sdk/cwl/arvados_cwl/arvcontainer.py

   1 from future import standard_library
   2 standard_library.install_aliases()
   3 from builtins import str
   4 # Copyright (C) The Arvados Authors. All rights reserved.
   5 #
   6 # SPDX-License-Identifier: Apache-2.0
   7
   8 import logging
   9 import json
  10 import os
  11 import urllib.request, urllib.parse, urllib.error
  12 import time
  13 import datetime
  14 import ciso8601
  15 import uuid
  16 import math
  17
  18 import arvados_cwl.util
  19 import ruamel.yaml as yaml
  20
  21 from cwltool.errors import WorkflowException
  22 from cwltool.process import UnsupportedRequirement, shortname
  23 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  24 from cwltool.utils import aslist
  25 from cwltool.job import JobBase
  26
  27 import arvados.collection
  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
  32 from .fsaccess import CollectionFetcher
  33 from .pathmapper import NoFollowPathMapper, trim_listing
  34 from .perf import Perf
  35
  36 logger = logging.getLogger('arvados.cwl-runner')
  37 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  38
  39 class ArvadosContainer(JobBase):
  40     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  41
  42     def __init__(self, runner, job_runtime,
  43                  builder,   # type: Builder
  44                  joborder,  # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
  45                  make_path_mapper,  # type: Callable[..., PathMapper]
  46                  requirements,      # type: List[Dict[Text, Text]]
  47                  hints,     # type: List[Dict[Text, Text]]
  48                  name       # type: Text
  49     ):
  50         super(ArvadosContainer, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
  51         self.arvrunner = runner
  52         self.job_runtime = job_runtime
  53         self.running = False
  54         self.uuid = None
  55
  56     def update_pipeline_component(self, r):
  57         pass
  58
  59     def run(self, runtimeContext):
  60         # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
  61         # which calls makeJobRunner() to get a new ArvadosContainer
  62         # object.  The fields that define execution such as
  63         # command_line, environment, etc are set on the
  64         # ArvadosContainer object by CommandLineTool.job() before
  65         # run() is called.
  66
  67         runtimeContext = self.job_runtime
  68
  69         container_request = {
  70             "command": self.command_line,
  71             "name": self.name,
  72             "output_path": self.outdir,
  73             "cwd": self.outdir,
  74             "priority": runtimeContext.priority,
  75             "state": "Committed",
  76             "properties": {},
  77         }
  78         runtime_constraints = {}
  79
  80         if runtimeContext.project_uuid:
  81             container_request["owner_uuid"] = runtimeContext.project_uuid
  82
  83         if self.arvrunner.secret_store.has_secret(self.command_line):
  84             raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
  85
  86         if self.arvrunner.secret_store.has_secret(self.environment):
  87             raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
  88
  89         resources = self.builder.resources
  90         if resources is not None:
  91             runtime_constraints["vcpus"] = math.ceil(resources.get("cores", 1))
  92             runtime_constraints["ram"] = math.ceil(resources.get("ram") * 2**20)
  93
  94         mounts = {
  95             self.outdir: {
  96                 "kind": "tmp",
  97                 "capacity": math.ceil(resources.get("outdirSize", 0) * 2**20)
  98             },
  99             self.tmpdir: {
 100                 "kind": "tmp",
 101                 "capacity": math.ceil(resources.get("tmpdirSize", 0) * 2**20)
 102             }
 103         }
 104         secret_mounts = {}
 105         scheduling_parameters = {}
 106
 107         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
 108         rf.sort(key=lambda k: k.resolved)
 109         prevdir = None
 110         for resolved, target, tp, stg in rf:
 111             if not stg:
 112                 continue
 113             if prevdir and target.startswith(prevdir):
 114                 continue
 115             if tp == "Directory":
 116                 targetdir = target
 117             else:
 118                 targetdir = os.path.dirname(target)
 119             sp = resolved.split("/", 1)
 120             pdh = sp[0][5:]   # remove "keep:"
 121             mounts[targetdir] = {
 122                 "kind": "collection",
 123                 "portable_data_hash": pdh
 124             }
 125             if len(sp) == 2:
 126                 if tp == "Directory":
 127                     path = sp[1]
 128                 else:
 129                     path = os.path.dirname(sp[1])
 130                 if path and path != "/":
 131                     mounts[targetdir]["path"] = path
 132             prevdir = targetdir + "/"
 133
 134         with Perf(metrics, "generatefiles %s" % self.name):
 135             if self.generatefiles["listing"]:
 136                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
 137                                                     keep_client=self.arvrunner.keep_client,
 138                                                     num_retries=self.arvrunner.num_retries)
 139                 generatemapper = NoFollowPathMapper(self.generatefiles["listing"], "", "",
 140                                                     separateDirs=False)
 141
 142                 sorteditems = sorted(list(generatemapper.items()), None, key=lambda n: n[1].target)
 143
 144                 logger.debug("generatemapper is %s", sorteditems)
 145
 146                 with Perf(metrics, "createfiles %s" % self.name):
 147                     for f, p in sorteditems:
 148                         if not p.target:
 149                             pass
 150                         elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 151                             if p.resolved.startswith("_:"):
 152                                 vwd.mkdirs(p.target)
 153                             else:
 154                                 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 155                                 vwd.copy(path, p.target, source_collection=source)
 156                         elif p.type == "CreateFile":
 157                             if self.arvrunner.secret_store.has_secret(p.resolved):
 158                                 secret_mounts["%s/%s" % (self.outdir, p.target)] = {
 159                                     "kind": "text",
 160                                     "content": self.arvrunner.secret_store.retrieve(p.resolved)
 161                                 }
 162                             else:
 163                                 with vwd.open(p.target, "w") as n:
 164                                     n.write(p.resolved.encode("utf-8"))
 165
 166                 def keepemptydirs(p):
 167                     if isinstance(p, arvados.collection.RichCollectionBase):
 168                         if len(p) == 0:
 169                             p.open(".keep", "w").close()
 170                         else:
 171                             for c in p:
 172                                 keepemptydirs(p[c])
 173
 174                 keepemptydirs(vwd)
 175
 176                 if not runtimeContext.current_container:
 177                     runtimeContext.current_container = arvados_cwl.util.get_current_container(self.arvrunner.api, self.arvrunner.num_retries, logger)
 178                 info = arvados_cwl.util.get_intermediate_collection_info(self.name, runtimeContext.current_container, runtimeContext.intermediate_output_ttl)
 179                 vwd.save_new(name=info["name"],
 180                              owner_uuid=runtimeContext.project_uuid,
 181                              ensure_unique_name=True,
 182                              trash_at=info["trash_at"],
 183                              properties=info["properties"])
 184
 185                 prev = None
 186                 for f, p in sorteditems:
 187                     if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
 188                         (prev is not None and p.target.startswith(prev))):
 189                         continue
 190                     mountpoint = "%s/%s" % (self.outdir, p.target)
 191                     mounts[mountpoint] = {"kind": "collection",
 192                                           "portable_data_hash": vwd.portable_data_hash(),
 193                                           "path": p.target}
 194                     if p.type.startswith("Writable"):
 195                         mounts[mountpoint]["writable"] = True
 196                     prev = p.target + "/"
 197
 198         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 199         if self.environment:
 200             container_request["environment"].update(self.environment)
 201
 202         if self.stdin:
 203             sp = self.stdin[6:].split("/", 1)
 204             mounts["stdin"] = {"kind": "collection",
 205                                 "portable_data_hash": sp[0],
 206                                 "path": sp[1]}
 207
 208         if self.stderr:
 209             mounts["stderr"] = {"kind": "file",
 210                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 211
 212         if self.stdout:
 213             mounts["stdout"] = {"kind": "file",
 214                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 215
 216         (docker_req, docker_is_req) = self.get_requirement("DockerRequirement")
 217         if not docker_req:
 218             docker_req = {"dockerImageId": "arvados/jobs"}
 219
 220         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 221                                                                     docker_req,
 222                                                                     runtimeContext.pull_image,
 223                                                                     runtimeContext.project_uuid)
 224
 225         api_req, _ = self.get_requirement("http://arvados.org/cwl#APIRequirement")
 226         if api_req:
 227             runtime_constraints["API"] = True
 228
 229         runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints")
 230         if runtime_req:
 231             if "keep_cache" in runtime_req:
 232                 runtime_constraints["keep_cache_ram"] = math.ceil(runtime_req["keep_cache"] * 2**20)
 233             if "outputDirType" in runtime_req:
 234                 if runtime_req["outputDirType"] == "local_output_dir":
 235                     # Currently the default behavior.
 236                     pass
 237                 elif runtime_req["outputDirType"] == "keep_output_dir":
 238                     mounts[self.outdir]= {
 239                         "kind": "collection",
 240                         "writable": True
 241                     }
 242
 243         partition_req, _ = self.get_requirement("http://arvados.org/cwl#PartitionRequirement")
 244         if partition_req:
 245             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 246
 247         intermediate_output_req, _ = self.get_requirement("http://arvados.org/cwl#IntermediateOutput")
 248         if intermediate_output_req:
 249             self.output_ttl = intermediate_output_req["outputTTL"]
 250         else:
 251             self.output_ttl = self.arvrunner.intermediate_output_ttl
 252
 253         if self.output_ttl < 0:
 254             raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 255
 256         if self.timelimit is not None:
 257             scheduling_parameters["max_run_time"] = self.timelimit
 258
 259         extra_submit_params = {}
 260         if runtimeContext.submit_runner_cluster:
 261             extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 262
 263         container_request["output_name"] = "Output for step %s" % (self.name)
 264         container_request["output_ttl"] = self.output_ttl
 265         container_request["mounts"] = mounts
 266         container_request["secret_mounts"] = secret_mounts
 267         container_request["runtime_constraints"] = runtime_constraints
 268         container_request["scheduling_parameters"] = scheduling_parameters
 269
 270         enable_reuse = runtimeContext.enable_reuse
 271         if enable_reuse:
 272             reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
 273             if reuse_req:
 274                 enable_reuse = reuse_req["enableReuse"]
 275         container_request["use_existing"] = enable_reuse
 276
 277         if runtimeContext.runnerjob.startswith("arvwf:"):
 278             wfuuid = runtimeContext.runnerjob[6:runtimeContext.runnerjob.index("#")]
 279             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 280             if container_request["name"] == "main":
 281                 container_request["name"] = wfrecord["name"]
 282             container_request["properties"]["template_uuid"] = wfuuid
 283
 284         self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 285
 286         try:
 287             if runtimeContext.submit_request_uuid:
 288                 response = self.arvrunner.api.container_requests().update(
 289                     uuid=runtimeContext.submit_request_uuid,
 290                     body=container_request,
 291                     **extra_submit_params
 292                 ).execute(num_retries=self.arvrunner.num_retries)
 293             else:
 294                 response = self.arvrunner.api.container_requests().create(
 295                     body=container_request,
 296                     **extra_submit_params
 297                 ).execute(num_retries=self.arvrunner.num_retries)
 298
 299             self.uuid = response["uuid"]
 300             self.arvrunner.process_submitted(self)
 301
 302             if response["state"] == "Final":
 303                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 304             else:
 305                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 306         except Exception as e:
 307             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 308             self.output_callback({}, "permanentFail")
 309
 310     def done(self, record):
 311         outputs = {}
 312         try:
 313             container = self.arvrunner.api.containers().get(
 314                 uuid=record["container_uuid"]
 315             ).execute(num_retries=self.arvrunner.num_retries)
 316             if container["state"] == "Complete":
 317                 rcode = container["exit_code"]
 318                 if self.successCodes and rcode in self.successCodes:
 319                     processStatus = "success"
 320                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 321                     processStatus = "temporaryFail"
 322                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 323                     processStatus = "permanentFail"
 324                 elif rcode == 0:
 325                     processStatus = "success"
 326                 else:
 327                     processStatus = "permanentFail"
 328             else:
 329                 processStatus = "permanentFail"
 330
 331             if processStatus == "permanentFail":
 332                 logc = arvados.collection.CollectionReader(container["log"],
 333                                                            api_client=self.arvrunner.api,
 334                                                            keep_client=self.arvrunner.keep_client,
 335                                                            num_retries=self.arvrunner.num_retries)
 336                 label = self.arvrunner.label(self)
 337                 done.logtail(
 338                     logc, logger.error,
 339                     "%s (%s) error log:" % (label, record["uuid"]), maxlen=40)
 340
 341             if record["output_uuid"]:
 342                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 343                     # Compute the trash time to avoid requesting the collection record.
 344                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 345                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 346                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 347                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 348                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 349                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 350                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 351
 352             if container["output"]:
 353                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 354         except WorkflowException as e:
 355             logger.error("%s unable to collect output from %s:\n%s",
 356                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 357             processStatus = "permanentFail"
 358         except Exception as e:
 359             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 360             processStatus = "permanentFail"
 361         finally:
 362             self.output_callback(outputs, processStatus)
 363
 364
 365 class RunnerContainer(Runner):
 366     """Submit and manage a container that runs arvados-cwl-runner."""
 367
 368     def arvados_job_spec(self, runtimeContext):
 369         """Create an Arvados container request for this workflow.
 370
 371         The returned dict can be used to create a container passed as
 372         the +body+ argument to container_requests().create().
 373         """
 374
 375         adjustDirObjs(self.job_order, trim_listing)
 376         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 377         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 378
 379         secret_mounts = {}
 380         for param in sorted(self.job_order.keys()):
 381             if self.secret_store.has_secret(self.job_order[param]):
 382                 mnt = "/secrets/s%d" % len(secret_mounts)
 383                 secret_mounts[mnt] = {
 384                     "kind": "text",
 385                     "content": self.secret_store.retrieve(self.job_order[param])
 386                 }
 387                 self.job_order[param] = {"$include": mnt}
 388
 389         container_req = {
 390             "name": self.name,
 391             "output_path": "/var/spool/cwl",
 392             "cwd": "/var/spool/cwl",
 393             "priority": self.priority,
 394             "state": "Committed",
 395             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 396             "mounts": {
 397                 "/var/lib/cwl/cwl.input.json": {
 398                     "kind": "json",
 399                     "content": self.job_order
 400                 },
 401                 "stdout": {
 402                     "kind": "file",
 403                     "path": "/var/spool/cwl/cwl.output.json"
 404                 },
 405                 "/var/spool/cwl": {
 406                     "kind": "collection",
 407                     "writable": True
 408                 }
 409             },
 410             "secret_mounts": secret_mounts,
 411             "runtime_constraints": {
 412                 "vcpus": math.ceil(self.submit_runner_cores),
 413                 "ram": 1024*1024 * (math.ceil(self.submit_runner_ram) + math.ceil(self.collection_cache_size)),
 414                 "API": True
 415             },
 416             "use_existing": self.enable_reuse,
 417             "properties": {}
 418         }
 419
 420         if self.embedded_tool.tool.get("id", "").startswith("keep:"):
 421             sp = self.embedded_tool.tool["id"].split('/')
 422             workflowcollection = sp[0][5:]
 423             workflowname = "/".join(sp[1:])
 424             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 425             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 426                 "kind": "collection",
 427                 "portable_data_hash": "%s" % workflowcollection
 428             }
 429         else:
 430             packed = packed_workflow(self.arvrunner, self.embedded_tool, self.merged_map)
 431             workflowpath = "/var/lib/cwl/workflow.json#main"
 432             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 433                 "kind": "json",
 434                 "content": packed
 435             }
 436             if self.embedded_tool.tool.get("id", "").startswith("arvwf:"):
 437                 container_req["properties"]["template_uuid"] = self.embedded_tool.tool["id"][6:33]
 438
 439
 440         # --local means execute the workflow instead of submitting a container request
 441         # --api=containers means use the containers API
 442         # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
 443         # --disable-validate because we already validated so don't need to do it again
 444         # --eval-timeout is the timeout for javascript invocation
 445         # --parallel-task-count is the number of threads to use for job submission
 446         # --enable/disable-reuse sets desired job reuse
 447         # --collection-cache-size sets aside memory to store collections
 448         command = ["arvados-cwl-runner",
 449                    "--local",
 450                    "--api=containers",
 451                    "--no-log-timestamps",
 452                    "--disable-validate",
 453                    "--eval-timeout=%s" % self.arvrunner.eval_timeout,
 454                    "--thread-count=%s" % self.arvrunner.thread_count,
 455                    "--enable-reuse" if self.enable_reuse else "--disable-reuse",
 456                    "--collection-cache-size=%s" % self.collection_cache_size]
 457
 458         if self.output_name:
 459             command.append("--output-name=" + self.output_name)
 460             container_req["output_name"] = self.output_name
 461
 462         if self.output_tags:
 463             command.append("--output-tags=" + self.output_tags)
 464
 465         if runtimeContext.debug:
 466             command.append("--debug")
 467
 468         if runtimeContext.storage_classes != "default":
 469             command.append("--storage-classes=" + runtimeContext.storage_classes)
 470
 471         if self.on_error:
 472             command.append("--on-error=" + self.on_error)
 473
 474         if self.intermediate_output_ttl:
 475             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 476
 477         if self.arvrunner.trash_intermediate:
 478             command.append("--trash-intermediate")
 479
 480         if self.arvrunner.project_uuid:
 481             command.append("--project-uuid="+self.arvrunner.project_uuid)
 482
 483         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 484
 485         container_req["command"] = command
 486
 487         return container_req
 488
 489
 490     def run(self, runtimeContext):
 491         runtimeContext.keepprefix = "keep:"
 492         job_spec = self.arvados_job_spec(runtimeContext)
 493         if self.arvrunner.project_uuid:
 494             job_spec["owner_uuid"] = self.arvrunner.project_uuid
 495
 496         extra_submit_params = {}
 497         if runtimeContext.submit_runner_cluster:
 498             extra_submit_params["cluster_id"] = runtimeContext.submit_runner_cluster
 499
 500         if runtimeContext.submit_request_uuid:
 501             response = self.arvrunner.api.container_requests().update(
 502                 uuid=runtimeContext.submit_request_uuid,
 503                 body=job_spec,
 504                 **extra_submit_params
 505             ).execute(num_retries=self.arvrunner.num_retries)
 506         else:
 507             response = self.arvrunner.api.container_requests().create(
 508                 body=job_spec,
 509                 **extra_submit_params
 510             ).execute(num_retries=self.arvrunner.num_retries)
 511
 512         self.uuid = response["uuid"]
 513         self.arvrunner.process_submitted(self)
 514
 515         logger.info("%s submitted container_request %s", self.arvrunner.label(self), response["uuid"])
 516
 517     def done(self, record):
 518         try:
 519             container = self.arvrunner.api.containers().get(
 520                 uuid=record["container_uuid"]
 521             ).execute(num_retries=self.arvrunner.num_retries)
 522         except Exception as e:
 523             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 524             self.arvrunner.output_callback({}, "permanentFail")
 525         else:
 526             super(RunnerContainer, self).done(container)