (using containers API),
  95     wait for them to complete, and report output.
  96
  97     """
  98
  99     def __init__(self, api_client,
 100                  arvargs=None,
 101                  keep_client=None,
 102                  num_retries=4,
 103                  thread_count=4,
 104                  stdout=sys.stdout):
 105
 106         if arvargs is None:
 107             arvargs = argparse.Namespace()
 108             arvargs.work_api = None
 109             arvargs.output_name = None
 110             arvargs.output_tags = None
 111             arvargs.thread_count = 1
 112             arvargs.collection_cache_size = None
 113
 114         self.api = api_client
 115         self.processes = {}
 116         self.workflow_eval_lock = threading.Condition(threading.RLock())
 117         self.final_output = None
 118         self.final_status = None
 119         self.num_retries = num_retries
 120         self.uuid = None
 121         self.stop_polling = threading.Event()
 122         self.poll_api = None
 123         self.pipeline = None
 124         self.final_output_collection = None
 125         self.output_name = arvargs.output_name
 126         self.output_tags = arvargs.output_tags
 127         self.project_uuid = None
 128         self.intermediate_output_ttl = 0
 129         self.intermediate_output_collections = []
 130         self.trash_intermediate = False
 131         self.thread_count = arvargs.thread_count
 132         self.poll_interval = 12
 133         self.loadingContext = None
 134         self.should_estimate_cache_size = True
 135         self.fs_access = None
 136         self.secret_store = None
 137         self.stdout = stdout
 138
 139         if keep_client is not None:
 140             self.keep_client = keep_client
 141         else:
 142             self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
 143
 144         if arvargs.collection_cache_size:
 145             collection_cache_size = arvargs.collection_cache_size*1024*1024
 146             self.should_estimate_cache_size = False
 147         else:
 148             collection_cache_size = 256*1024*1024
 149
 150         self.collection_cache = CollectionCache(self.api, self.keep_client, self.num_retries,
 151                                                 cap=collection_cache_size)
 152
 153         self.fetcher_constructor = partial(CollectionFetcher,
 154                                            api_client=self.api,
 155                                            fs_access=CollectionFsAccess("", collection_cache=self.collection_cache),
 156                                            num_retries=self.num_retries)
 157
 158         self.work_api = None
 159         expected_api = ["containers"]
 160         for api in expected_api:
 161             try:
 162                 methods = self.api._rootDesc.get('resources')[api]['methods']
 163                 if ('httpMethod' in methods['create'] and
 164                     (arvargs.work_api == api or arvargs.work_api is None)):
 165                     self.work_api = api
 166                     break
 167             except KeyError:
 168                 pass
 169
 170         if not self.work_api:
 171             if arvargs.work_api is None:
 172                 raise Exception("No supported APIs")
 173             else:
 174                 raise Exception("Unsupported API '%s', expected one of %s" % (arvargs.work_api, expected_api))
 175
 176         if self.work_api == "jobs":
 177             logger.error("""
 178 *******************************
 179 The 'jobs' API is no longer supported.
 180 *******************************""")
 181             exit(1)
 182
 183         self.loadingContext = ArvLoadingContext(vars(arvargs))
 184         self.loadingContext.fetcher_constructor = self.fetcher_constructor
 185         self.loadingContext.resolver = partial(collectionResolver, self.api, num_retries=self.num_retries)
 186         self.loadingContext.construct_tool_object = self.arv_make_tool
 187
 188         # Add a custom logging handler to the root logger for runtime status reporting
 189         # if running inside a container
 190         if arvados_cwl.util.get_current_container(self.api, self.num_retries, logger):
 191             root_logger = logging.getLogger('')
 192
 193             # Remove existing RuntimeStatusLoggingHandlers if they exist
 194             handlers = [h for h in root_logger.handlers if not isinstance(h, RuntimeStatusLoggingHandler)]
 195             root_logger.handlers = handlers
 196
 197             handler = RuntimeStatusLoggingHandler(self.runtime_status_update)
 198             root_logger.addHandler(handler)
 199
 200         self.runtimeContext = ArvRuntimeContext(vars(arvargs))
 201         self.runtimeContext.make_fs_access = partial(CollectionFsAccess,
 202                                                      collection_cache=self.collection_cache)
 203
 204         validate_cluster_target(self, self.runtimeContext)
 205
 206
 207     def arv_make_tool(self, toolpath_object, loadingContext):
 208         if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
 209             return ArvadosCommandTool(self, toolpath_object, loadingContext)
 210         elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
 211             return ArvadosWorkflow(self, toolpath_object, loadingContext)
 212         elif "class" in toolpath_object and toolpath_object["class"] == "ExpressionTool":
 213             return ArvadosExpressionTool(self, toolpath_object, loadingContext)
 214         else:
 215             raise Exception("Unknown tool %s" % toolpath_object.get("class"))
 216
 217     def output_callback(self, out, processStatus):
 218         with self.workflow_eval_lock:
 219             if processStatus == "success":
 220                 logger.info("Overall process status is %s", processStatus)
 221                 state = "Complete"
 222             else:
 223                 logger.error("Overall process status is %s", processStatus)
 224                 state = "Failed"
 225             if self.pipeline:
 226                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 227                                                         body={"state": state}).execute(num_retries=self.num_retries)
 228             self.final_status = processStatus
 229             self.final_output = out
 230             self.workflow_eval_lock.notifyAll()
 231
 232
 233     def start_run(self, runnable, runtimeContext):
 234         self.task_queue.add(partial(runnable.run, runtimeContext),
 235                             self.workflow_eval_lock, self.stop_polling)
 236
 237     def process_submitted(self, container):
 238         with self.workflow_eval_lock:
 239             self.processes[container.uuid] = container
 240
 241     def process_done(self, uuid, record):
 242         with self.workflow_eval_lock:
 243             j = self.processes[uuid]
 244             logger.info("%s %s is %s", self.label(j), uuid, record["state"])
 245             self.task_queue.add(partial(j.done, record),
 246                                 self.workflow_eval_lock, self.stop_polling)
 247             del self.processes[uuid]
 248
 249     def runtime_status_update(self, kind, message, detail=None):
 250         """
 251         Updates the runtime_status field on the runner container.
 252         Called when there's a need to report errors, warnings or just
 253         activity statuses, for example in the RuntimeStatusLoggingHandler.
 254         """
 255         with self.workflow_eval_lock:
 256             current = None
 257             try:
 258                 current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 259             except Exception as e:
 260                 logger.info("Couldn't get current container: %s", e)
 261             if current is None:
 262                 return
 263             runtime_status = current.get('runtime_status', {})
 264             if kind in ('error', 'warning'):
 265                 updatemessage = runtime_status.get(kind, "")
 266                 if not updatemessage:
 267                     updatemessage = message
 268
 269                 # Subsequent messages tacked on in detail
 270                 updatedetail = runtime_status.get(kind+'Detail', "")
 271                 maxlines = 40
 272                 if updatedetail.count("\n") < maxlines:
 273                     if updatedetail:
 274                         updatedetail += "\n"
 275                     updatedetail += message + "\n"
 276
 277                     if detail:
 278                         updatedetail += detail + "\n"
 279
 280                     if updatedetail.count("\n") >= maxlines:
 281                         updatedetail += "\nSome messages may have been omitted.  Check the full log."
 282
 283                 runtime_status.update({
 284                     kind: updatemessage,
 285                     kind+'Detail': updatedetail,
 286                 })
 287             else:
 288                 # Ignore any other status kind
 289                 return
 290             try:
 291                 self.api.containers().update(uuid=current['uuid'],
 292                                             body={
 293                                                 'runtime_status': runtime_status,
 294                                             }).execute(num_retries=self.num_retries)
 295             except Exception as e:
 296                 logger.info("Couldn't update runtime_status: %s", e)
 297
 298     def wrapped_callback(self, cb, obj, st):
 299         with self.workflow_eval_lock:
 300             cb(obj, st)
 301             self.workflow_eval_lock.notifyAll()
 302
 303     def get_wrapped_callback(self, cb):
 304         return partial(self.wrapped_callback, cb)
 305
 306     def on_message(self, event):
 307         if event.get("object_uuid") in self.processes and event["event_type"] == "update":
 308             uuid = event["object_uuid"]
 309             if event["properties"]["new_attributes"]["state"] == "Running":
 310                 with self.workflow_eval_lock:
 311                     j = self.processes[uuid]
 312                     if j.running is False:
 313                         j.running = True
 314                         j.update_pipeline_component(event["properties"]["new_attributes"])
 315                         logger.info("%s %s is Running", self.label(j), uuid)
 316             elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled", "Final"):
 317                 self.process_done(uuid, event["properties"]["new_attributes"])
 318
 319     def label(self, obj):
 320         return "[%s %s]" % (self.work_api[0:-1], obj.name)
 321
 322     def poll_states(self):
 323         """Poll status of containers listed in the processes dict.
 324
 325         Runs in a separate thread.
 326         """
 327
 328         try:
 329             remain_wait = self.poll_interval
 330             while True:
 331                 if remain_wait > 0:
 332                     self.stop_polling.wait(remain_wait)
 333                 if self.stop_polling.is_set():
 334                     break
 335                 with self.workflow_eval_lock:
 336                     keys = list(self.processes)
 337                 if not keys:
 338                     remain_wait = self.poll_interval
 339                     continue
 340
 341                 begin_poll = time.time()
 342                 if self.work_api == "containers":
 343                     table = self.poll_api.container_requests()
 344
 345                 pageSize = self.poll_api._rootDesc.get('maxItemsPerResponse', 1000)
 346
 347                 while keys:
 348                     page = keys[:pageSize]
 349                     try:
 350                         proc_states = table.list(filters=[["uuid", "in", page]]).execute(num_retries=self.num_retries)
 351                     except Exception:
 352                         logger.exception("Error checking states on API server: %s")
 353                         remain_wait = self.poll_interval
 354                         continue
 355
 356                     for p in proc_states["items"]:
 357                         self.on_message({
 358                             "object_uuid": p["uuid"],
 359                             "event_type": "update",
 360                             "properties": {
 361                                 "new_attributes": p
 362                             }
 363                         })
 364                     keys = keys[pageSize:]
 365
 366                 finish_poll = time.time()
 367                 remain_wait = self.poll_interval - (finish_poll - begin_poll)
 368         except:
 369             logger.exception("Fatal error in state polling thread.")
 370             with self.workflow_eval_lock:
 371                 self.processes.clear()
 372                 self.workflow_eval_lock.notifyAll()
 373         finally:
 374             self.stop_polling.set()
 375
 376     def add_intermediate_output(self, uuid):
 377         if uuid:
 378             self.intermediate_output_collections.append(uuid)
 379
 380     def trash_intermediate_output(self):
 381         logger.info("Cleaning up intermediate output collections")
 382         for i in self.intermediate_output_collections:
 383             try:
 384                 self.api.collections().delete(uuid=i).execute(num_retries=self.num_retries)
 385             except Exception:
 386                 logger.warning("Failed to delete intermediate output: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
 387             except (KeyboardInterrupt, SystemExit):
 388                 break
 389
 390     def check_features(self, obj, parentfield=""):
 391         if isinstance(obj, dict):
 392             if obj.get("class") == "DockerRequirement":
 393                 if obj.get("dockerOutputDirectory"):
 394                     if not obj.get("dockerOutputDirectory").startswith('/'):
 395                         raise SourceLine(obj, "dockerOutputDirectory", validate.ValidationException).makeError(
 396                             "Option 'dockerOutputDirectory' must be an absolute path.")
 397             if obj.get("class") == "InplaceUpdateRequirement":
 398                 if obj["inplaceUpdate"] and parentfield == "requirements":
 399                     raise SourceLine(obj, "class", UnsupportedRequirement).makeError("InplaceUpdateRequirement not supported for keep collections.")
 400             for k,v in viewitems(obj):
 401                 self.check_features(v, parentfield=k)
 402         elif isinstance(obj, list):
 403             for i,v in enumerate(obj):
 404                 with SourceLine(obj, i, UnsupportedRequirement, logger.isEnabledFor(logging.DEBUG)):
 405                     self.check_features(v, parentfield=parentfield)
 406
 407     def make_output_collection(self, name, storage_classes, tagsString, outputObj):
 408         outputObj = copy.deepcopy(outputObj)
 409
 410         files = []
 411         def capture(fileobj):
 412             files.append(fileobj)
 413
 414         adjustDirObjs(outputObj, capture)
 415         adjustFileObjs(outputObj, capture)
 416
 417         generatemapper = NoFollowPathMapper(files, "", "", separateDirs=False)
 418
 419         final = arvados.collection.Collection(api_client=self.api,
 420                                               keep_client=self.keep_client,
 421                                               num_retries=self.num_retries)
 422
 423         for k,v in generatemapper.items():
 424             if v.type == "Directory" and v.resolved.startswith("_:"):
 425                     continue
 426             if v.type == "CreateFile" and (k.startswith("_:") or v.resolved.startswith("_:")):
 427                 with final.open(v.target, "wb") as f:
 428                     f.write(v.resolved.encode("utf-8"))
 429                     continue
 430
 431             if not v.resolved.startswith("keep:"):
 432                 raise Exception("Output source is not in keep or a literal")
 433             sp = v.resolved.split("/")
 434             srccollection = sp[0][5:]
 435             try:
 436                 reader = self.collection_cache.get(srccollection)
 437                 srcpath = urllib.parse.unquote("/".join(sp[1:]) if len(sp) > 1 else ".")
 438                 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
 439             except arvados.errors.ArgumentError as e:
 440                 logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e)
 441                 raise
 442             except IOError as e:
 443                 logger.error("While preparing output collection: %s", e)
 444                 raise
 445
 446         def rewrite(fileobj):
 447             fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
 448             for k in ("listing", "contents", "nameext", "nameroot", "dirname"):
 449                 if k in fileobj:
 450                     del fileobj[k]
 451
 452         adjustDirObjs(outputObj, rewrite)
 453         adjustFileObjs(outputObj, rewrite)
 454
 455         with final.open("cwl.output.json", "w") as f:
 456             res = str(json.dumps(outputObj, sort_keys=True, indent=4, separators=(',',': '), ensure_ascii=False))
 457             f.write(res)
 458
 459         final.save_new(name=name, owner_uuid=self.project_uuid, storage_classes=storage_classes, ensure_unique_name=True)
 460
 461         logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
 462                     final.api_response()["name"],
 463                     final.manifest_locator())
 464
 465         final_uuid = final.manifest_locator()
 466         tags = tagsString.split(',')
 467         for tag in tags:
 468              self.api.links().create(body={
 469                 "head_uuid": final_uuid, "link_class": "tag", "name": tag
 470                 }).execute(num_retries=self.num_retries)
 471
 472         def finalcollection(fileobj):
 473             fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"])
 474
 475         adjustDirObjs(outputObj, finalcollection)
 476         adjustFileObjs(outputObj, finalcollection)
 477
 478         return (outputObj, final)
 479
 480     def set_crunch_output(self):
 481         if self.work_api == "containers":
 482             current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 483             if current is None:
 484                 return
 485             try:
 486                 self.api.containers().update(uuid=current['uuid'],
 487                                              body={
 488                                                  'output': self.final_output_collection.portable_data_hash(),
 489                                              }).execute(num_retries=self.num_retries)
 490                 self.api.collections().update(uuid=self.final_output_collection.manifest_locator(),
 491                                               body={
 492                                                   'is_trashed': True
 493                                               }).execute(num_retries=self.num_retries)
 494             except Exception:
 495                 logger.exception("Setting container output")
 496                 raise
 497
 498     def apply_reqs(self, job_order_object, tool):
 499         if "https://w3id.org/cwl/cwl#requirements" in job_order_object:
 500             if tool.metadata.get("http://commonwl.org/cwltool#original_cwlVersion") == 'v1.0':
 501                 raise WorkflowException(
 502                     "`cwl:requirements` in the input object is not part of CWL "
 503                     "v1.0. You can adjust to use `cwltool:overrides` instead; or you "
 504                     "can set the cwlVersion to v1.1 or greater and re-run with "
 505                     "--enable-dev.")
 506             job_reqs = job_order_object["https://w3id.org/cwl/cwl#requirements"]
 507             for req in job_reqs:
 508                 tool.requirements.append(req)
 509
 510     def arv_executor(self, updated_tool, job_order, runtimeContext, logger=None):
 511         self.debug = runtimeContext.debug
 512
 513         workbench1 = self.api.config()["Services"]["Workbench1"]["ExternalURL"]
 514         workbench2 = self.api.config()["Services"]["Workbench2"]["ExternalURL"]
 515         controller = self.api.config()["Services"]["Controller"]["ExternalURL"]
 516         logger.info("Using cluster %s (%s)", self.api.config()["ClusterID"], workbench2 or workbench1 or controller)
 517
 518         updated_tool.visit(self.check_features)
 519
 520         self.project_uuid = runtimeContext.project_uuid
 521         self.pipeline = None
 522         self.fs_access = runtimeContext.make_fs_access(runtimeContext.basedir)
 523         self.secret_store = runtimeContext.secret_store
 524
 525         self.trash_intermediate = runtimeContext.trash_intermediate
 526         if self.trash_intermediate and self.work_api != "containers":
 527             raise Exception("--trash-intermediate is only supported with --api=containers.")
 528
 529         self.intermediate_output_ttl = runtimeContext.intermediate_output_ttl
 530         if self.intermediate_output_ttl and self.work_api != "containers":
 531             raise Exception("--intermediate-output-ttl is only supported with --api=containers.")
 532         if self.intermediate_output_ttl < 0:
 533             raise Exception("Invalid value %d for --intermediate-output-ttl, cannot be less than zero" % self.intermediate_output_ttl)
 534
 535         if runtimeContext.submit_request_uuid and self.work_api != "containers":
 536             raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api))
 537
 538         default_storage_classes = ",".join([k for k,v in self.api.config().get("StorageClasses", {"default": {"Default": True}}).items() if v.get("Default") is True])
 539         if runtimeContext.storage_classes == "default":
 540             runtimeContext.storage_classes = default_storage_classes
 541         if runtimeContext.intermediate_storage_classes == "default":
 542             runtimeContext.intermediate_storage_classes = default_storage_classes
 543
 544         if not runtimeContext.name:
 545             runtimeContext.name = self.name = updated_tool.tool.get("label") or updated_tool.metadata.get("label") or os.path.basename(updated_tool.tool["id"])
 546
 547         # Upload local file references in the job order.
 548         job_order = upload_job_order(self, "%s input" % runtimeContext.name,
 549                                      updated_tool, job_order)
 550
 551         # the last clause means: if it is a command line tool, and we
 552         # are going to wait for the result, and always_submit_runner
 553         # is false, then we don't submit a runner process.
 554
 555         submitting = (runtimeContext.update_workflow or
 556                       runtimeContext.create_workflow or
 557                       (runtimeContext.submit and not
 558                        (updated_tool.tool["class"] == "CommandLineTool" and
 559                         runtimeContext.wait and
 560                         not runtimeContext.always_submit_runner)))
 561
 562         loadingContext = self.loadingContext.copy()
 563         loadingContext.do_validate = False
 564         if submitting:
 565             loadingContext.do_update = False
 566             # Document may have been auto-updated. Reload the original
 567             # document with updating disabled because we want to
 568             # submit the document with its original CWL version, not
 569             # the auto-updated one.
 570             tool = load_tool(updated_tool.tool["id"], loadingContext)
 571         else:
 572             tool = updated_tool
 573
 574         # Upload direct dependencies of workflow steps, get back mapping of files to keep references.
 575         # Also uploads docker images.
 576         merged_map = upload_workflow_deps(self, tool)
 577
 578         # Recreate process object (ArvadosWorkflow or
 579         # ArvadosCommandTool) because tool document may have been
 580         # updated by upload_workflow_deps in ways that modify
 581         # inheritance of hints or requirements.
 582         loadingContext.loader = tool.doc_loader
 583         loadingContext.avsc_names = tool.doc_schema
 584         loadingContext.metadata = tool.metadata
 585         tool = load_tool(tool.tool, loadingContext)
 586
 587         existing_uuid = runtimeContext.update_workflow
 588         if existing_uuid or runtimeContext.create_workflow:
 589             # Create a pipeline template or workflow record and exit.
 590             if self.work_api == "containers":
 591                 uuid = upload_workflow(self, tool, job_order,
 592                                         self.project_uuid,
 593                                         uuid=existing_uuid,
 594                                         submit_runner_ram=runtimeContext.submit_runner_ram,
 595                                         name=runtimeContext.name,
 596                                         merged_map=merged_map,
 597                                         submit_runner_image=runtimeContext.submit_runner_image)
 598                 self.stdout.write(uuid + "\n")
 599                 return (None, "success")
 600
 601         self.apply_reqs(job_order, tool)
 602
 603         self.ignore_docker_for_reuse = runtimeContext.ignore_docker_for_reuse
 604         self.eval_timeout = runtimeContext.eval_timeout
 605
 606         runtimeContext = runtimeContext.copy()
 607         runtimeContext.use_container = True
 608         runtimeContext.tmpdir_prefix = "tmp"
 609         runtimeContext.work_api = self.work_api
 610
 611         if self.work_api == "containers":
 612             if self.ignore_docker_for_reuse:
 613                 raise Exception("--ignore-docker-for-reuse not supported with containers API.")
 614             runtimeContext.outdir = "/var/spool/cwl"
 615             runtimeContext.docker_outdir = "/var/spool/cwl"
 616             runtimeContext.tmpdir = "/tmp"
 617             runtimeContext.docker_tmpdir = "/tmp"
 618
 619         if runtimeContext.priority < 1 or runtimeContext.priority > 1000:
 620             raise Exception("--priority must be in the range 1..1000.")
 621
 622         if self.should_estimate_cache_size:
 623             visited = set()
 624             estimated_size = [0]
 625             def estimate_collection_cache(obj):
 626                 if obj.get("location", "").startswith("keep:"):
 627                     m = pdh_size.match(obj["location"][5:])
 628                     if m and m.group(1) not in visited:
 629                         visited.add(m.group(1))
 630                         estimated_size[0] += int(m.group(2))
 631             visit_class(job_order, ("File", "Directory"), estimate_collection_cache)
 632             runtimeContext.collection_cache_size = max(((estimated_size[0]*192) // (1024*1024))+1, 256)
 633             self.collection_cache.set_cap(runtimeContext.collection_cache_size*1024*1024)
 634
 635         logger.info("Using collection cache size %s MiB", runtimeContext.collection_cache_size)
 636
 637         runnerjob = None
 638         if runtimeContext.submit:
 639             # Submit a runner job to run the workflow for us.
 640             if self.work_api == "containers":
 641                 if submitting:
 642                     tool = RunnerContainer(self, updated_tool,
 643                                            tool, loadingContext, runtimeContext.enable_reuse,
 644                                            self.output_name,
 645                                            self.output_tags,
 646                                            submit_runner_ram=runtimeContext.submit_runner_ram,
 647                                            name=runtimeContext.name,
 648                                            on_error=runtimeContext.on_error,
 649                                            submit_runner_image=runtimeContext.submit_runner_image,
 650                                            intermediate_output_ttl=runtimeContext.intermediate_output_ttl,
 651                                            merged_map=merged_map,
 652                                            priority=runtimeContext.priority,
 653                                            secret_store=self.secret_store,
 654                                            collection_cache_size=runtimeContext.collection_cache_size,
 655                                            collection_cache_is_default=self.should_estimate_cache_size)
 656                 else:
 657                     runtimeContext.runnerjob = tool.tool["id"]
 658
 659         if runtimeContext.cwl_runner_job is not None:
 660             self.uuid = runtimeContext.cwl_runner_job.get('uuid')
 661
 662         jobiter = tool.job(job_order,
 663                            self.output_callback,
 664                            runtimeContext)
 665
 666         if runtimeContext.submit and not runtimeContext.wait:
 667             runnerjob = next(jobiter)
 668             runnerjob.run(runtimeContext)
 669             self.stdout.write(runnerjob.uuid+"\n")
 670             return (None, "success")
 671
 672         current_container = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 673         if current_container:
 674             logger.info("Running inside container %s", current_container.get("uuid"))
 675
 676         self.poll_api = arvados.api('v1', timeout=runtimeContext.http_timeout)
 677         self.polling_thread = threading.Thread(target=self.poll_states)
 678         self.polling_thread.start()
 679
 680         self.task_queue = TaskQueue(self.workflow_eval_lock, self.thread_count)
 681
 682         try:
 683             self.workflow_eval_lock.acquire()
 684
 685             # Holds the lock while this code runs and releases it when
 686             # it is safe to do so in self.workflow_eval_lock.wait(),
 687             # at which point on_message can update job state and
 688             # process output callbacks.
 689
 690             loopperf = Perf(metrics, "jobiter")
 691             loopperf.__enter__()
 692             for runnable in jobiter:
 693                 loopperf.__exit__()
 694
 695                 if self.stop_polling.is_set():
 696                     break
 697
 698                 if self.task_queue.error is not None:
 699                     raise self.task_queue.error
 700
 701                 if runnable:
 702                     with Perf(metrics, "run"):
 703                         self.start_run(runnable, runtimeContext)
 704                 else:
 705                     if (self.task_queue.in_flight + len(self.processes)) > 0:
 706                         self.workflow_eval_lock.wait(3)
 707                     else:
 708                         logger.error("Workflow is deadlocked, no runnable processes and not waiting on any pending processes.")
 709                         break
 710
 711                 if self.stop_polling.is_set():
 712                     break
 713
 714                 loopperf.__enter__()
 715             loopperf.__exit__()
 716
 717             while (self.task_queue.in_flight + len(self.processes)) > 0:
 718                 if self.task_queue.error is not None:
 719                     raise self.task_queue.error
 720                 self.workflow_eval_lock.wait(3)
 721
 722         except UnsupportedRequirement:
 723             raise
 724         except:
 725             if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit:
 726                 logger.error("Interrupted, workflow will be cancelled")
 727             elif isinstance(sys.exc_info()[1], WorkflowException):
 728                 logger.error("Workflow execution failed:\n%s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
 729             else:
 730                 logger.exception("Workflow execution failed")
 731
 732             if self.pipeline:
 733                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 734                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
 735
 736             if self.work_api == "containers" and not current_container:
 737                 # Not running in a crunch container, so cancel any outstanding processes.
 738                 for p in self.processes:
 739                     try:
 740                         self.api.container_requests().update(uuid=p,
 741                                                              body={"priority": "0"}
 742                         ).execute(num_retries=self.num_retries)
 743                     except Exception:
 744                         pass
 745         finally:
 746             self.workflow_eval_lock.release()
 747             self.task_queue.drain()
 748             self.stop_polling.set()
 749             self.polling_thread.join()
 750             self.task_queue.join()
 751
 752         if self.final_status == "UnsupportedRequirement":
 753             raise UnsupportedRequirement("Check log for details.")
 754
 755         if self.final_output is None:
 756             raise WorkflowException("Workflow did not return a result.")
 757
 758         if runtimeContext.submit and isinstance(tool, Runner):
 759             logger.info("Final output collection %s", tool.final_output)
 760             if workbench2 or workbench1:
 761                 logger.info("Output at %scollections/%s", workbench2 or workbench1, tool.final_output)
 762         else:
 763             if self.output_name is None:
 764                 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
 765             if self.output_tags is None:
 766                 self.output_tags = ""
 767
 768             storage_classes = ""
 769             storage_class_req, _ = tool.get_requirement("http://arvados.org/cwl#OutputStorageClass")
 770             if storage_class_req and storage_class_req.get("finalStorageClass"):
 771                 storage_classes = aslist(storage_class_req["finalStorageClass"])
 772             else:
 773                 storage_classes = runtimeContext.storage_classes.strip().split(",")
 774
 775             self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, storage_classes, self.output_tags, self.final_output)
 776             self.set_crunch_output()
 777
 778         if runtimeContext.compute_checksum:
 779             adjustDirObjs(self.final_output, partial(get_listing, self.fs_access))
 780             adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
 781
 782         if self.trash_intermediate and self.final_status == "success":
 783             self.trash_intermediate_output()
 784
 785         return (self.final_output, self.final_status)