sdk/cwl/arvados_cwl/arvcontainer.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 from future import standard_library
   6 standard_library.install_aliases()
   7 from builtins import str
   8
   9 import logging
  10 import json
  11 import os
  12 import urllib.request, urllib.parse, urllib.error
  13 import time
  14 import datetime
  15 import ciso8601
  16 import uuid
  17 import math
  18 import re
  19
  20 import arvados_cwl.util
  21 import ruamel.yaml
  22
  23 from cwltool.errors import WorkflowException
  24 from cwltool.process import UnsupportedRequirement, shortname
  25 from cwltool.utils import aslist, adjustFileObjs, adjustDirObjs, visit_class
  26 from cwltool.job import JobBase
  27
  28 import arvados.collection
  29
  30 from .arvdocker import arv_docker_get_image
  31 from . import done
  32 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields, make_builder
  33 from .fsaccess import CollectionFetcher
  34 from .pathmapper import NoFollowPathMapper, trim_listing
  35 from .perf import Perf
  36 from ._version import __version__
  37
  38 logger = logging.getLogger('arvados.cwl-runner')
  39 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  40
  41 def cleanup_name_for_collection(name):
  42     return name.replace("/", " ")
  43
  44 class ArvadosContainer(JobBase):
  45     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  46
  47     def __init__(self, runner, job_runtime,
  48                  builder,   # type: Builder
  49                  joborder,  # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
  50                  make_path_mapper,  # type: Callable[..., PathMapper]
  51                  requirements,      # type: List[Dict[Text, Text]]
  52                  hints,     # type: List[Dict[Text, Text]]
  53                  name       # type: Text
  54     ):
  55         super(ArvadosContainer, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
  56         self.arvrunner = runner
  57         self.job_runtime = job_runtime
  58         self.running = False
  59         self.uuid = None
  60         self.attempt_count = 0
  61
  62     def update_pipeline_component(self, r):
  63         pass
  64
  65     def _required_env(self):
  66         env = {}
  67         env["HOME"] = self.outdir
  68         env["TMPDIR"] = self.tmpdir
  69         return env
  70
  71     def run(self, toplevelRuntimeContext):
  72         # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
  73         # which calls makeJobRunner() to get a new ArvadosContainer
  74         # object.  The fields that define execution such as
  75         # command_line, environment, etc are set on the
  76         # ArvadosContainer object by CommandLineTool.job() before
  77         # run() is called.
  78
  79         runtimeContext = self.job_runtime
  80
  81         if runtimeContext.submit_request_uuid:
  82             container_request = self.arvrunner.api.container_requests().get(
  83                 uuid=runtimeContext.submit_request_uuid
  84             ).execute(num_retries=self.arvrunner.num_retries)
  85         else:
  86             container_request = {}
  87
  88         container_request["command"] = self.command_line
  89         container_request["name"] = self.name
  90         container_request["output_path"] = self.outdir
  91         container_request["cwd"] = self.outdir
  92         container_request["priority"] = runtimeContext.priority
  93         container_request["state"] = "Uncommitted"
  94         container_request.setdefault("properties", {})
  95
  96         container_request["properties"]["cwl_input"] = self.joborder
  97
  98         runtime_constraints = {}
  99
 100         if runtimeContext.project_uuid:
 101             container_request["owner_uuid"] = runtimeContext.project_uuid
 102
 103         if self.arvrunner.secret_store.has_secret(self.command_line):
 104             raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
 105
 106         if self.arvrunner.secret_store.has_secret(self.environment):
 107             raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
 108
 109         resources = self.builder.resources
 110         if resources is not None:
 111             runtime_constraints["vcpus"] = math.ceil(resources.get("cores", 1))
 112             runtime_constraints["ram"] = math.ceil(resources.get("ram") * 2**20)
 113
 114         mounts = {
 115             self.outdir: {
 116                 "kind": "tmp",
 117                 "capacity": math.ceil(resources.get("outdirSize", 0) * 2**20)
 118             },
 119             self.tmpdir: {
 120                 "kind": "tmp",
 121                 "capacity": math.ceil(resources.get("tmpdirSize", 0) * 2**20)
 122             }
 123         }
 124         secret_mounts = {}
 125         scheduling_parameters = {}
 126
 127         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
 128         rf.sort(key=lambda k: k.resolved)
 129         prevdir = None
 130         for resolved, target, tp, stg in rf:
 131             if not stg:
 132                 continue
 133             if prevdir and target.startswith(prevdir):
 134                 continue
 135             if tp == "Directory":
 136                 targetdir = target
 137             else:
 138                 targetdir = os.path.dirname(target)
 139             sp = resolved.split("/", 1)
 140             pdh = sp[0][5:]   # remove "keep:"
 141             mounts[targetdir] = {
 142                 "kind": "collection",
 143                 "portable_data_hash": pdh
 144             }
 145             if pdh in self.pathmapper.pdh_to_uuid:
 146                 mounts[targetdir]["uuid"] = self.pathmapper.pdh_to_uuid[pdh]
 147             if len(sp) == 2:
 148                 if tp == "Directory":
 149                     path = sp[1]
 150                 else:
 151                     path = os.path.dirname(sp[1])
 152                 if path and path != "/":
 153                     mounts[targetdir]["path"] = path
 154             prevdir = targetdir + "/"
 155
 156         intermediate_collection_info = arvados_cwl.util.get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl)
 157
 158         with Perf(metrics, "generatefiles %s" % self.name):
 159             if self.generatefiles["listing"]:
 160                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
 161                                                     keep_client=self.arvrunner.keep_client,
 162                                                     num_retries=self.arvrunner.num_retries)
 163                 generatemapper = NoFollowPathMapper(self.generatefiles["listing"], "", "",
 164                                                     separateDirs=False)
 165
 166                 sorteditems = sorted(generatemapper.items(), key=lambda n: n[1].target)
 167
 168                 logger.debug("generatemapper is %s", sorteditems)
 169
 170                 with Perf(metrics, "createfiles %s" % self.name):
 171                     for f, p in sorteditems:
 172                         if not p.target:
 173                             continue
 174
 175                         if p.target.startswith("/"):
 176                             dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:]
 177                         else:
 178                             dst = p.target
 179
 180                         if p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 181                             if p.resolved.startswith("_:"):
 182                                 vwd.mkdirs(dst)
 183                             else:
 184                                 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 185                                 vwd.copy(path or ".", dst, source_collection=source)
 186                         elif p.type == "CreateFile":
 187                             if self.arvrunner.secret_store.has_secret(p.resolved):
 188                                 mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target)
 189                                 secret_mounts[mountpoint] = {
 190                                     "kind": "text",
 191                                     "content": self.arvrunner.secret_store.retrieve(p.resolved)
 192                                 }
 193                             else:
 194                                 with vwd.open(dst, "w") as n:
 195                                     n.write(p.resolved)
 196
 197                 def keepemptydirs(p):
 198                     if isinstance(p, arvados.collection.RichCollectionBase):
 199                         if len(p) == 0:
 200                             p.open(".keep", "w").close()
 201                         else:
 202                             for c in p:
 203                                 keepemptydirs(p[c])
 204
 205                 keepemptydirs(vwd)
 206
 207                 if not runtimeContext.current_container:
 208                     runtimeContext.current_container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
 209                 vwd.save_new(name=intermediate_collection_info["name"],
 210                              owner_uuid=runtimeContext.project_uuid,
 211                              ensure_unique_name=True,
 212                              trash_at=intermediate_collection_info["trash_at"],
 213                              properties=intermediate_collection_info["properties"])
 214
 215                 prev = None
 216                 for f, p in sorteditems:
 217                     if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
 218                         (prev is not None and p.target.startswith(prev))):
 219                         continue
 220                     if p.target.startswith("/"):
 221                         dst = p.target[len(self.outdir)+1:] if p.target.startswith(self.outdir+"/") else p.target[1:]
 222                     else:
 223                         dst = p.target
 224                     mountpoint = p.target if p.target.startswith("/") else os.path.join(self.outdir, p.target)
 225                     mounts[mountpoint] = {"kind": "collection",
 226                                           "portable_data_hash": vwd.portable_data_hash(),
 227                                           "path": dst}
 228                     if p.type.startswith("Writable"):
 229                         mounts[mountpoint]["writable"] = True
 230                     prev = p.target + "/"
 231
 232         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 233         if self.environment:
 234             container_request["environment"].update(self.environment)
 235
 236         if self.stdin:
 237             sp = self.stdin[6:].split("/", 1)
 238             mounts["stdin"] = {"kind": "collection",
 239                                 "portable_data_hash": sp[0],
 240                                 "path": sp[1]}
 241
 242         if self.stderr:
 243             mounts["stderr"] = {"kind": "file",
 244                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 245
 246         if self.stdout:
 247             mounts["stdout"] = {"kind": "file",
 248                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 249
 250         (docker_req, docker_is_req) = self.get_requirement("DockerRequirement")
 251
 252         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 253                                                                     docker_req,
 254                                                                     runtimeContext.pull_image,
 255                                                                     runtimeContext)
 256
 257         network_req, _ = self.get_requirement("NetworkAccess")
 258         if network_req:
 259             runtime_constraints["API"] = network_req["networkAccess"]
 260
 261         api_req, _ = self.get_requirement("http://arvados.org/cwl#APIRequirement")
 262         if api_req:
 263             runtime_constraints["API"] = True
 264
 265         use_disk_cache = (self.arvrunner.api.config()["Containers"].get("DefaultKeepCacheRAM", 0) == 0)
 266
 267         keep_cache_type_req, _ = self.get_requirement("http://arvados.org/cwl#KeepCacheTypeRequirement")
 268         if keep_cache_type_req:
 269             if "keepCacheType" in keep_cache_type_req:
 270                 if keep_cache_type_req["keepCacheType"] == "ram_cache":
 271                     use_disk_cache = False
 272
 273         runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints")
 274         if runtime_req:
 275             if "keep_cache" in runtime_req:
 276                 if use_disk_cache:
 277                     # If DefaultKeepCacheRAM is zero it means we should use disk cache.
 278                     runtime_constraints["keep_cache_disk"] = math.ceil(runtime_req["keep_cache"] * 2**20)
 279                 else:
 280                     runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20)
 281             if "outputDirType" in runtime_req:
 282                 if runtime_req["outputDirType"] == "local_output_dir":
 283                     # Currently the default behavior.
 284                     pass
 285                 elif runtime_req["outputDirType"] == "keep_output_dir":
 286                     mounts[self.outdir]= {
 287                         "kind": "collection",
 288                         "writable": True
 289                     }
 290
 291         partition_req, _ = self.get_requirement("http://arvados.org/cwl#PartitionRequirement")
 292         if partition_req:
 293             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 294
 295         intermediate_output_req, _ = self.get_requirement("http://arvados.org/cwl#IntermediateOutput")
 296         if intermediate_output_req:
 297             self.output_ttl = intermediate_output_req["outputTTL"]
 298         else:
 299             self.output_ttl = self.arvrunner.intermediate_output_ttl
 300
 301         if self.output_ttl < 0:
 302             raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 303
 304
 305         if self.arvrunner.api._rootDesc["revision"] >= "20210628":
 306             storage_class_req, _ = self.get_requirement("http://arvados.org/cwl#OutputStorageClass")
 307             if storage_class_req and storage_class_req.get("intermediateStorageClass"):
 308                 container_request["output_storage_classes"] = aslist(storage_class_req["intermediateStorageClass"])
 309             else:
 310                 container_request["output_storage_classes"] = runtimeContext.intermediate_storage_classes.strip().split(",")
 311
 312         cuda_req, _ = self.get_requirement("http://commonwl.org/cwltool#CUDARequirement")
 313         if cuda_req:
 314             runtime_constraints["cuda"] = {
 315                 "device_count": resources.get("cudaDeviceCount", 1),
 316                 "driver_version": cuda_req["cudaVersionMin"],
 317                 "hardware_capability": aslist(cuda_req["cudaComputeCapability"])[0]
 318             }
 319
 320         if runtimeContext.enable_preemptible is False:
 321             scheduling_parameters["preemptible"] = False
 322         else:
 323             preemptible_req, _ = self.get_requirement("http://arvados.org/cwl#UsePreemptible")
 324             if preemptible_req:
 325                 scheduling_parameters["preemptible"] = preemptible_req["usePreemptible"]
 326             elif runtimeContext.enable_preemptible is True:
 327                 scheduling_parameters["preemptible"] = True
 328             elif runtimeContext.enable_preemptible is None:
 329                 pass
 330
 331         if self.timelimit is not None and self.timelimit > 0:
 332             scheduling_parameters["max_run_time"] = self.timelimit
 333
 334         extra_submit_params = {}
 335         if runtimeContext.submit_runner_cluster:
 336             extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 337
 338         container_request["output_name"] = cleanup_name_for_collection("Output from step %s" % (self.name))
 339         container_request["output_ttl"] = self.output_ttl
 340         container_request["mounts"] = mounts
 341         container_request["secret_mounts"] = secret_mounts
 342         container_request["runtime_constraints"] = runtime_constraints
 343         container_request["scheduling_parameters"] = scheduling_parameters
 344
 345         enable_reuse = runtimeContext.enable_reuse
 346         if enable_reuse:
 347             reuse_req, _ = self.get_requirement("WorkReuse")
 348             if reuse_req:
 349                 enable_reuse = reuse_req["enableReuse"]
 350             reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
 351             if reuse_req:
 352                 enable_reuse = reuse_req["enableReuse"]
 353         container_request["use_existing"] = enable_reuse
 354
 355         properties_req, _ = self.get_requirement("http://arvados.org/cwl#ProcessProperties")
 356         if properties_req:
 357             for pr in properties_req["processProperties"]:
 358                 container_request["properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
 359
 360         output_properties_req, _ = self.get_requirement("http://arvados.org/cwl#OutputCollectionProperties")
 361         if output_properties_req:
 362             if self.arvrunner.api._rootDesc["revision"] >= "20220510":
 363                 container_request["output_properties"] = {}
 364                 for pr in output_properties_req["outputProperties"]:
 365                     container_request["output_properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
 366             else:
 367                 logger.warning("%s API revision is %s, revision %s is required to support setting properties on output collections.",
 368                                self.arvrunner.label(self), self.arvrunner.api._rootDesc["revision"], "20220510")
 369
 370         if runtimeContext.runnerjob.startswith("arvwf:"):
 371             wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")]
 372             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 373             if container_request["name"] == "main":
 374                 container_request["name"] = wfrecord["name"]
 375             container_request["properties"]["template_uuid"] = wfuuid
 376
 377         if self.attempt_count == 0:
 378             self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 379
 380         try:
 381             ram = runtime_constraints["ram"]
 382
 383             for i in range(1, 4):
 384                 runtime_constraints["ram"] = ram * i
 385
 386                 if runtimeContext.submit_request_uuid:
 387                     response = self.arvrunner.api.container_requests().update(
 388                         uuid=runtimeContext.submit_request_uuid,
 389                         body=container_request,
 390                         **extra_submit_params
 391                     ).execute(num_retries=self.arvrunner.num_retries)
 392                 else:
 393                     response = self.arvrunner.api.container_requests().create(
 394                         body=container_request,
 395                         **extra_submit_params
 396                     ).execute(num_retries=self.arvrunner.num_retries)
 397                     runtimeContext.submit_request_uuid = response["uuid"]
 398
 399                 if response["container_uuid"] is not None:
 400                     break
 401
 402             if response["container_uuid"] is None:
 403                 runtime_constraints["ram"] = ram * (self.attempt_count+1)
 404
 405             container_request["state"] = "Committed"
 406             response = self.arvrunner.api.container_requests().update(
 407                 uuid=runtimeContext.submit_request_uuid,
 408                 body=container_request,
 409                 **extra_submit_params
 410             ).execute(num_retries=self.arvrunner.num_retries)
 411
 412             self.uuid = response["uuid"]
 413             self.arvrunner.process_submitted(self)
 414             self.attempt_count += 1
 415
 416             if response["state"] == "Final":
 417                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 418             else:
 419                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 420         except Exception as e:
 421             logger.exception("%s error submitting container\n%s", self.arvrunner.label(self), e)
 422             logger.debug("Container request was %s", container_request)
 423             self.output_callback({}, "permanentFail")
 424
 425     def out_of_memory_retry(self, record, container):
 426         logc = arvados.collection.CollectionReader(record["log_uuid"],
 427                                                    api_client=self.arvrunner.api,
 428                                                    keep_client=self.arvrunner.keep_client,
 429                                                    num_retries=self.arvrunner.num_retries)
 430
 431         loglines = [""]
 432         def callback(v1, v2, v3):
 433             loglines[0] = v3
 434
 435         done.logtail(logc, callback, "", maxlen=1000)
 436
 437         # Check OOM killed
 438         oom_matches = r'container using over 9.% of memory'
 439         if container["exit_code"] == 137 and re.search(oom_matches, loglines[0], re.IGNORECASE | re.MULTILINE):
 440             return True
 441
 442         # Check allocation failure
 443         bad_alloc_matches = r'(bad_alloc|out ?of ?memory)'
 444         if re.search(bad_alloc_matches, loglines[0], re.IGNORECASE | re.MULTILINE):
 445             return True
 446
 447         return False
 448
 449     def done(self, record):
 450         outputs = {}
 451         retried = False
 452         try:
 453             container = self.arvrunner.api.containers().get(
 454                 uuid=record["container_uuid"]
 455             ).execute(num_retries=self.arvrunner.num_retries)
 456             if container["state"] == "Complete":
 457                 rcode = container["exit_code"]
 458                 if self.successCodes and rcode in self.successCodes:
 459                     processStatus = "success"
 460                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 461                     processStatus = "temporaryFail"
 462                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 463                     processStatus = "permanentFail"
 464                 elif rcode == 0:
 465                     processStatus = "success"
 466                 else:
 467                     processStatus = "permanentFail"
 468
 469                 if processStatus == "permanentFail" and self.out_of_memory_retry(record, container):
 470                     logger.info("%s Container failed with out of memory error, retrying with more RAM.",
 471                                  self.arvrunner.label(self))
 472                     self.job_runtime.submit_request_uuid = None
 473                     self.uuid = None
 474                     self.run(None)
 475                     retried = True
 476                     return
 477
 478                 if rcode == 137:
 479                     logger.warning("%s Container may have been killed for using too much RAM.  Try resubmitting with a higher 'ramMin'.",
 480                                  self.arvrunner.label(self))
 481             else:
 482                 processStatus = "permanentFail"
 483
 484             if processStatus == "permanentFail" and record["log_uuid"]:
 485                 logc = arvados.collection.CollectionReader(record["log_uuid"],
 486                                                            api_client=self.arvrunner.api,
 487                                                            keep_client=self.arvrunner.keep_client,
 488                                                            num_retries=self.arvrunner.num_retries)
 489                 label = self.arvrunner.label(self)
 490                 done.logtail(
 491                     logc, logger.error,
 492                     "%s (%s) error log:" % (label, record["uuid"]), maxlen=40)
 493
 494             if record["output_uuid"]:
 495                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 496                     # Compute the trash time to avoid requesting the collection record.
 497                     trash_at = ciso8601.parse_datetime_as_naive(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 498                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 499                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 500                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 501                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 502                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 503                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 504
 505             if container["output"]:
 506                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 507
 508             properties = record["properties"].copy()
 509             properties["cwl_output"] = outputs
 510             self.arvrunner.api.container_requests().update(
 511                 uuid=self.uuid,
 512                 body={"container_request": {"properties": properties}}
 513             ).execute(num_retries=self.arvrunner.num_retries)
 514         except WorkflowException as e:
 515             # Only include a stack trace if in debug mode.
 516             # A stack trace may obfuscate more useful output about the workflow.
 517             logger.error("%s unable to collect output from %s:\n%s",
 518                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 519             processStatus = "permanentFail"
 520         except Exception:
 521             logger.exception("%s while getting output object:", self.arvrunner.label(self))
 522             processStatus = "permanentFail"
 523         finally:
 524             if not retried:
 525                 self.output_callback(outputs, processStatus)
 526
 527
 528 class RunnerContainer(Runner):
 529     """Submit and manage a container that runs arvados-cwl-runner."""
 530
 531     def arvados_job_spec(self, runtimeContext, git_info):
 532         """Create an Arvados container request for this workflow.
 533
 534         The returned dict can be used to create a container passed as
 535         the +body+ argument to container_requests().create().
 536         """
 537
 538         adjustDirObjs(self.job_order, trim_listing)
 539         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 540         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 541
 542         secret_mounts = {}
 543         for param in sorted(self.job_order.keys()):
 544             if self.secret_store.has_secret(self.job_order[param]):
 545                 mnt = "/secrets/s%d" % len(secret_mounts)
 546                 secret_mounts[mnt] = {
 547                     "kind": "text",
 548                     "content": self.secret_store.retrieve(self.job_order[param])
 549                 }
 550                 self.job_order[param] = {"$include": mnt}
 551
 552         container_req = {
 553             "name": self.name,
 554             "output_path": "/var/spool/cwl",
 555             "cwd": "/var/spool/cwl",
 556             "priority": self.priority,
 557             "state": "Committed",
 558             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image, runtimeContext),
 559             "mounts": {
 560                 "/var/lib/cwl/cwl.input.json": {
 561                     "kind": "json",
 562                     "content": self.job_order
 563                 },
 564                 "stdout": {
 565                     "kind": "file",
 566                     "path": "/var/spool/cwl/cwl.output.json"
 567                 },
 568                 "/var/spool/cwl": {
 569                     "kind": "collection",
 570                     "writable": True
 571                 }
 572             },
 573             "secret_mounts": secret_mounts,
 574             "runtime_constraints": {
 575                 "vcpus": math.ceil(self.submit_runner_cores),
 576                 "ram": 1024*1024 * (math.ceil(self.submit_runner_ram) + math.ceil(self.collection_cache_size)),
 577                 "API": True
 578             },
 579             "use_existing": False, # Never reuse the runner container - see #15497.
 580             "properties": {}
 581         }
 582
 583         if self.embedded_tool.tool.get("id", "").startswith("keep:"):
 584             sp = self.embedded_tool.tool["id"].split('/')
 585             workflowcollection = sp[0][5:]
 586             workflowname = "/".join(sp[1:])
 587             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 588             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 589                 "kind": "collection",
 590                 "portable_data_hash": "%s" % workflowcollection
 591             }
 592         elif self.embedded_tool.tool.get("id", "").startswith("arvwf:"):
 593             uuid, frg = urllib.parse.urldefrag(self.embedded_tool.tool["id"])
 594             workflowpath = "/var/lib/cwl/workflow.json#" + frg
 595             packedtxt = self.loadingContext.loader.fetch_text(uuid)
 596             yaml = ruamel.yaml.YAML(typ='safe', pure=True)
 597             packed = yaml.load(packedtxt)
 598             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 599                 "kind": "json",
 600                 "content": packed
 601             }
 602             container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33]
 603         else:
 604             main = self.loadingContext.loader.idx["_:main"]
 605             if main.get("id") == "_:main":
 606                 del main["id"]
 607             workflowpath = "/var/lib/cwl/workflow.json#main"
 608             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 609                 "kind": "json",
 610                 "content": main
 611             }
 612
 613         container_req["properties"].update({k.replace("http://arvados.org/cwl#", "arv:"): v for k, v in git_info.items()})
 614
 615         properties_req, _ = self.embedded_tool.get_requirement("http://arvados.org/cwl#ProcessProperties")
 616         if properties_req:
 617             builder = make_builder(self.job_order, self.embedded_tool.hints, self.embedded_tool.requirements, runtimeContext, self.embedded_tool.metadata)
 618             for pr in properties_req["processProperties"]:
 619                 container_req["properties"][pr["propertyName"]] = builder.do_eval(pr["propertyValue"])
 620
 621         # --local means execute the workflow instead of submitting a container request
 622         # --api=containers means use the containers API
 623         # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
 624         # --disable-validate because we already validated so don't need to do it again
 625         # --eval-timeout is the timeout for javascript invocation
 626         # --parallel-task-count is the number of threads to use for job submission
 627         # --enable/disable-reuse sets desired job reuse
 628         # --collection-cache-size sets aside memory to store collections
 629         command = ["arvados-cwl-runner",
 630                    "--local",
 631                    "--api=containers",
 632                    "--no-log-timestamps",
 633                    "--disable-validate",
 634                    "--disable-color",
 635                    "--eval-timeout=%s" % self.arvrunner.eval_timeout,
 636                    "--thread-count=%s" % self.arvrunner.thread_count,
 637                    "--enable-reuse" if self.enable_reuse else "--disable-reuse",
 638                    "--collection-cache-size=%s" % self.collection_cache_size]
 639
 640         if self.output_name:
 641             command.append("--output-name=" + self.output_name)
 642             container_req["output_name"] = self.output_name
 643
 644         if self.output_tags:
 645             command.append("--output-tags=" + self.output_tags)
 646
 647         if runtimeContext.debug:
 648             command.append("--debug")
 649
 650         if runtimeContext.storage_classes != "default" and runtimeContext.storage_classes:
 651             command.append("--storage-classes=" + runtimeContext.storage_classes)
 652
 653         if runtimeContext.intermediate_storage_classes != "default" and runtimeContext.intermediate_storage_classes:
 654             command.append("--intermediate-storage-classes=" + runtimeContext.intermediate_storage_classes)
 655
 656         if runtimeContext.on_error:
 657             command.append("--on-error=" + self.on_error)
 658
 659         if runtimeContext.intermediate_output_ttl:
 660             command.append("--intermediate-output-ttl=%d" % runtimeContext.intermediate_output_ttl)
 661
 662         if runtimeContext.trash_intermediate:
 663             command.append("--trash-intermediate")
 664
 665         if runtimeContext.project_uuid:
 666             command.append("--project-uuid="+runtimeContext.project_uuid)
 667
 668         if self.enable_dev:
 669             command.append("--enable-dev")
 670
 671         if runtimeContext.enable_preemptible is True:
 672             command.append("--enable-preemptible")
 673
 674         if runtimeContext.enable_preemptible is False:
 675             command.append("--disable-preemptible")
 676
 677         if runtimeContext.varying_url_params:
 678             command.append("--varying-url-params="+runtimeContext.varying_url_params)
 679
 680         if runtimeContext.prefer_cached_downloads:
 681             command.append("--prefer-cached-downloads")
 682
 683         if self.fast_parser:
 684             command.append("--fast-parser")
 685
 686         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 687
 688         container_req["command"] = command
 689
 690         return container_req
 691
 692
 693     def run(self, runtimeContext):
 694         runtimeContext.keepprefix = "keep:"
 695         job_spec = self.arvados_job_spec(runtimeContext, self.git_info)
 696         if runtimeContext.project_uuid:
 697             job_spec["owner_uuid"] = runtimeContext.project_uuid
 698
 699         extra_submit_params = {}
 700         if runtimeContext.submit_runner_cluster:
 701             extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 702
 703         if runtimeContext.submit_request_uuid:
 704             if "cluster_id" in extra_submit_params:
 705                 # Doesn't make sense for "update" and actually fails
 706                 del extra_submit_params["cluster_id"]
 707             response = self.arvrunner.api.container_requests().update(
 708                 uuid=runtimeContext.submit_request_uuid,
 709                 body=job_spec,
 710                 **extra_submit_params
 711             ).execute(num_retries=self.arvrunner.num_retries)
 712         else:
 713             response = self.arvrunner.api.container_requests().create(
 714                 body=job_spec,
 715                 **extra_submit_params
 716             ).execute(num_retries=self.arvrunner.num_retries)
 717
 718         self.uuid = response["uuid"]
 719         self.arvrunner.process_submitted(self)
 720
 721         logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"])
 722
 723         workbench1 = self.arvrunner.api.config()["Services"]["Workbench1"]["ExternalURL"]
 724         workbench2 = self.arvrunner.api.config()["Services"]["Workbench2"]["ExternalURL"]
 725         url = ""
 726         if workbench2:
 727             url = "{}processes/{}".format(workbench2, response["uuid"])
 728         elif workbench1:
 729             url = "{}container_requests/{}".format(workbench1, response["uuid"])
 730         if url:
 731             logger.info("Monitor workflow progress at %s", url)
 732
 733
 734     def done(self, record):
 735         try:
 736             container = self.arvrunner.api.containers().get(
 737                 uuid=record["container_uuid"]
 738             ).execute(num_retries=self.arvrunner.num_retries)
 739             container["log"] = record["log_uuid"]
 740         except Exception:
 741             logger.exception("%s while getting runner container", self.arvrunner.label(self))
 742             self.arvrunner.output_callback({}, "permanentFail")
 743         else:
 744             super(RunnerContainer, self).done(container)