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 import logging
   6 import json
   7 import os
   8 import urllib.request, urllib.parse, urllib.error
   9 import time
  10 import datetime
  11 import ciso8601
  12 import uuid
  13 import math
  14 import re
  15
  16 import arvados_cwl.util
  17 import ruamel.yaml
  18
  19 from cwltool.errors import WorkflowException
  20 from cwltool.process import UnsupportedRequirement, shortname
  21 from cwltool.utils import aslist, adjustFileObjs, adjustDirObjs, visit_class
  22 from cwltool.job import JobBase
  23
  24 import arvados.collection
  25
  26 import crunchstat_summary.summarizer
  27 import crunchstat_summary.reader
  28
  29 from .arvdocker import arv_docker_get_image
  30 from . import done
  31 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields, make_builder
  32 from .fsaccess import CollectionFetcher
  33 from .pathmapper import NoFollowPathMapper, trim_listing
  34 from .perf import Perf
  35 from ._version import __version__
  36
  37 logger = logging.getLogger('arvados.cwl-runner')
  38 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  39
  40 def cleanup_name_for_collection(name):
  41     return name.replace("/", " ")
  42
  43 class ArvadosContainer(JobBase):
  44     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  45
  46     def __init__(self, runner, job_runtime, globpatterns,
  47                  builder,   # type: Builder
  48                  joborder,  # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
  49                  make_path_mapper,  # type: Callable[..., PathMapper]
  50                  requirements,      # type: List[Dict[Text, Text]]
  51                  hints,     # type: List[Dict[Text, Text]]
  52                  name       # type: Text
  53     ):
  54         super(ArvadosContainer, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
  55         self.arvrunner = runner
  56         self.job_runtime = job_runtime
  57         self.running = False
  58         self.uuid = None
  59         self.attempt_count = 0
  60         self.globpatterns = globpatterns
  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             if self.arvrunner.api._rootDesc["revision"] >= "20250128":
 315                 # Arvados 3.1+ API
 316                 runtime_constraints["gpu"] = {
 317                     "stack": "cuda",
 318                     "device_count": resources.get("cudaDeviceCount", 1),
 319                     "driver_version": cuda_req["cudaVersionMin"],
 320                     "hardware_target": aslist(cuda_req["cudaComputeCapability"]),
 321                     "vram": cuda_req["cudaVram"]*1024*1024,
 322                 }
 323             else:
 324                 # Legacy API
 325                 runtime_constraints["cuda"] = {
 326                     "device_count": resources.get("cudaDeviceCount", 1),
 327                     "driver_version": cuda_req["cudaVersionMin"],
 328                     "hardware_capability": aslist(cuda_req["cudaComputeCapability"])[0]
 329                 }
 330
 331         rocm_req, _ = self.get_requirement("http://arvados.org/cwl#ROCmRequirement")
 332         if rocm_req:
 333             if self.arvrunner.api._rootDesc["revision"] >= "20250128":
 334                 runtime_constraints["gpu"] = {
 335                     "stack": "rocm",
 336                     "device_count": rocm_req["rocmDeviceCountMin"],
 337                     "driver_version": rocm_req["rocmDriverVersion"],
 338                     "hardware_target": aslist(rocm_req["rocmTarget"]),
 339                     "vram": rocm_req["rocmVram"]*1024*1024,
 340                 }
 341             else:
 342                 raise WorkflowException("Arvados API server does not support ROCm (requires Arvados 3.1+)")
 343
 344         if runtimeContext.enable_preemptible is False:
 345             scheduling_parameters["preemptible"] = False
 346         else:
 347             preemptible_req, _ = self.get_requirement("http://arvados.org/cwl#UsePreemptible")
 348             if preemptible_req:
 349                 scheduling_parameters["preemptible"] = preemptible_req["usePreemptible"]
 350             elif runtimeContext.enable_preemptible is True:
 351                 scheduling_parameters["preemptible"] = True
 352             elif runtimeContext.enable_preemptible is None:
 353                 pass
 354
 355         if scheduling_parameters.get("preemptible") and self.may_resubmit_non_preemptible():
 356             # Only make one attempt, because if it is preempted we
 357             # will resubmit and ask for a non-preemptible instance.
 358             container_request["container_count_max"] = 1
 359
 360         if self.timelimit is not None and self.timelimit > 0:
 361             scheduling_parameters["max_run_time"] = self.timelimit
 362
 363         extra_submit_params = {}
 364         if runtimeContext.submit_runner_cluster:
 365             extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 366
 367         container_request["output_name"] = cleanup_name_for_collection("Output from step %s" % (self.name))
 368         container_request["output_ttl"] = self.output_ttl
 369         container_request["mounts"] = mounts
 370         container_request["secret_mounts"] = secret_mounts
 371         container_request["runtime_constraints"] = runtime_constraints
 372         container_request["scheduling_parameters"] = scheduling_parameters
 373
 374         enable_reuse = runtimeContext.enable_reuse
 375         if enable_reuse:
 376             reuse_req, _ = self.get_requirement("WorkReuse")
 377             if reuse_req:
 378                 enable_reuse = reuse_req["enableReuse"]
 379             reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
 380             if reuse_req:
 381                 enable_reuse = reuse_req["enableReuse"]
 382         container_request["use_existing"] = enable_reuse
 383
 384         properties_req, _ = self.get_requirement("http://arvados.org/cwl#ProcessProperties")
 385         if properties_req:
 386             for pr in properties_req["processProperties"]:
 387                 container_request["properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
 388
 389         output_properties_req, _ = self.get_requirement("http://arvados.org/cwl#OutputCollectionProperties")
 390         if output_properties_req:
 391             if self.arvrunner.api._rootDesc["revision"] >= "20220510":
 392                 container_request["output_properties"] = {}
 393                 for pr in output_properties_req["outputProperties"]:
 394                     container_request["output_properties"][pr["propertyName"]] = self.builder.do_eval(pr["propertyValue"])
 395             else:
 396                 logger.warning("%s API revision is %s, revision %s is required to support setting properties on output collections.",
 397                                self.arvrunner.label(self), self.arvrunner.api._rootDesc["revision"], "20220510")
 398
 399         if self.arvrunner.api._rootDesc["revision"] >= "20240502" and self.globpatterns:
 400             output_glob = []
 401             for gb in self.globpatterns:
 402                 gb = self.builder.do_eval(gb)
 403                 if not gb:
 404                     continue
 405                 for gbeval in aslist(gb):
 406                     if gbeval.startswith(self.outdir+"/"):
 407                         gbeval = gbeval[len(self.outdir)+1:]
 408                     while gbeval.startswith("./"):
 409                         gbeval = gbeval[2:]
 410
 411                     if gbeval in (self.outdir, "", "."):
 412                         output_glob.append("**")
 413                     elif gbeval.endswith("/"):
 414                         output_glob.append(gbeval+"**")
 415                     else:
 416                         output_glob.append(gbeval)
 417                         output_glob.append(gbeval + "/**")
 418
 419             if "**" in output_glob:
 420                 # if it's going to match all, prefer not to provide it
 421                 # at all.
 422                 output_glob.clear()
 423
 424             if output_glob:
 425                 # Tools should either use cwl.output.json or
 426                 # outputBinding globs. However, one CWL conformance
 427                 # test has both, so we need to make sure we collect
 428                 # cwl.output.json in this case. That test uses
 429                 # cwl.output.json return a string, but also uses
 430                 # outputBinding.
 431                 output_glob.append("cwl.output.json")
 432
 433                 # It could happen that a tool creates cwl.output.json,
 434                 # references a file, but also uses a outputBinding
 435                 # glob that doesn't include the file being referenced.
 436                 #
 437                 # In this situation, output_glob will only match the
 438                 # pattern we know about.  If cwl.output.json referred
 439                 # to other files in the output, those would be
 440                 # missing.  We could upload the entire output, but we
 441                 # currently have no way of knowing at this point
 442                 # whether cwl.output.json will be used this way.
 443                 #
 444                 # Because this is a corner case, I'm inclined to leave
 445                 # this as a known issue for now.  No conformance tests
 446                 # do this and I'd even be inclined to have it ruled
 447                 # incompatible in the CWL spec if it did come up.
 448                 # That said, in retrospect it would have been good to
 449                 # require CommandLineTool to declare when it expects
 450                 # cwl.output.json.
 451
 452                 container_request["output_glob"] = output_glob
 453
 454         ram_multiplier = [1]
 455
 456         oom_retry_req, _ = self.get_requirement("http://arvados.org/cwl#OutOfMemoryRetry")
 457         if oom_retry_req:
 458             if oom_retry_req.get('memoryRetryMultiplier'):
 459                 ram_multiplier.append(oom_retry_req.get('memoryRetryMultiplier'))
 460             elif oom_retry_req.get('memoryRetryMultipler'):
 461                 ram_multiplier.append(oom_retry_req.get('memoryRetryMultipler'))
 462             else:
 463                 ram_multiplier.append(2)
 464
 465         if runtimeContext.runnerjob.startswith("arvwf:"):
 466             wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")]
 467             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 468             if container_request["name"] == "main":
 469                 container_request["name"] = wfrecord["name"]
 470             container_request["properties"]["template_uuid"] = wfuuid
 471
 472         if self.attempt_count == 0:
 473             self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 474
 475         try:
 476             ram = runtime_constraints["ram"]
 477
 478             self.uuid = runtimeContext.submit_request_uuid
 479
 480             for i in ram_multiplier:
 481                 runtime_constraints["ram"] = ram * i
 482
 483                 if self.uuid:
 484                     response = self.arvrunner.api.container_requests().update(
 485                         uuid=self.uuid,
 486                         body=container_request,
 487                         **extra_submit_params
 488                     ).execute(num_retries=self.arvrunner.num_retries)
 489                 else:
 490                     response = self.arvrunner.api.container_requests().create(
 491                         body=container_request,
 492                         **extra_submit_params
 493                     ).execute(num_retries=self.arvrunner.num_retries)
 494                     self.uuid = response["uuid"]
 495
 496                 if response["container_uuid"] is not None:
 497                     break
 498
 499             if response["container_uuid"] is None:
 500                 runtime_constraints["ram"] = ram * ram_multiplier[self.attempt_count]
 501
 502             container_request["state"] = "Committed"
 503             response = self.arvrunner.api.container_requests().update(
 504                 uuid=self.uuid,
 505                 body=container_request,
 506                 **extra_submit_params
 507             ).execute(num_retries=self.arvrunner.num_retries)
 508
 509             self.arvrunner.process_submitted(self)
 510             self.attempt_count += 1
 511
 512             if response["state"] == "Final":
 513                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 514             else:
 515                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 516         except Exception as e:
 517             logger.exception("%s error submitting container\n%s", self.arvrunner.label(self), e)
 518             logger.debug("Container request was %s", container_request)
 519             self.output_callback({}, "permanentFail")
 520
 521     def may_resubmit_non_preemptible(self):
 522         if self.job_runtime.enable_resubmit_non_preemptible is False:
 523             # explicitly disabled
 524             return False
 525
 526         spot_instance_retry_req, _ = self.get_requirement("http://arvados.org/cwl#PreemptionBehavior")
 527         if spot_instance_retry_req:
 528             if spot_instance_retry_req["resubmitNonPreemptible"] is False:
 529                 # explicitly disabled by hint
 530                 return False
 531         elif self.job_runtime.enable_resubmit_non_preemptible is None:
 532             # default behavior is we don't retry
 533             return False
 534
 535         # At this point, by process of elimination either
 536         # resubmitNonPreemptible or enable_resubmit_non_preemptible
 537         # must be True, so now check if the container was actually
 538         # preempted.
 539
 540         return True
 541
 542     def spot_instance_retry(self, record, container):
 543         return self.may_resubmit_non_preemptible() and bool(container["runtime_status"].get("preemptionNotice"))
 544
 545     def out_of_memory_retry(self, record, container):
 546         oom_retry_req, _ = self.get_requirement("http://arvados.org/cwl#OutOfMemoryRetry")
 547         if oom_retry_req is None:
 548             return False
 549
 550         # Sometimes it gets killed with no warning
 551         if container["exit_code"] == 137:
 552             return True
 553
 554         logc = arvados.collection.CollectionReader(record["log_uuid"],
 555                                                    api_client=self.arvrunner.api,
 556                                                    keep_client=self.arvrunner.keep_client,
 557                                                    num_retries=self.arvrunner.num_retries)
 558
 559         loglines = [""]
 560         def callback(v1, v2, v3):
 561             loglines[0] = v3
 562
 563         done.logtail(logc, callback, "", maxlen=1000)
 564
 565         # Check allocation failure
 566         oom_matches = oom_retry_req.get('memoryErrorRegex') or r'(bad_alloc|out ?of ?memory|memory ?error|container using over 9.% of memory)'
 567         if re.search(oom_matches, loglines[0], re.IGNORECASE | re.MULTILINE):
 568             return True
 569
 570         return False
 571
 572     def done(self, record):
 573         outputs = {}
 574         retried = False
 575         rcode = None
 576         do_retry = False
 577
 578         try:
 579             container = self.arvrunner.api.containers().get(
 580                 uuid=record["container_uuid"]
 581             ).execute(num_retries=self.arvrunner.num_retries)
 582
 583             if container["state"] == "Complete":
 584                 rcode = container["exit_code"]
 585                 if self.successCodes and rcode in self.successCodes:
 586                     processStatus = "success"
 587                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 588                     processStatus = "temporaryFail"
 589                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 590                     processStatus = "permanentFail"
 591                 elif rcode == 0:
 592                     processStatus = "success"
 593                 else:
 594                     processStatus = "permanentFail"
 595
 596                 if processStatus == "permanentFail" and self.attempt_count == 1 and self.out_of_memory_retry(record, container):
 597                     logger.info("%s Container failed with out of memory error.  Retrying container with more RAM.",
 598                                  self.arvrunner.label(self))
 599                     self.job_runtime = self.job_runtime.copy()
 600                     do_retry = True
 601
 602                 if rcode == 137 and not do_retry:
 603                     logger.warning("%s Container may have been killed for using too much RAM.  Try resubmitting with a higher 'ramMin' or use the arv:OutOfMemoryRetry feature.",
 604                                  self.arvrunner.label(self))
 605             else:
 606                 processStatus = "permanentFail"
 607
 608             if processStatus == "permanentFail" and self.attempt_count == 1 and self.spot_instance_retry(record, container):
 609                 logger.info("%s Container failed because the preemptible instance it was running on was reclaimed.  Retrying container on a non-preemptible instance.")
 610                 self.job_runtime = self.job_runtime.copy()
 611                 self.job_runtime.enable_preemptible = False
 612                 do_retry = True
 613
 614             if do_retry:
 615                 # Add a property indicating that this container was resubmitted.
 616                 updateproperties = record["properties"].copy()
 617                 olduuid = self.uuid
 618                 self.job_runtime.submit_request_uuid = None
 619                 self.uuid = None
 620                 self.run(None)
 621                 # this flag suppresses calling the output callback, we only want to set this
 622                 # when we're sure that the resubmission has happened without issue.
 623                 retried = True
 624                 # Add a property to the old container request indicating it
 625                 # was retried
 626                 updateproperties["arv:failed_container_resubmitted"] = self.uuid
 627                 self.arvrunner.api.container_requests().update(uuid=olduuid,
 628                                                                body={"properties": updateproperties}).execute()
 629                 return
 630
 631             logc = None
 632             if record["log_uuid"]:
 633                 logc = arvados.collection.Collection(record["log_uuid"],
 634                                                      api_client=self.arvrunner.api,
 635                                                      keep_client=self.arvrunner.keep_client,
 636                                                      num_retries=self.arvrunner.num_retries)
 637
 638             if processStatus == "permanentFail" and logc is not None:
 639                 label = self.arvrunner.label(self)
 640                 done.logtail(
 641                     logc, logger.error,
 642                     "%s (%s) error log:" % (label, record["uuid"]), maxlen=40, include_crunchrun=(rcode is None or rcode > 127))
 643
 644             if record["output_uuid"]:
 645                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 646                     # Compute the trash time to avoid requesting the collection record.
 647                     trash_at = ciso8601.parse_datetime_as_naive(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 648                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 649                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 650                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 651                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 652                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 653                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 654
 655             if container["output"]:
 656                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 657
 658             properties = record["properties"].copy()
 659             properties["cwl_output"] = outputs
 660             self.arvrunner.api.container_requests().update(
 661                 uuid=self.uuid,
 662                 body={"container_request": {"properties": properties}}
 663             ).execute(num_retries=self.arvrunner.num_retries)
 664
 665             if logc is not None and self.job_runtime.enable_usage_report is not False:
 666                 try:
 667                     summarizer = crunchstat_summary.summarizer.ContainerRequestSummarizer(
 668                         record,
 669                         collection_object=logc,
 670                         label=self.name,
 671                         arv=self.arvrunner.api)
 672                     summarizer.run()
 673                     with logc.open("usage_report.html", "wt") as mr:
 674                         mr.write(summarizer.html_report())
 675                     logc.save()
 676
 677                     # Post warnings about nodes that are under-utilized.
 678                     for rc in summarizer._recommend_gen(lambda x: x):
 679                         self.job_runtime.usage_report_notes.append(rc)
 680
 681                 except Exception as e:
 682                     logger.warning("%s unable to generate resource usage report",
 683                                  self.arvrunner.label(self),
 684                                  exc_info=(e if self.arvrunner.debug else False))
 685
 686         except WorkflowException as e:
 687             # Only include a stack trace if in debug mode.
 688             # A stack trace may obfuscate more useful output about the workflow.
 689             logger.error("%s unable to collect output from %s:\n%s",
 690                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 691             processStatus = "permanentFail"
 692         except Exception:
 693             logger.exception("%s while getting output object:", self.arvrunner.label(self))
 694             processStatus = "permanentFail"
 695         finally:
 696             if not retried:
 697                 self.output_callback(outputs, processStatus)
 698
 699
 700 class RunnerContainer(Runner):
 701     """Submit and manage a container that runs arvados-cwl-runner."""
 702
 703     def arvados_job_spec(self, runtimeContext, git_info):
 704         """Create an Arvados container request for this workflow.
 705
 706         The returned dict can be used to create a container passed as
 707         the +body+ argument to container_requests().create().
 708         """
 709
 710         adjustDirObjs(self.job_order, trim_listing)
 711         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 712         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 713
 714         secret_mounts = {}
 715         for param in sorted(self.job_order.keys()):
 716             if self.secret_store.has_secret(self.job_order[param]):
 717                 mnt = "/secrets/s%d" % len(secret_mounts)
 718                 secret_mounts[mnt] = {
 719                     "kind": "text",
 720                     "content": self.secret_store.retrieve(self.job_order[param])
 721                 }
 722                 self.job_order[param] = {"$include": mnt}
 723
 724         container_image = arvados_jobs_image(self.arvrunner, self.jobs_image, runtimeContext)
 725
 726         workflow_runner_req, _ = self.embedded_tool.get_requirement("http://arvados.org/cwl#WorkflowRunnerResources")
 727         if workflow_runner_req and workflow_runner_req.get("acrContainerImage"):
 728             container_image = workflow_runner_req.get("acrContainerImage")
 729
 730         container_req = {
 731             "name": self.name,
 732             "output_path": "/var/spool/cwl",
 733             "cwd": "/var/spool/cwl",
 734             "priority": self.priority,
 735             "state": "Committed",
 736             "container_image": container_image,
 737             "mounts": {
 738                 "/var/lib/cwl/cwl.input.json": {
 739                     "kind": "json",
 740                     "content": self.job_order
 741                 },
 742                 "stdout": {
 743                     "kind": "file",
 744                     "path": "/var/spool/cwl/cwl.output.json"
 745                 },
 746                 "/var/spool/cwl": {
 747                     "kind": "collection",
 748                     "writable": True
 749                 }
 750             },
 751             "secret_mounts": secret_mounts,
 752             "runtime_constraints": {
 753                 "vcpus": math.ceil(self.submit_runner_cores),
 754                 "ram": 1024*1024 * (math.ceil(self.submit_runner_ram) + math.ceil(self.collection_cache_size)),
 755                 "API": True
 756             },
 757             "use_existing": self.reuse_runner,
 758             "properties": {}
 759         }
 760
 761         if self.embedded_tool.tool.get("id", "").startswith("keep:"):
 762             sp = self.embedded_tool.tool["id"].split('/')
 763             workflowcollection = sp[0][5:]
 764             workflowname = "/".join(sp[1:])
 765             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 766             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 767                 "kind": "collection",
 768                 "portable_data_hash": "%s" % workflowcollection
 769             }
 770         elif self.embedded_tool.tool.get("id", "").startswith("arvwf:"):
 771             uuid, frg = urllib.parse.urldefrag(self.embedded_tool.tool["id"])
 772             workflowpath = "/var/lib/cwl/workflow.json#" + frg
 773             packedtxt = self.loadingContext.loader.fetch_text(uuid)
 774             yaml = ruamel.yaml.YAML(typ='safe', pure=True)
 775             packed = yaml.load(packedtxt)
 776             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 777                 "kind": "json",
 778                 "content": packed
 779             }
 780             container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33]
 781         elif self.embedded_tool.tool.get("id", "").startswith("file:"):
 782             raise WorkflowException("Tool id '%s' is a local file but expected keep: or arvwf:" % self.embedded_tool.tool.get("id"))
 783         else:
 784             main = self.loadingContext.loader.idx["_:main"]
 785             if main.get("id") == "_:main":
 786                 del main["id"]
 787             workflowpath = "/var/lib/cwl/workflow.json#main"
 788             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 789                 "kind": "json",
 790                 "content": main
 791             }
 792
 793         container_req["properties"].update({k.replace("http://arvados.org/cwl#", "arv:"): v for k, v in git_info.items()})
 794
 795         properties_req, _ = self.embedded_tool.get_requirement("http://arvados.org/cwl#ProcessProperties")
 796         if properties_req:
 797             builder = make_builder(self.job_order, self.embedded_tool.hints, self.embedded_tool.requirements, runtimeContext, self.embedded_tool.metadata)
 798             for pr in properties_req["processProperties"]:
 799                 container_req["properties"][pr["propertyName"]] = builder.do_eval(pr["propertyValue"])
 800
 801         # --local means execute the workflow instead of submitting a container request
 802         # --api=containers means use the containers API
 803         # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
 804         # --disable-validate because we already validated so don't need to do it again
 805         # --eval-timeout is the timeout for javascript invocation
 806         # --parallel-task-count is the number of threads to use for job submission
 807         # --enable/disable-reuse sets desired job reuse
 808         # --collection-cache-size sets aside memory to store collections
 809         command = ["arvados-cwl-runner",
 810                    "--local",
 811                    "--api=containers",
 812                    "--no-log-timestamps",
 813                    "--disable-validate",
 814                    "--disable-color",
 815                    "--eval-timeout=%s" % self.arvrunner.eval_timeout,
 816                    "--thread-count=%s" % self.arvrunner.thread_count,
 817                    "--enable-reuse" if self.enable_reuse else "--disable-reuse",
 818                    "--collection-cache-size=%s" % self.collection_cache_size]
 819
 820         if self.output_name:
 821             command.append("--output-name=" + self.output_name)
 822             container_req["output_name"] = self.output_name
 823
 824         if self.output_tags:
 825             command.append("--output-tags=" + self.output_tags)
 826
 827         if runtimeContext.debug:
 828             command.append("--debug")
 829
 830         if runtimeContext.storage_classes != "default" and runtimeContext.storage_classes:
 831             command.append("--storage-classes=" + runtimeContext.storage_classes)
 832
 833         if runtimeContext.intermediate_storage_classes != "default" and runtimeContext.intermediate_storage_classes:
 834             command.append("--intermediate-storage-classes=" + runtimeContext.intermediate_storage_classes)
 835
 836         if runtimeContext.on_error:
 837             command.append("--on-error=" + self.on_error)
 838
 839         if runtimeContext.intermediate_output_ttl:
 840             command.append("--intermediate-output-ttl=%d" % runtimeContext.intermediate_output_ttl)
 841
 842         if runtimeContext.trash_intermediate:
 843             command.append("--trash-intermediate")
 844
 845         if runtimeContext.project_uuid:
 846             command.append("--project-uuid="+runtimeContext.project_uuid)
 847
 848         if self.enable_dev:
 849             command.append("--enable-dev")
 850
 851         if runtimeContext.enable_preemptible is True:
 852             command.append("--enable-preemptible")
 853
 854         if runtimeContext.enable_preemptible is False:
 855             command.append("--disable-preemptible")
 856
 857         if runtimeContext.varying_url_params:
 858             command.append("--varying-url-params="+runtimeContext.varying_url_params)
 859
 860         if runtimeContext.prefer_cached_downloads:
 861             command.append("--prefer-cached-downloads")
 862
 863         if runtimeContext.enable_usage_report is True:
 864             command.append("--enable-usage-report")
 865
 866         if runtimeContext.enable_usage_report is False:
 867             command.append("--disable-usage-report")
 868
 869         if self.fast_parser:
 870             command.append("--fast-parser")
 871
 872         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 873
 874         container_req["command"] = command
 875
 876         return container_req
 877
 878
 879     def run(self, runtimeContext):
 880         runtimeContext.keepprefix = "keep:"
 881         job_spec = self.arvados_job_spec(runtimeContext, self.git_info)
 882         if runtimeContext.project_uuid:
 883             job_spec["owner_uuid"] = runtimeContext.project_uuid
 884
 885         extra_submit_params = {}
 886         if runtimeContext.submit_runner_cluster:
 887             extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 888
 889         if runtimeContext.submit_request_uuid:
 890             if "cluster_id" in extra_submit_params:
 891                 # Doesn't make sense for "update" and actually fails
 892                 del extra_submit_params["cluster_id"]
 893             response = self.arvrunner.api.container_requests().update(
 894                 uuid=runtimeContext.submit_request_uuid,
 895                 body=job_spec,
 896                 **extra_submit_params
 897             ).execute(num_retries=self.arvrunner.num_retries)
 898         else:
 899             response = self.arvrunner.api.container_requests().create(
 900                 body=job_spec,
 901                 **extra_submit_params
 902             ).execute(num_retries=self.arvrunner.num_retries)
 903
 904         self.uuid = response["uuid"]
 905         self.arvrunner.process_submitted(self)
 906
 907         logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"])
 908
 909         workbench2 = self.arvrunner.api.config()["Services"]["Workbench2"]["ExternalURL"]
 910         if workbench2:
 911             url = "{}processes/{}".format(workbench2, response["uuid"])
 912             logger.info("Monitor workflow progress at %s", url)
 913
 914
 915     def done(self, record):
 916         try:
 917             container = self.arvrunner.api.containers().get(
 918                 uuid=record["container_uuid"]
 919             ).execute(num_retries=self.arvrunner.num_retries)
 920             container["log"] = record["log_uuid"]
 921         except Exception:
 922             logger.exception("%s while getting runner container", self.arvrunner.label(self))
 923             self.arvrunner.output_callback({}, "permanentFail")
 924         else:
 925             super(RunnerContainer, self).done(container)