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 subprocess
  21 import time
  22 import urllib
  23
  24 from cwltool.errors import WorkflowException
  25 import cwltool.workflow
  26 from schema_salad.sourceline import SourceLine, cmap
  27 import schema_salad.validate as validate
  28 from schema_salad.ref_resolver import file_uri, uri_file_path
  29
  30 import arvados
  31 import arvados.config
  32 from arvados.keep import KeepClient
  33 from arvados.errors import ApiError
  34
  35 import arvados_cwl.util
  36 from .arvcontainer import RunnerContainer, cleanup_name_for_collection
  37 from .runner import Runner, upload_docker, upload_job_order, upload_workflow_deps, make_builder, update_from_merged_map
  38 from .arvtool import ArvadosCommandTool, validate_cluster_target, ArvadosExpressionTool
  39 from .arvworkflow import ArvadosWorkflow, upload_workflow, make_workflow_record
  40 from .fsaccess import CollectionFsAccess, CollectionFetcher, collectionResolver, CollectionCache, pdh_size
  41 from .perf import Perf
  42 from .pathmapper import NoFollowPathMapper
  43 from cwltool.task_queue import TaskQueue
  44 from .context import ArvLoadingContext, ArvRuntimeContext
  45 from ._version import __version__
  46
  47 from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema
  48 from cwltool.utils import adjustFileObjs, adjustDirObjs, get_listing, visit_class, aslist
  49 from cwltool.command_line_tool import compute_checksums
  50 from cwltool.load_tool import load_tool
  51
  52 logger = logging.getLogger('arvados.cwl-runner')
  53 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  54
  55 DEFAULT_PRIORITY = 500
  56
  57 class RuntimeStatusLoggingHandler(logging.Handler):
  58     """
  59     Intercepts logging calls and report them as runtime statuses on runner
  60     containers.
  61     """
  62     def __init__(self, runtime_status_update_func):
  63         super(RuntimeStatusLoggingHandler, self).__init__()
  64         self.runtime_status_update = runtime_status_update_func
  65         self.updatingRuntimeStatus = False
  66
  67     def emit(self, record):
  68         kind = None
  69         if record.levelno >= logging.ERROR:
  70             kind = 'error'
  71         elif record.levelno >= logging.WARNING:
  72             kind = 'warning'
  73         if kind == 'warning' and record.name == "salad":
  74             # Don't send validation warnings to runtime status,
  75             # they're noisy and unhelpful.
  76             return
  77         if kind is not None and self.updatingRuntimeStatus is not True:
  78             self.updatingRuntimeStatus = True
  79             try:
  80                 log_msg = record.getMessage()
  81                 if '\n' in log_msg:
  82                     # If the logged message is multi-line, use its first line as status
  83                     # and the rest as detail.
  84                     status, detail = log_msg.split('\n', 1)
  85                     self.runtime_status_update(
  86                         kind,
  87                         "%s: %s" % (record.name, status),
  88                         detail
  89                     )
  90                 else:
  91                     self.runtime_status_update(
  92                         kind,
  93                         "%s: %s" % (record.name, record.getMessage())
  94                     )
  95             finally:
  96                 self.updatingRuntimeStatus = False
  97
  98
  99 class ArvCwlExecutor(object):
 100     """Execute a CWL tool or workflow, submit work (using containers API),
 101     wait for them to complete, and report output.
 102
 103     """
 104
 105     def __init__(self, api_client,
 106                  arvargs=None,
 107                  keep_client=None,
 108                  num_retries=4,
 109                  thread_count=4,
 110                  stdout=sys.stdout):
 111
 112         if arvargs is None:
 113             arvargs = argparse.Namespace()
 114             arvargs.work_api = None
 115             arvargs.output_name = None
 116             arvargs.output_tags = None
 117             arvargs.thread_count = 1
 118             arvargs.collection_cache_size = None
 119             arvargs.git_info = True
 120             arvargs.submit = False
 121             arvargs.defer_downloads = False
 122
 123         self.api = api_client
 124         self.processes = {}
 125         self.workflow_eval_lock = threading.Condition(threading.RLock())
 126         self.final_output = None
 127         self.final_status = None
 128         self.num_retries = num_retries
 129         self.uuid = None
 130         self.stop_polling = threading.Event()
 131         self.poll_api = None
 132         self.pipeline = None
 133         self.final_output_collection = None
 134         self.output_name = arvargs.output_name
 135         self.output_tags = arvargs.output_tags
 136         self.project_uuid = None
 137         self.intermediate_output_ttl = 0
 138         self.intermediate_output_collections = []
 139         self.trash_intermediate = False
 140         self.thread_count = arvargs.thread_count
 141         self.poll_interval = 12
 142         self.loadingContext = None
 143         self.should_estimate_cache_size = True
 144         self.fs_access = None
 145         self.secret_store = None
 146         self.stdout = stdout
 147         self.fast_submit = False
 148         self.git_info = arvargs.git_info
 149
 150         if keep_client is not None:
 151             self.keep_client = keep_client
 152         else:
 153             self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
 154
 155         if arvargs.collection_cache_size:
 156             collection_cache_size = arvargs.collection_cache_size*1024*1024
 157             self.should_estimate_cache_size = False
 158         else:
 159             collection_cache_size = 256*1024*1024
 160
 161         self.collection_cache = CollectionCache(self.api, self.keep_client, self.num_retries,
 162                                                 cap=collection_cache_size)
 163
 164         self.fetcher_constructor = partial(CollectionFetcher,
 165                                            api_client=self.api,
 166                                            fs_access=CollectionFsAccess("", collection_cache=self.collection_cache),
 167                                            num_retries=self.num_retries)
 168
 169         self.work_api = None
 170         expected_api = ["containers"]
 171         for api in expected_api:
 172             try:
 173                 methods = self.api._rootDesc.get('resources')[api]['methods']
 174                 if ('httpMethod' in methods['create'] and
 175                     (arvargs.work_api == api or arvargs.work_api is None)):
 176                     self.work_api = api
 177                     break
 178             except KeyError:
 179                 pass
 180
 181         if not self.work_api:
 182             if arvargs.work_api is None:
 183                 raise Exception("No supported APIs")
 184             else:
 185                 raise Exception("Unsupported API '%s', expected one of %s" % (arvargs.work_api, expected_api))
 186
 187         if self.work_api == "jobs":
 188             logger.error("""
 189 *******************************
 190 The 'jobs' API is no longer supported.
 191 *******************************""")
 192             exit(1)
 193
 194         self.loadingContext = ArvLoadingContext(vars(arvargs))
 195         self.loadingContext.fetcher_constructor = self.fetcher_constructor
 196         self.loadingContext.resolver = partial(collectionResolver, self.api, num_retries=self.num_retries)
 197         self.loadingContext.construct_tool_object = self.arv_make_tool
 198
 199         # Add a custom logging handler to the root logger for runtime status reporting
 200         # if running inside a container
 201         if arvados_cwl.util.get_current_container(self.api, self.num_retries, logger):
 202             root_logger = logging.getLogger('')
 203
 204             # Remove existing RuntimeStatusLoggingHandlers if they exist
 205             handlers = [h for h in root_logger.handlers if not isinstance(h, RuntimeStatusLoggingHandler)]
 206             root_logger.handlers = handlers
 207
 208             handler = RuntimeStatusLoggingHandler(self.runtime_status_update)
 209             root_logger.addHandler(handler)
 210
 211         self.toplevel_runtimeContext = ArvRuntimeContext(vars(arvargs))
 212         self.toplevel_runtimeContext.make_fs_access = partial(CollectionFsAccess,
 213                                                      collection_cache=self.collection_cache)
 214
 215         self.defer_downloads = arvargs.submit and arvargs.defer_downloads
 216
 217         validate_cluster_target(self, self.toplevel_runtimeContext)
 218
 219
 220     def arv_make_tool(self, toolpath_object, loadingContext):
 221         if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
 222             return ArvadosCommandTool(self, toolpath_object, loadingContext)
 223         elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
 224             return ArvadosWorkflow(self, toolpath_object, loadingContext)
 225         elif "class" in toolpath_object and toolpath_object["class"] == "ExpressionTool":
 226             return ArvadosExpressionTool(self, toolpath_object, loadingContext)
 227         else:
 228             raise Exception("Unknown tool %s" % toolpath_object.get("class"))
 229
 230     def output_callback(self, out, processStatus):
 231         with self.workflow_eval_lock:
 232             if processStatus == "success":
 233                 logger.info("Overall process status is %s", processStatus)
 234                 state = "Complete"
 235             else:
 236                 logger.error("Overall process status is %s", processStatus)
 237                 state = "Failed"
 238             if self.pipeline:
 239                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 240                                                         body={"state": state}).execute(num_retries=self.num_retries)
 241             self.final_status = processStatus
 242             self.final_output = out
 243             self.workflow_eval_lock.notifyAll()
 244
 245
 246     def start_run(self, runnable, runtimeContext):
 247         self.task_queue.add(partial(runnable.run, runtimeContext),
 248                             self.workflow_eval_lock, self.stop_polling)
 249
 250     def process_submitted(self, container):
 251         with self.workflow_eval_lock:
 252             self.processes[container.uuid] = container
 253
 254     def process_done(self, uuid, record):
 255         with self.workflow_eval_lock:
 256             j = self.processes[uuid]
 257             logger.info("%s %s is %s", self.label(j), uuid, record["state"])
 258             self.task_queue.add(partial(j.done, record),
 259                                 self.workflow_eval_lock, self.stop_polling)
 260             del self.processes[uuid]
 261
 262     def runtime_status_update(self, kind, message, detail=None):
 263         """
 264         Updates the runtime_status field on the runner container.
 265         Called when there's a need to report errors, warnings or just
 266         activity statuses, for example in the RuntimeStatusLoggingHandler.
 267         """
 268
 269         if kind not in ('error', 'warning', 'activity'):
 270             # Ignore any other status kind
 271             return
 272
 273         with self.workflow_eval_lock:
 274             current = None
 275             try:
 276                 current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 277             except Exception as e:
 278                 logger.info("Couldn't get current container: %s", e)
 279             if current is None:
 280                 return
 281             runtime_status = current.get('runtime_status', {})
 282
 283             original_updatemessage = updatemessage = runtime_status.get(kind, "")
 284             if kind == "activity" or not updatemessage:
 285                 updatemessage = message
 286
 287             # Subsequent messages tacked on in detail
 288             original_updatedetail = updatedetail = runtime_status.get(kind+'Detail', "")
 289             maxlines = 40
 290             if updatedetail.count("\n") < maxlines:
 291                 if updatedetail:
 292                     updatedetail += "\n"
 293                 updatedetail += message + "\n"
 294
 295                 if detail:
 296                     updatedetail += detail + "\n"
 297
 298                 if updatedetail.count("\n") >= maxlines:
 299                     updatedetail += "\nSome messages may have been omitted.  Check the full log."
 300
 301             if updatemessage == original_updatemessage and updatedetail == original_updatedetail:
 302                 # don't waste time doing an update if nothing changed
 303                 # (usually because we exceeded the max lines)
 304                 return
 305
 306             runtime_status.update({
 307                 kind: updatemessage,
 308                 kind+'Detail': updatedetail,
 309             })
 310
 311             try:
 312                 self.api.containers().update(uuid=current['uuid'],
 313                                             body={
 314                                                 'runtime_status': runtime_status,
 315                                             }).execute(num_retries=self.num_retries)
 316             except Exception as e:
 317                 logger.info("Couldn't update runtime_status: %s", e)
 318
 319     def wrapped_callback(self, cb, obj, st):
 320         with self.workflow_eval_lock:
 321             cb(obj, st)
 322             self.workflow_eval_lock.notifyAll()
 323
 324     def get_wrapped_callback(self, cb):
 325         return partial(self.wrapped_callback, cb)
 326
 327     def on_message(self, event):
 328         if event.get("object_uuid") in self.processes and event["event_type"] == "update":
 329             uuid = event["object_uuid"]
 330             if event["properties"]["new_attributes"]["state"] == "Running":
 331                 with self.workflow_eval_lock:
 332                     j = self.processes[uuid]
 333                     if j.running is False:
 334                         j.running = True
 335                         j.update_pipeline_component(event["properties"]["new_attributes"])
 336                         logger.info("%s %s is Running", self.label(j), uuid)
 337             elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled", "Final"):
 338                 self.process_done(uuid, event["properties"]["new_attributes"])
 339
 340     def label(self, obj):
 341         return "[%s %s]" % (self.work_api[0:-1], obj.name)
 342
 343     def poll_states(self):
 344         """Poll status of containers listed in the processes dict.
 345
 346         Runs in a separate thread.
 347         """
 348
 349         try:
 350             remain_wait = self.poll_interval
 351             while True:
 352                 if remain_wait > 0:
 353                     self.stop_polling.wait(remain_wait)
 354                 if self.stop_polling.is_set():
 355                     break
 356                 with self.workflow_eval_lock:
 357                     keys = list(self.processes)
 358                 if not keys:
 359                     remain_wait = self.poll_interval
 360                     continue
 361
 362                 begin_poll = time.time()
 363                 if self.work_api == "containers":
 364                     table = self.poll_api.container_requests()
 365
 366                 pageSize = self.poll_api._rootDesc.get('maxItemsPerResponse', 1000)
 367
 368                 while keys:
 369                     page = keys[:pageSize]
 370                     try:
 371                         proc_states = table.list(filters=[["uuid", "in", page]], select=["uuid", "container_uuid", "state", "log_uuid",
 372                                                                                          "output_uuid", "modified_at", "properties"]).execute(num_retries=self.num_retries)
 373                     except Exception as e:
 374                         logger.warning("Temporary error checking states on API server: %s", e)
 375                         remain_wait = self.poll_interval
 376                         continue
 377
 378                     for p in proc_states["items"]:
 379                         self.on_message({
 380                             "object_uuid": p["uuid"],
 381                             "event_type": "update",
 382                             "properties": {
 383                                 "new_attributes": p
 384                             }
 385                         })
 386                     keys = keys[pageSize:]
 387
 388                 finish_poll = time.time()
 389                 remain_wait = self.poll_interval - (finish_poll - begin_poll)
 390         except:
 391             logger.exception("Fatal error in state polling thread.")
 392             with self.workflow_eval_lock:
 393                 self.processes.clear()
 394                 self.workflow_eval_lock.notifyAll()
 395         finally:
 396             self.stop_polling.set()
 397
 398     def add_intermediate_output(self, uuid):
 399         if uuid:
 400             self.intermediate_output_collections.append(uuid)
 401
 402     def trash_intermediate_output(self):
 403         logger.info("Cleaning up intermediate output collections")
 404         for i in self.intermediate_output_collections:
 405             try:
 406                 self.api.collections().delete(uuid=i).execute(num_retries=self.num_retries)
 407             except Exception:
 408                 logger.warning("Failed to delete intermediate output: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
 409             except (KeyboardInterrupt, SystemExit):
 410                 break
 411
 412     def check_features(self, obj, parentfield=""):
 413         if isinstance(obj, dict):
 414             if obj.get("class") == "DockerRequirement":
 415                 if obj.get("dockerOutputDirectory"):
 416                     if not obj.get("dockerOutputDirectory").startswith('/'):
 417                         raise SourceLine(obj, "dockerOutputDirectory", validate.ValidationException).makeError(
 418                             "Option 'dockerOutputDirectory' must be an absolute path.")
 419             if obj.get("class") == "InplaceUpdateRequirement":
 420                 if obj["inplaceUpdate"] and parentfield == "requirements":
 421                     raise SourceLine(obj, "class", UnsupportedRequirement).makeError("InplaceUpdateRequirement not supported for keep collections.")
 422             for k,v in viewitems(obj):
 423                 self.check_features(v, parentfield=k)
 424         elif isinstance(obj, list):
 425             for i,v in enumerate(obj):
 426                 with SourceLine(obj, i, UnsupportedRequirement, logger.isEnabledFor(logging.DEBUG)):
 427                     self.check_features(v, parentfield=parentfield)
 428
 429     def make_output_collection(self, name, storage_classes, tagsString, output_properties, outputObj):
 430         outputObj = copy.deepcopy(outputObj)
 431
 432         files = []
 433         def capture(fileobj):
 434             files.append(fileobj)
 435
 436         adjustDirObjs(outputObj, capture)
 437         adjustFileObjs(outputObj, capture)
 438
 439         generatemapper = NoFollowPathMapper(files, "", "", separateDirs=False)
 440
 441         final = arvados.collection.Collection(api_client=self.api,
 442                                               keep_client=self.keep_client,
 443                                               num_retries=self.num_retries)
 444
 445         for k,v in generatemapper.items():
 446             if v.type == "Directory" and v.resolved.startswith("_:"):
 447                     continue
 448             if v.type == "CreateFile" and (k.startswith("_:") or v.resolved.startswith("_:")):
 449                 with final.open(v.target, "wb") as f:
 450                     f.write(v.resolved.encode("utf-8"))
 451                     continue
 452
 453             if not v.resolved.startswith("keep:"):
 454                 raise Exception("Output source is not in keep or a literal")
 455             sp = v.resolved.split("/")
 456             srccollection = sp[0][5:]
 457             try:
 458                 reader = self.collection_cache.get(srccollection)
 459                 srcpath = urllib.parse.unquote("/".join(sp[1:]) if len(sp) > 1 else ".")
 460                 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
 461             except arvados.errors.ArgumentError as e:
 462                 logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e)
 463                 raise
 464             except IOError as e:
 465                 logger.error("While preparing output collection: %s", e)
 466                 raise
 467
 468         def rewrite(fileobj):
 469             fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
 470             for k in ("listing", "contents", "nameext", "nameroot", "dirname"):
 471                 if k in fileobj:
 472                     del fileobj[k]
 473
 474         adjustDirObjs(outputObj, rewrite)
 475         adjustFileObjs(outputObj, rewrite)
 476
 477         with final.open("cwl.output.json", "w") as f:
 478             res = str(json.dumps(outputObj, sort_keys=True, indent=4, separators=(',',': '), ensure_ascii=False))
 479             f.write(res)
 480
 481
 482         final.save_new(name=name, owner_uuid=self.project_uuid, storage_classes=storage_classes,
 483                        ensure_unique_name=True, properties=output_properties)
 484
 485         logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
 486                     final.api_response()["name"],
 487                     final.manifest_locator())
 488
 489         final_uuid = final.manifest_locator()
 490         tags = tagsString.split(',')
 491         for tag in tags:
 492              self.api.links().create(body={
 493                 "head_uuid": final_uuid, "link_class": "tag", "name": tag
 494                 }).execute(num_retries=self.num_retries)
 495
 496         def finalcollection(fileobj):
 497             fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"])
 498
 499         adjustDirObjs(outputObj, finalcollection)
 500         adjustFileObjs(outputObj, finalcollection)
 501
 502         return (outputObj, final)
 503
 504     def set_crunch_output(self):
 505         if self.work_api == "containers":
 506             current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 507             if current is None:
 508                 return
 509             try:
 510                 self.api.containers().update(uuid=current['uuid'],
 511                                              body={
 512                                                  'output': self.final_output_collection.portable_data_hash(),
 513                                                  'output_properties': self.final_output_collection.get_properties(),
 514                                              }).execute(num_retries=self.num_retries)
 515                 self.api.collections().update(uuid=self.final_output_collection.manifest_locator(),
 516                                               body={
 517                                                   'is_trashed': True
 518                                               }).execute(num_retries=self.num_retries)
 519             except Exception:
 520                 logger.exception("Setting container output")
 521                 raise
 522
 523     def apply_reqs(self, job_order_object, tool):
 524         if "https://w3id.org/cwl/cwl#requirements" in job_order_object:
 525             if tool.metadata.get("http://commonwl.org/cwltool#original_cwlVersion") == 'v1.0':
 526                 raise WorkflowException(
 527                     "`cwl:requirements` in the input object is not part of CWL "
 528                     "v1.0. You can adjust to use `cwltool:overrides` instead; or you "
 529                     "can set the cwlVersion to v1.1 or greater and re-run with "
 530                     "--enable-dev.")
 531             job_reqs = job_order_object["https://w3id.org/cwl/cwl#requirements"]
 532             for req in job_reqs:
 533                 tool.requirements.append(req)
 534
 535     @staticmethod
 536     def get_git_info(tool):
 537         in_a_git_repo = False
 538         cwd = None
 539         filepath = None
 540
 541         if tool.tool["id"].startswith("file://"):
 542             # check if git is installed
 543             try:
 544                 filepath = uri_file_path(tool.tool["id"])
 545                 cwd = os.path.dirname(filepath)
 546                 subprocess.run(["git", "log", "--format=%H", "-n1", "HEAD"], cwd=cwd, check=True, capture_output=True, text=True)
 547                 in_a_git_repo = True
 548             except Exception as e:
 549                 pass
 550
 551         gitproperties = {}
 552
 553         if in_a_git_repo:
 554             git_commit = subprocess.run(["git", "log", "--format=%H", "-n1", "HEAD"], cwd=cwd, capture_output=True, text=True).stdout
 555             git_date = subprocess.run(["git", "log", "--format=%cD", "-n1", "HEAD"], cwd=cwd, capture_output=True, text=True).stdout
 556             git_committer = subprocess.run(["git", "log", "--format=%cn <%ce>", "-n1", "HEAD"], cwd=cwd, capture_output=True, text=True).stdout
 557             git_branch = subprocess.run(["git", "rev-parse", "--abbrev-ref", "HEAD"], cwd=cwd, capture_output=True, text=True).stdout
 558             git_origin = subprocess.run(["git", "remote", "get-url", "origin"], cwd=cwd, capture_output=True, text=True).stdout
 559             git_status = subprocess.run(["git", "status", "--untracked-files=no", "--porcelain"], cwd=cwd, capture_output=True, text=True).stdout
 560             git_describe = subprocess.run(["git", "describe", "--always", "--tags"], cwd=cwd, capture_output=True, text=True).stdout
 561             git_toplevel = subprocess.run(["git", "rev-parse", "--show-toplevel"], cwd=cwd, capture_output=True, text=True).stdout
 562             git_path = filepath[len(git_toplevel):]
 563
 564             gitproperties = {
 565                 "http://arvados.org/cwl#gitCommit": git_commit.strip(),
 566                 "http://arvados.org/cwl#gitDate": git_date.strip(),
 567                 "http://arvados.org/cwl#gitCommitter": git_committer.strip(),
 568                 "http://arvados.org/cwl#gitBranch": git_branch.strip(),
 569                 "http://arvados.org/cwl#gitOrigin": git_origin.strip(),
 570                 "http://arvados.org/cwl#gitStatus": git_status.strip(),
 571                 "http://arvados.org/cwl#gitDescribe": git_describe.strip(),
 572                 "http://arvados.org/cwl#gitPath": git_path.strip(),
 573             }
 574         else:
 575             for g in ("http://arvados.org/cwl#gitCommit",
 576                       "http://arvados.org/cwl#gitDate",
 577                       "http://arvados.org/cwl#gitCommitter",
 578                       "http://arvados.org/cwl#gitBranch",
 579                       "http://arvados.org/cwl#gitOrigin",
 580                       "http://arvados.org/cwl#gitStatus",
 581                       "http://arvados.org/cwl#gitDescribe",
 582                       "http://arvados.org/cwl#gitPath"):
 583                 if g in tool.metadata:
 584                     gitproperties[g] = tool.metadata[g]
 585
 586         return gitproperties
 587
 588     def set_container_request_properties(self, container, properties):
 589         resp = self.api.container_requests().list(filters=[["container_uuid", "=", container["uuid"]]], select=["uuid", "properties"]).execute(num_retries=self.num_retries)
 590         for cr in resp["items"]:
 591             cr["properties"].update({k.replace("http://arvados.org/cwl#", "arv:"): v for k, v in properties.items()})
 592             self.api.container_requests().update(uuid=cr["uuid"], body={"container_request": {"properties": cr["properties"]}}).execute(num_retries=self.num_retries)
 593
 594     def arv_executor(self, updated_tool, job_order, runtimeContext, logger=None):
 595         self.debug = runtimeContext.debug
 596
 597         self.runtime_status_update("activity", "initialization")
 598
 599         git_info = self.get_git_info(updated_tool) if self.git_info else {}
 600         if git_info:
 601             logger.info("Git provenance")
 602             for g in git_info:
 603                 if git_info[g]:
 604                     logger.info("  %s: %s", g.split("#", 1)[1], git_info[g])
 605
 606         runtimeContext.git_info = git_info
 607
 608         workbench1 = self.api.config()["Services"]["Workbench1"]["ExternalURL"]
 609         workbench2 = self.api.config()["Services"]["Workbench2"]["ExternalURL"]
 610         controller = self.api.config()["Services"]["Controller"]["ExternalURL"]
 611         logger.info("Using cluster %s (%s)", self.api.config()["ClusterID"], workbench2 or workbench1 or controller)
 612
 613         if not self.fast_submit:
 614             updated_tool.visit(self.check_features)
 615
 616         self.pipeline = None
 617         self.fs_access = runtimeContext.make_fs_access(runtimeContext.basedir)
 618         self.secret_store = runtimeContext.secret_store
 619
 620         self.trash_intermediate = runtimeContext.trash_intermediate
 621         if self.trash_intermediate and self.work_api != "containers":
 622             raise Exception("--trash-intermediate is only supported with --api=containers.")
 623
 624         self.intermediate_output_ttl = runtimeContext.intermediate_output_ttl
 625         if self.intermediate_output_ttl and self.work_api != "containers":
 626             raise Exception("--intermediate-output-ttl is only supported with --api=containers.")
 627         if self.intermediate_output_ttl < 0:
 628             raise Exception("Invalid value %d for --intermediate-output-ttl, cannot be less than zero" % self.intermediate_output_ttl)
 629
 630         if runtimeContext.submit_request_uuid and self.work_api != "containers":
 631             raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api))
 632
 633         runtimeContext = runtimeContext.copy()
 634
 635         default_storage_classes = ",".join([k for k,v in self.api.config().get("StorageClasses", {"default": {"Default": True}}).items() if v.get("Default") is True])
 636         if runtimeContext.storage_classes == "default":
 637             runtimeContext.storage_classes = default_storage_classes
 638         if runtimeContext.intermediate_storage_classes == "default":
 639             runtimeContext.intermediate_storage_classes = default_storage_classes
 640
 641         if not runtimeContext.name:
 642             self.name = updated_tool.tool.get("label") or updated_tool.metadata.get("label") or os.path.basename(updated_tool.tool["id"])
 643             if git_info.get("http://arvados.org/cwl#gitDescribe"):
 644                 self.name = "%s (%s)" % (self.name, git_info.get("http://arvados.org/cwl#gitDescribe"))
 645             runtimeContext.name = self.name
 646
 647         if runtimeContext.copy_deps is None and (runtimeContext.create_workflow or runtimeContext.update_workflow):
 648             # When creating or updating workflow record, by default
 649             # always copy dependencies and ensure Docker images are up
 650             # to date.
 651             runtimeContext.copy_deps = True
 652             runtimeContext.match_local_docker = True
 653
 654         if runtimeContext.update_workflow and self.project_uuid is None:
 655             # If we are updating a workflow, make sure anything that
 656             # gets uploaded goes into the same parent project, unless
 657             # an alternate --project-uuid was provided.
 658             existing_wf = self.api.workflows().get(uuid=runtimeContext.update_workflow).execute()
 659             runtimeContext.project_uuid = existing_wf["owner_uuid"]
 660
 661         self.project_uuid = runtimeContext.project_uuid
 662
 663         self.runtime_status_update("activity", "data transfer")
 664
 665         # Upload local file references in the job order.
 666         with Perf(metrics, "upload_job_order"):
 667             job_order, jobmapper = upload_job_order(self, "%s input" % runtimeContext.name,
 668                                          updated_tool, job_order, runtimeContext)
 669
 670         # determine if we are submitting or directly executing the workflow.
 671         #
 672         # the last clause means: if it is a command line tool, and we
 673         # are going to wait for the result, and always_submit_runner
 674         # is false, then we don't submit a runner process.
 675
 676         submitting = (runtimeContext.update_workflow or
 677                       runtimeContext.create_workflow or
 678                       (runtimeContext.submit and not
 679                        (updated_tool.tool["class"] == "CommandLineTool" and
 680                         runtimeContext.wait and
 681                         not runtimeContext.always_submit_runner)))
 682
 683         loadingContext = self.loadingContext.copy()
 684         loadingContext.do_validate = False
 685         loadingContext.disable_js_validation = True
 686         tool = updated_tool
 687
 688         # Upload direct dependencies of workflow steps, get back mapping of files to keep references.
 689         # Also uploads docker images.
 690         if not self.fast_submit:
 691             logger.info("Uploading workflow dependencies")
 692             with Perf(metrics, "upload_workflow_deps"):
 693                 merged_map = upload_workflow_deps(self, tool, runtimeContext)
 694         else:
 695             # in the fast submit case, we are running a workflow that
 696             # has already been uploaded to Arvados, so we assume all
 697             # the dependencies have been pinned to keep references and
 698             # there is nothing to do.
 699             merged_map = {}
 700
 701         loadingContext.loader = tool.doc_loader
 702         loadingContext.avsc_names = tool.doc_schema
 703         loadingContext.metadata = tool.metadata
 704         loadingContext.skip_resolve_all = True
 705
 706         workflow_wrapper = None
 707         if submitting and not self.fast_submit:
 708             # upload workflow and get back the workflow wrapper
 709
 710             workflow_wrapper = upload_workflow(self, tool, job_order,
 711                                                runtimeContext.project_uuid,
 712                                                runtimeContext,
 713                                                uuid=runtimeContext.update_workflow,
 714                                                submit_runner_ram=runtimeContext.submit_runner_ram,
 715                                                name=runtimeContext.name,
 716                                                merged_map=merged_map,
 717                                                submit_runner_image=runtimeContext.submit_runner_image,
 718                                                git_info=git_info,
 719                                                set_defaults=(runtimeContext.update_workflow or runtimeContext.create_workflow),
 720                                                jobmapper=jobmapper)
 721
 722             if runtimeContext.update_workflow or runtimeContext.create_workflow:
 723                 # We're registering the workflow, so create or update
 724                 # the workflow record and then exit.
 725                 uuid = make_workflow_record(self, workflow_wrapper, runtimeContext.name, tool,
 726                                             runtimeContext.project_uuid, runtimeContext.update_workflow)
 727                 self.stdout.write(uuid + "\n")
 728                 return (None, "success")
 729
 730             # Did not register a workflow, we're going to submit
 731             # it instead.
 732             loadingContext.loader.idx.clear()
 733             loadingContext.loader.idx["_:main"] = workflow_wrapper
 734             workflow_wrapper["id"] = "_:main"
 735
 736             # Reload the minimal wrapper workflow.
 737             self.fast_submit = True
 738             tool = load_tool(workflow_wrapper, loadingContext)
 739             loadingContext.loader.idx["_:main"] = workflow_wrapper
 740
 741         if not submitting:
 742             # If we are going to run the workflow now (rather than
 743             # submit it), we need to update the workflow document
 744             # replacing file references with keep references.  If we
 745             # are just going to construct a run submission, we don't
 746             # need to do this.
 747             update_from_merged_map(tool, merged_map)
 748
 749         self.apply_reqs(job_order, tool)
 750
 751         self.ignore_docker_for_reuse = runtimeContext.ignore_docker_for_reuse
 752         self.eval_timeout = runtimeContext.eval_timeout
 753
 754         runtimeContext.use_container = True
 755         runtimeContext.tmpdir_prefix = "tmp"
 756         runtimeContext.work_api = self.work_api
 757
 758         if not self.output_name:
 759              self.output_name = "Output from workflow %s" % runtimeContext.name
 760
 761         self.output_name  = cleanup_name_for_collection(self.output_name)
 762
 763         if self.work_api == "containers":
 764             if self.ignore_docker_for_reuse:
 765                 raise Exception("--ignore-docker-for-reuse not supported with containers API.")
 766             runtimeContext.outdir = "/var/spool/cwl"
 767             runtimeContext.docker_outdir = "/var/spool/cwl"
 768             runtimeContext.tmpdir = "/tmp"
 769             runtimeContext.docker_tmpdir = "/tmp"
 770
 771         if runtimeContext.priority < 1 or runtimeContext.priority > 1000:
 772             raise Exception("--priority must be in the range 1..1000.")
 773
 774         if self.should_estimate_cache_size:
 775             visited = set()
 776             estimated_size = [0]
 777             def estimate_collection_cache(obj):
 778                 if obj.get("location", "").startswith("keep:"):
 779                     m = pdh_size.match(obj["location"][5:])
 780                     if m and m.group(1) not in visited:
 781                         visited.add(m.group(1))
 782                         estimated_size[0] += int(m.group(2))
 783             visit_class(job_order, ("File", "Directory"), estimate_collection_cache)
 784             runtimeContext.collection_cache_size = max(((estimated_size[0]*192) // (1024*1024))+1, 256)
 785             self.collection_cache.set_cap(runtimeContext.collection_cache_size*1024*1024)
 786
 787         logger.info("Using collection cache size %s MiB", runtimeContext.collection_cache_size)
 788
 789         runnerjob = None
 790         if runtimeContext.submit:
 791             # We are submitting instead of running immediately.
 792             #
 793             # Create a "Runner job" that when run() is invoked,
 794             # creates the container request to run the workflow.
 795             if self.work_api == "containers":
 796                 if submitting:
 797                     loadingContext.metadata = updated_tool.metadata.copy()
 798                     tool = RunnerContainer(self, tool, loadingContext, runtimeContext.enable_reuse,
 799                                            self.output_name,
 800                                            self.output_tags,
 801                                            submit_runner_ram=runtimeContext.submit_runner_ram,
 802                                            name=runtimeContext.name,
 803                                            on_error=runtimeContext.on_error,
 804                                            submit_runner_image=runtimeContext.submit_runner_image,
 805                                            intermediate_output_ttl=runtimeContext.intermediate_output_ttl,
 806                                            merged_map=merged_map,
 807                                            priority=runtimeContext.priority,
 808                                            secret_store=self.secret_store,
 809                                            collection_cache_size=runtimeContext.collection_cache_size,
 810                                            collection_cache_is_default=self.should_estimate_cache_size,
 811                                            git_info=git_info)
 812                 else:
 813                     runtimeContext.runnerjob = tool.tool["id"]
 814
 815         if runtimeContext.cwl_runner_job is not None:
 816             self.uuid = runtimeContext.cwl_runner_job.get('uuid')
 817
 818         jobiter = tool.job(job_order,
 819                            self.output_callback,
 820                            runtimeContext)
 821
 822         if runtimeContext.submit and not runtimeContext.wait:
 823             # User provided --no-wait so submit the container request,
 824             # get the container request uuid, print it out, and exit.
 825             runnerjob = next(jobiter)
 826             runnerjob.run(runtimeContext)
 827             self.stdout.write(runnerjob.uuid+"\n")
 828             return (None, "success")
 829
 830         # We either running the workflow directly, or submitting it
 831         # and will wait for a final result.
 832
 833         self.runtime_status_update("activity", "workflow execution")
 834
 835         current_container = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger)
 836         if current_container:
 837             logger.info("Running inside container %s", current_container.get("uuid"))
 838             self.set_container_request_properties(current_container, git_info)
 839
 840         self.poll_api = arvados.api('v1', timeout=runtimeContext.http_timeout)
 841         self.polling_thread = threading.Thread(target=self.poll_states)
 842         self.polling_thread.start()
 843
 844         self.task_queue = TaskQueue(self.workflow_eval_lock, self.thread_count)
 845
 846         try:
 847             self.workflow_eval_lock.acquire()
 848
 849             # Holds the lock while this code runs and releases it when
 850             # it is safe to do so in self.workflow_eval_lock.wait(),
 851             # at which point on_message can update job state and
 852             # process output callbacks.
 853
 854             loopperf = Perf(metrics, "jobiter")
 855             loopperf.__enter__()
 856             for runnable in jobiter:
 857                 loopperf.__exit__()
 858
 859                 if self.stop_polling.is_set():
 860                     break
 861
 862                 if self.task_queue.error is not None:
 863                     raise self.task_queue.error
 864
 865                 if runnable:
 866                     with Perf(metrics, "run"):
 867                         self.start_run(runnable, runtimeContext)
 868                 else:
 869                     if (self.task_queue.in_flight + len(self.processes)) > 0:
 870                         self.workflow_eval_lock.wait(3)
 871                     else:
 872                         if self.final_status is None:
 873                             logger.error("Workflow is deadlocked, no runnable processes and not waiting on any pending processes.")
 874                         break
 875
 876                 if self.stop_polling.is_set():
 877                     break
 878
 879                 loopperf.__enter__()
 880             loopperf.__exit__()
 881
 882             while (self.task_queue.in_flight + len(self.processes)) > 0:
 883                 if self.task_queue.error is not None:
 884                     raise self.task_queue.error
 885                 self.workflow_eval_lock.wait(3)
 886
 887         except UnsupportedRequirement:
 888             raise
 889         except:
 890             if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit:
 891                 logger.error("Interrupted, workflow will be cancelled")
 892             elif isinstance(sys.exc_info()[1], WorkflowException):
 893                 logger.error("Workflow execution failed:\n%s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
 894             else:
 895                 logger.exception("Workflow execution failed")
 896
 897             if self.pipeline:
 898                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 899                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
 900
 901             if self.work_api == "containers" and not current_container:
 902                 # Not running in a crunch container, so cancel any outstanding processes.
 903                 for p in self.processes:
 904                     try:
 905                         self.api.container_requests().update(uuid=p,
 906                                                              body={"priority": "0"}
 907                         ).execute(num_retries=self.num_retries)
 908                     except Exception:
 909                         pass
 910         finally:
 911             self.workflow_eval_lock.release()
 912             self.task_queue.drain()
 913             self.stop_polling.set()
 914             self.polling_thread.join()
 915             self.task_queue.join()
 916
 917         if self.final_status == "UnsupportedRequirement":
 918             raise UnsupportedRequirement("Check log for details.")
 919
 920         if self.final_output is None:
 921             raise WorkflowException("Workflow did not return a result.")
 922
 923         if runtimeContext.submit and isinstance(tool, Runner):
 924             logger.info("Final output collection %s", tool.final_output)
 925             if workbench2 or workbench1:
 926                 logger.info("Output at %scollections/%s", workbench2 or workbench1, tool.final_output)
 927         else:
 928             if self.output_tags is None:
 929                 self.output_tags = ""
 930
 931             storage_classes = ""
 932             storage_class_req, _ = tool.get_requirement("http://arvados.org/cwl#OutputStorageClass")
 933             if storage_class_req and storage_class_req.get("finalStorageClass"):
 934                 storage_classes = aslist(storage_class_req["finalStorageClass"])
 935             else:
 936                 storage_classes = runtimeContext.storage_classes.strip().split(",")
 937
 938             output_properties = {}
 939             output_properties_req, _ = tool.get_requirement("http://arvados.org/cwl#OutputCollectionProperties")
 940             if output_properties_req:
 941                 builder = make_builder(job_order, tool.hints, tool.requirements, runtimeContext, tool.metadata)
 942                 for pr in output_properties_req["outputProperties"]:
 943                     output_properties[pr["propertyName"]] = builder.do_eval(pr["propertyValue"])
 944
 945             self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, storage_classes,
 946                                                                                           self.output_tags, output_properties,
 947                                                                                           self.final_output)
 948             self.set_crunch_output()
 949
 950         if runtimeContext.compute_checksum:
 951             adjustDirObjs(self.final_output, partial(get_listing, self.fs_access))
 952             adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
 953
 954         if self.trash_intermediate and self.final_status == "success":
 955             self.trash_intermediate_output()
 956
 957         return (self.final_output, self.final_status)