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, cleanup_name_for_collection
  35 from .runner import Runner, upload_docker, upload_job_order, upload_workflow_deps, make_builder
  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.toplevel_runtimeContext = ArvRuntimeContext(vars(arvargs))
 201         self.toplevel_runtimeContext.make_fs_access = partial(CollectionFsAccess,
 202                                                      collection_cache=self.collection_cache)
 203
 204         validate_cluster_target(self, self.toplevel_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, output_properties, 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
 460         final.save_new(name=name, owner_uuid=self.project_uuid, storage_classes=storage_classes,
 461                        ensure_unique_name=True, properties=output_properties)
 462
 463         logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
 464                     final.api_response()["name"],
 465                     final.manifest_locator())
 466
 467         final_uuid = final.manifest_locator()
 468         tags = tagsString.split(',')
 469         for tag in tags:
 470              self.api.links().create(body={
 471                 "head_uuid": final_uuid, "link_class": "tag", "name": tag
 472                 }).execute(num_retries=self.num_retries)
 473
 474         def finalcollection(fileobj):
 475             fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"])
 476
 477         adjustDirObjs(outputObj, finalcollection)
 478         adjustFileObjs(outputObj, finalcollection)
 479
 480         return (outputObj, final)
 481
 482     def set_crunch_output(self):
 483         if self.work_api == "containers":
 484             current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 485             if current is None:
 486                 return
 487             try:
 488                 self.api.containers().update(uuid=current['uuid'],
 489                                              body={
 490                                                  'output': self.final_output_collection.portable_data_hash(),
 491                                                  'output_properties': self.final_output_collection.get_properties(),
 492                                              }).execute(num_retries=self.num_retries)
 493                 self.api.collections().update(uuid=self.final_output_collection.manifest_locator(),
 494                                               body={
 495                                                   'is_trashed': True
 496                                               }).execute(num_retries=self.num_retries)
 497             except Exception:
 498                 logger.exception("Setting container output")
 499                 raise
 500
 501     def apply_reqs(self, job_order_object, tool):
 502         if "https://w3id.org/cwl/cwl#requirements" in job_order_object:
 503             if tool.metadata.get("http://commonwl.org/cwltool#original_cwlVersion") == 'v1.0':
 504                 raise WorkflowException(
 505                     "`cwl:requirements` in the input object is not part of CWL "
 506                     "v1.0. You can adjust to use `cwltool:overrides` instead; or you "
 507                     "can set the cwlVersion to v1.1 or greater and re-run with "
 508                     "--enable-dev.")
 509             job_reqs = job_order_object["https://w3id.org/cwl/cwl#requirements"]
 510             for req in job_reqs:
 511                 tool.requirements.append(req)
 512
 513     def arv_executor(self, updated_tool, job_order, runtimeContext, logger=None):
 514         self.debug = runtimeContext.debug
 515
 516         workbench1 = self.api.config()["Services"]["Workbench1"]["ExternalURL"]
 517         workbench2 = self.api.config()["Services"]["Workbench2"]["ExternalURL"]
 518         controller = self.api.config()["Services"]["Controller"]["ExternalURL"]
 519         logger.info("Using cluster %s (%s)", self.api.config()["ClusterID"], workbench2 or workbench1 or controller)
 520
 521         updated_tool.visit(self.check_features)
 522
 523         self.pipeline = None
 524         self.fs_access = runtimeContext.make_fs_access(runtimeContext.basedir)
 525         self.secret_store = runtimeContext.secret_store
 526
 527         self.trash_intermediate = runtimeContext.trash_intermediate
 528         if self.trash_intermediate and self.work_api != "containers":
 529             raise Exception("--trash-intermediate is only supported with --api=containers.")
 530
 531         self.intermediate_output_ttl = runtimeContext.intermediate_output_ttl
 532         if self.intermediate_output_ttl and self.work_api != "containers":
 533             raise Exception("--intermediate-output-ttl is only supported with --api=containers.")
 534         if self.intermediate_output_ttl < 0:
 535             raise Exception("Invalid value %d for --intermediate-output-ttl, cannot be less than zero" % self.intermediate_output_ttl)
 536
 537         if runtimeContext.submit_request_uuid and self.work_api != "containers":
 538             raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api))
 539
 540         runtimeContext = runtimeContext.copy()
 541
 542         default_storage_classes = ",".join([k for k,v in self.api.config().get("StorageClasses", {"default": {"Default": True}}).items() if v.get("Default") is True])
 543         if runtimeContext.storage_classes == "default":
 544             runtimeContext.storage_classes = default_storage_classes
 545         if runtimeContext.intermediate_storage_classes == "default":
 546             runtimeContext.intermediate_storage_classes = default_storage_classes
 547
 548         if not runtimeContext.name:
 549             runtimeContext.name = self.name = updated_tool.tool.get("label") or updated_tool.metadata.get("label") or os.path.basename(updated_tool.tool["id"])
 550
 551         if runtimeContext.copy_deps is None and (runtimeContext.create_workflow or runtimeContext.update_workflow):
 552             # When creating or updating workflow record, by default
 553             # always copy dependencies and ensure Docker images are up
 554             # to date.
 555             runtimeContext.copy_deps = True
 556             runtimeContext.match_local_docker = True
 557
 558         if runtimeContext.update_workflow and self.project_uuid is None:
 559             # If we are updating a workflow, make sure anything that
 560             # gets uploaded goes into the same parent project, unless
 561             # an alternate --project-uuid was provided.
 562             existing_wf = self.api.workflows().get(uuid=runtimeContext.update_workflow).execute()
 563             runtimeContext.project_uuid = existing_wf["owner_uuid"]
 564
 565         self.project_uuid = runtimeContext.project_uuid
 566
 567         # Upload local file references in the job order.
 568         job_order = upload_job_order(self, "%s input" % runtimeContext.name,
 569                                      updated_tool, job_order, runtimeContext)
 570
 571         # the last clause means: if it is a command line tool, and we
 572         # are going to wait for the result, and always_submit_runner
 573         # is false, then we don't submit a runner process.
 574
 575         submitting = (runtimeContext.update_workflow or
 576                       runtimeContext.create_workflow or
 577                       (runtimeContext.submit and not
 578                        (updated_tool.tool["class"] == "CommandLineTool" and
 579                         runtimeContext.wait and
 580                         not runtimeContext.always_submit_runner)))
 581
 582         loadingContext = self.loadingContext.copy()
 583         loadingContext.do_validate = False
 584         loadingContext.disable_js_validation = True
 585         if submitting:
 586             loadingContext.do_update = False
 587             # Document may have been auto-updated. Reload the original
 588             # document with updating disabled because we want to
 589             # submit the document with its original CWL version, not
 590             # the auto-updated one.
 591             tool = load_tool(updated_tool.tool["id"], loadingContext)
 592         else:
 593             tool = updated_tool
 594
 595         # Upload direct dependencies of workflow steps, get back mapping of files to keep references.
 596         # Also uploads docker images.
 597         merged_map = upload_workflow_deps(self, tool, runtimeContext)
 598
 599         # Recreate process object (ArvadosWorkflow or
 600         # ArvadosCommandTool) because tool document may have been
 601         # updated by upload_workflow_deps in ways that modify
 602         # inheritance of hints or requirements.
 603         loadingContext.loader = tool.doc_loader
 604         loadingContext.avsc_names = tool.doc_schema
 605         loadingContext.metadata = tool.metadata
 606         tool = load_tool(tool.tool, loadingContext)
 607
 608         if runtimeContext.update_workflow or runtimeContext.create_workflow:
 609             # Create a pipeline template or workflow record and exit.
 610             if self.work_api == "containers":
 611                 uuid = upload_workflow(self, tool, job_order,
 612                                        runtimeContext.project_uuid,
 613                                        runtimeContext,
 614                                        uuid=runtimeContext.update_workflow,
 615                                        submit_runner_ram=runtimeContext.submit_runner_ram,
 616                                        name=runtimeContext.name,
 617                                        merged_map=merged_map,
 618                                        submit_runner_image=runtimeContext.submit_runner_image)
 619                 self.stdout.write(uuid + "\n")
 620                 return (None, "success")
 621
 622         self.apply_reqs(job_order, tool)
 623
 624         self.ignore_docker_for_reuse = runtimeContext.ignore_docker_for_reuse
 625         self.eval_timeout = runtimeContext.eval_timeout
 626
 627         runtimeContext.use_container = True
 628         runtimeContext.tmpdir_prefix = "tmp"
 629         runtimeContext.work_api = self.work_api
 630
 631         if not self.output_name:
 632              self.output_name = "Output from workflow %s" % runtimeContext.name
 633
 634         self.output_name  = cleanup_name_for_collection(self.output_name)
 635
 636         if self.work_api == "containers":
 637             if self.ignore_docker_for_reuse:
 638                 raise Exception("--ignore-docker-for-reuse not supported with containers API.")
 639             runtimeContext.outdir = "/var/spool/cwl"
 640             runtimeContext.docker_outdir = "/var/spool/cwl"
 641             runtimeContext.tmpdir = "/tmp"
 642             runtimeContext.docker_tmpdir = "/tmp"
 643
 644         if runtimeContext.priority < 1 or runtimeContext.priority > 1000:
 645             raise Exception("--priority must be in the range 1..1000.")
 646
 647         if self.should_estimate_cache_size:
 648             visited = set()
 649             estimated_size = [0]
 650             def estimate_collection_cache(obj):
 651                 if obj.get("location", "").startswith("keep:"):
 652                     m = pdh_size.match(obj["location"][5:])
 653                     if m and m.group(1) not in visited:
 654                         visited.add(m.group(1))
 655                         estimated_size[0] += int(m.group(2))
 656             visit_class(job_order, ("File", "Directory"), estimate_collection_cache)
 657             runtimeContext.collection_cache_size = max(((estimated_size[0]*192) // (1024*1024))+1, 256)
 658             self.collection_cache.set_cap(runtimeContext.collection_cache_size*1024*1024)
 659
 660         logger.info("Using collection cache size %s MiB", runtimeContext.collection_cache_size)
 661
 662         runnerjob = None
 663         if runtimeContext.submit:
 664             # Submit a runner job to run the workflow for us.
 665             if self.work_api == "containers":
 666                 if submitting:
 667                     tool = RunnerContainer(self, updated_tool,
 668                                            tool, loadingContext, runtimeContext.enable_reuse,
 669                                            self.output_name,
 670                                            self.output_tags,
 671                                            submit_runner_ram=runtimeContext.submit_runner_ram,
 672                                            name=runtimeContext.name,
 673                                            on_error=runtimeContext.on_error,
 674                                            submit_runner_image=runtimeContext.submit_runner_image,
 675                                            intermediate_output_ttl=runtimeContext.intermediate_output_ttl,
 676                                            merged_map=merged_map,
 677                                            priority=runtimeContext.priority,
 678                                            secret_store=self.secret_store,
 679                                            collection_cache_size=runtimeContext.collection_cache_size,
 680                                            collection_cache_is_default=self.should_estimate_cache_size)
 681                 else:
 682                     runtimeContext.runnerjob = tool.tool["id"]
 683
 684         if runtimeContext.cwl_runner_job is not None:
 685             self.uuid = runtimeContext.cwl_runner_job.get('uuid')
 686
 687         jobiter = tool.job(job_order,
 688                            self.output_callback,
 689                            runtimeContext)
 690
 691         if runtimeContext.submit and not runtimeContext.wait:
 692             runnerjob = next(jobiter)
 693             runnerjob.run(runtimeContext)
 694             self.stdout.write(runnerjob.uuid+"\n")
 695             return (None, "success")
 696
 697         current_container = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 698         if current_container:
 699             logger.info("Running inside container %s", current_container.get("uuid"))
 700
 701         self.poll_api = arvados.api('v1', timeout=runtimeContext.http_timeout)
 702         self.polling_thread = threading.Thread(target=self.poll_states)
 703         self.polling_thread.start()
 704
 705         self.task_queue = TaskQueue(self.workflow_eval_lock, self.thread_count)
 706
 707         try:
 708             self.workflow_eval_lock.acquire()
 709
 710             # Holds the lock while this code runs and releases it when
 711             # it is safe to do so in self.workflow_eval_lock.wait(),
 712             # at which point on_message can update job state and
 713             # process output callbacks.
 714
 715             loopperf = Perf(metrics, "jobiter")
 716             loopperf.__enter__()
 717             for runnable in jobiter:
 718                 loopperf.__exit__()
 719
 720                 if self.stop_polling.is_set():
 721                     break
 722
 723                 if self.task_queue.error is not None:
 724                     raise self.task_queue.error
 725
 726                 if runnable:
 727                     with Perf(metrics, "run"):
 728                         self.start_run(runnable, runtimeContext)
 729                 else:
 730                     if (self.task_queue.in_flight + len(self.processes)) > 0:
 731                         self.workflow_eval_lock.wait(3)
 732                     else:
 733                         logger.error("Workflow is deadlocked, no runnable processes and not waiting on any pending processes.")
 734                         break
 735
 736                 if self.stop_polling.is_set():
 737                     break
 738
 739                 loopperf.__enter__()
 740             loopperf.__exit__()
 741
 742             while (self.task_queue.in_flight + len(self.processes)) > 0:
 743                 if self.task_queue.error is not None:
 744                     raise self.task_queue.error
 745                 self.workflow_eval_lock.wait(3)
 746
 747         except UnsupportedRequirement:
 748             raise
 749         except:
 750             if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit:
 751                 logger.error("Interrupted, workflow will be cancelled")
 752             elif isinstance(sys.exc_info()[1], WorkflowException):
 753                 logger.error("Workflow execution failed:\n%s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
 754             else:
 755                 logger.exception("Workflow execution failed")
 756
 757             if self.pipeline:
 758                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 759                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
 760
 761             if self.work_api == "containers" and not current_container:
 762                 # Not running in a crunch container, so cancel any outstanding processes.
 763                 for p in self.processes:
 764                     try:
 765                         self.api.container_requests().update(uuid=p,
 766                                                              body={"priority": "0"}
 767                         ).execute(num_retries=self.num_retries)
 768                     except Exception:
 769                         pass
 770         finally:
 771             self.workflow_eval_lock.release()
 772             self.task_queue.drain()
 773             self.stop_polling.set()
 774             self.polling_thread.join()
 775             self.task_queue.join()
 776
 777         if self.final_status == "UnsupportedRequirement":
 778             raise UnsupportedRequirement("Check log for details.")
 779
 780         if self.final_output is None:
 781             raise WorkflowException("Workflow did not return a result.")
 782
 783         if runtimeContext.submit and isinstance(tool, Runner):
 784             logger.info("Final output collection %s", tool.final_output)
 785             if workbench2 or workbench1:
 786                 logger.info("Output at %scollections/%s", workbench2 or workbench1, tool.final_output)
 787         else:
 788             if self.output_tags is None:
 789                 self.output_tags = ""
 790
 791             storage_classes = ""
 792             storage_class_req, _ = tool.get_requirement("http://arvados.org/cwl#OutputStorageClass")
 793             if storage_class_req and storage_class_req.get("finalStorageClass"):
 794                 storage_classes = aslist(storage_class_req["finalStorageClass"])
 795             else:
 796                 storage_classes = runtimeContext.storage_classes.strip().split(",")
 797
 798             output_properties = {}
 799             output_properties_req, _ = tool.get_requirement("http://arvados.org/cwl#OutputCollectionProperties")
 800             if output_properties_req:
 801                 builder = make_builder(job_order, tool.hints, tool.requirements, runtimeContext, tool.metadata)
 802                 for pr in output_properties_req["outputProperties"]:
 803                     output_properties[pr["propertyName"]] = builder.do_eval(pr["propertyValue"])
 804
 805             self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, storage_classes,
 806                                                                                           self.output_tags, output_properties,
 807                                                                                           self.final_output)
 808             self.set_crunch_output()
 809
 810         if runtimeContext.compute_checksum:
 811             adjustDirObjs(self.final_output, partial(get_listing, self.fs_access))
 812             adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
 813
 814         if self.trash_intermediate and self.final_status == "success":
 815             self.trash_intermediate_output()
 816
 817         return (self.final_output, self.final_status)