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
   9 import time
  10 import datetime
  11 import ciso8601
  12 import uuid
  13
  14 import ruamel.yaml as yaml
  15
  16 from cwltool.errors import WorkflowException
  17 from cwltool.process import get_feature, UnsupportedRequirement, shortname
  18 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  19 from cwltool.utils import aslist
  20
  21 import arvados.collection
  22
  23 from .arvdocker import arv_docker_get_image
  24 from . import done
  25 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields
  26 from .fsaccess import CollectionFetcher
  27 from .pathmapper import NoFollowPathMapper, trim_listing
  28 from .perf import Perf
  29
  30 logger = logging.getLogger('arvados.cwl-runner')
  31 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  32
  33 class ArvadosContainer(object):
  34     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  35
  36     def __init__(self, runner):
  37         self.arvrunner = runner
  38         self.running = False
  39         self.uuid = None
  40
  41     def update_pipeline_component(self, r):
  42         pass
  43
  44     def run(self, dry_run=False, pull_image=True, **kwargs):
  45         # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
  46         # which calls makeJobRunner() to get a new ArvadosContainer
  47         # object.  The fields that define execution such as
  48         # command_line, environment, etc are set on the
  49         # ArvadosContainer object by CommandLineTool.job() before
  50         # run() is called.
  51
  52         container_request = {
  53             "command": self.command_line,
  54             "name": self.name,
  55             "output_path": self.outdir,
  56             "cwd": self.outdir,
  57             "priority": kwargs.get("priority"),
  58             "state": "Committed",
  59             "properties": {},
  60         }
  61         runtime_constraints = {}
  62
  63         if self.arvrunner.project_uuid:
  64             container_request["owner_uuid"] = self.arvrunner.project_uuid
  65
  66         if self.arvrunner.secret_store.has_secret(self.command_line):
  67             raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
  68
  69         if self.arvrunner.secret_store.has_secret(self.environment):
  70             raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
  71
  72         resources = self.builder.resources
  73         if resources is not None:
  74             runtime_constraints["vcpus"] = resources.get("cores", 1)
  75             runtime_constraints["ram"] = resources.get("ram") * 2**20
  76
  77         mounts = {
  78             self.outdir: {
  79                 "kind": "tmp",
  80                 "capacity": resources.get("outdirSize", 0) * 2**20
  81             },
  82             self.tmpdir: {
  83                 "kind": "tmp",
  84                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  85             }
  86         }
  87         secret_mounts = {}
  88         scheduling_parameters = {}
  89
  90         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  91         rf.sort(key=lambda k: k.resolved)
  92         prevdir = None
  93         for resolved, target, tp, stg in rf:
  94             if not stg:
  95                 continue
  96             if prevdir and target.startswith(prevdir):
  97                 continue
  98             if tp == "Directory":
  99                 targetdir = target
 100             else:
 101                 targetdir = os.path.dirname(target)
 102             sp = resolved.split("/", 1)
 103             pdh = sp[0][5:]   # remove "keep:"
 104             mounts[targetdir] = {
 105                 "kind": "collection",
 106                 "portable_data_hash": pdh
 107             }
 108             if len(sp) == 2:
 109                 if tp == "Directory":
 110                     path = sp[1]
 111                 else:
 112                     path = os.path.dirname(sp[1])
 113                 if path and path != "/":
 114                     mounts[targetdir]["path"] = path
 115             prevdir = targetdir + "/"
 116
 117         with Perf(metrics, "generatefiles %s" % self.name):
 118             if self.generatefiles["listing"]:
 119                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
 120                                                     keep_client=self.arvrunner.keep_client,
 121                                                     num_retries=self.arvrunner.num_retries)
 122                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
 123                                                     separateDirs=False)
 124
 125                 sorteditems = sorted(generatemapper.items(), None, key=lambda n: n[1].target)
 126
 127                 logger.debug("generatemapper is %s", sorteditems)
 128
 129                 with Perf(metrics, "createfiles %s" % self.name):
 130                     for f, p in sorteditems:
 131                         if not p.target:
 132                             pass
 133                         elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 134                             if p.resolved.startswith("_:"):
 135                                 vwd.mkdirs(p.target)
 136                             else:
 137                                 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 138                                 vwd.copy(path, p.target, source_collection=source)
 139                         elif p.type == "CreateFile":
 140                             if self.arvrunner.secret_store.has_secret(p.resolved):
 141                                 secret_mounts["%s/%s" % (self.outdir, p.target)] = {
 142                                     "kind": "text",
 143                                     "content": self.arvrunner.secret_store.retrieve(p.resolved)
 144                                 }
 145                             else:
 146                                 with vwd.open(p.target, "w") as n:
 147                                     n.write(p.resolved.encode("utf-8"))
 148
 149                 def keepemptydirs(p):
 150                     if isinstance(p, arvados.collection.RichCollectionBase):
 151                         if len(p) == 0:
 152                             p.open(".keep", "w").close()
 153                         else:
 154                             for c in p:
 155                                 keepemptydirs(p[c])
 156
 157                 keepemptydirs(vwd)
 158
 159                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 160                     vwd.save_new()
 161
 162                 prev = None
 163                 for f, p in sorteditems:
 164                     if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
 165                         (prev is not None and p.target.startswith(prev))):
 166                         continue
 167                     mountpoint = "%s/%s" % (self.outdir, p.target)
 168                     mounts[mountpoint] = {"kind": "collection",
 169                                           "portable_data_hash": vwd.portable_data_hash(),
 170                                           "path": p.target}
 171                     if p.type.startswith("Writable"):
 172                         mounts[mountpoint]["writable"] = True
 173                     prev = p.target + "/"
 174
 175         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 176         if self.environment:
 177             container_request["environment"].update(self.environment)
 178
 179         if self.stdin:
 180             sp = self.stdin[6:].split("/", 1)
 181             mounts["stdin"] = {"kind": "collection",
 182                                 "portable_data_hash": sp[0],
 183                                 "path": sp[1]}
 184
 185         if self.stderr:
 186             mounts["stderr"] = {"kind": "file",
 187                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 188
 189         if self.stdout:
 190             mounts["stdout"] = {"kind": "file",
 191                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 192
 193         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 194         if not docker_req:
 195             docker_req = {"dockerImageId": "arvados/jobs"}
 196
 197         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 198                                                                      docker_req,
 199                                                                      pull_image,
 200                                                                      self.arvrunner.project_uuid)
 201
 202         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 203         if api_req:
 204             runtime_constraints["API"] = True
 205
 206         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 207         if runtime_req:
 208             if "keep_cache" in runtime_req:
 209                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 210             if "outputDirType" in runtime_req:
 211                 if runtime_req["outputDirType"] == "local_output_dir":
 212                     # Currently the default behavior.
 213                     pass
 214                 elif runtime_req["outputDirType"] == "keep_output_dir":
 215                     mounts[self.outdir]= {
 216                         "kind": "collection",
 217                         "writable": True
 218                     }
 219
 220         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 221         if partition_req:
 222             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 223
 224         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 225         if intermediate_output_req:
 226             self.output_ttl = intermediate_output_req["outputTTL"]
 227         else:
 228             self.output_ttl = self.arvrunner.intermediate_output_ttl
 229
 230         if self.output_ttl < 0:
 231             raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 232
 233         container_request["output_ttl"] = self.output_ttl
 234         container_request["mounts"] = mounts
 235         container_request["secret_mounts"] = secret_mounts
 236         container_request["runtime_constraints"] = runtime_constraints
 237         container_request["scheduling_parameters"] = scheduling_parameters
 238
 239         enable_reuse = kwargs.get("enable_reuse", True)
 240         if enable_reuse:
 241             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 242             if reuse_req:
 243                 enable_reuse = reuse_req["enableReuse"]
 244         container_request["use_existing"] = enable_reuse
 245
 246         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 247             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 248             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 249             if container_request["name"] == "main":
 250                 container_request["name"] = wfrecord["name"]
 251             container_request["properties"]["template_uuid"] = wfuuid
 252
 253         self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 254
 255         try:
 256             if kwargs.get("submit_request_uuid"):
 257                 response = self.arvrunner.api.container_requests().update(
 258                     uuid=kwargs["submit_request_uuid"],
 259                     body=container_request
 260                 ).execute(num_retries=self.arvrunner.num_retries)
 261             else:
 262                 response = self.arvrunner.api.container_requests().create(
 263                     body=container_request
 264                 ).execute(num_retries=self.arvrunner.num_retries)
 265
 266             self.uuid = response["uuid"]
 267             self.arvrunner.process_submitted(self)
 268
 269             if response["state"] == "Final":
 270                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 271             else:
 272                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 273         except Exception as e:
 274             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 275             self.output_callback({}, "permanentFail")
 276
 277     def done(self, record):
 278         outputs = {}
 279         try:
 280             container = self.arvrunner.api.containers().get(
 281                 uuid=record["container_uuid"]
 282             ).execute(num_retries=self.arvrunner.num_retries)
 283             if container["state"] == "Complete":
 284                 rcode = container["exit_code"]
 285                 if self.successCodes and rcode in self.successCodes:
 286                     processStatus = "success"
 287                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 288                     processStatus = "temporaryFail"
 289                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 290                     processStatus = "permanentFail"
 291                 elif rcode == 0:
 292                     processStatus = "success"
 293                 else:
 294                     processStatus = "permanentFail"
 295             else:
 296                 processStatus = "permanentFail"
 297
 298             if processStatus == "permanentFail":
 299                 logc = arvados.collection.CollectionReader(container["log"],
 300                                                            api_client=self.arvrunner.api,
 301                                                            keep_client=self.arvrunner.keep_client,
 302                                                            num_retries=self.arvrunner.num_retries)
 303                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 304
 305             if record["output_uuid"]:
 306                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 307                     # Compute the trash time to avoid requesting the collection record.
 308                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 309                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 310                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 311                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 312                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 313                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 314                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 315
 316             if container["output"]:
 317                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 318         except WorkflowException as e:
 319             logger.error("%s unable to collect output from %s:\n%s",
 320                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 321             processStatus = "permanentFail"
 322         except Exception as e:
 323             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 324             processStatus = "permanentFail"
 325         finally:
 326             self.output_callback(outputs, processStatus)
 327
 328
 329 class RunnerContainer(Runner):
 330     """Submit and manage a container that runs arvados-cwl-runner."""
 331
 332     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 333         """Create an Arvados container request for this workflow.
 334
 335         The returned dict can be used to create a container passed as
 336         the +body+ argument to container_requests().create().
 337         """
 338
 339         adjustDirObjs(self.job_order, trim_listing)
 340         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 341         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 342
 343         secret_mounts = {}
 344         for param in sorted(self.job_order.keys()):
 345             if self.secret_store.has_secret(self.job_order[param]):
 346                 mnt = "/secrets/s%d" % len(secret_mounts)
 347                 secret_mounts[mnt] = {
 348                     "kind": "text",
 349                     "content": self.secret_store.retrieve(self.job_order[param])
 350                 }
 351                 self.job_order[param] = {"$include": mnt}
 352
 353         container_req = {
 354             "name": self.name,
 355             "output_path": "/var/spool/cwl",
 356             "cwd": "/var/spool/cwl",
 357             "priority": self.priority,
 358             "state": "Committed",
 359             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 360             "mounts": {
 361                 "/var/lib/cwl/cwl.input.json": {
 362                     "kind": "json",
 363                     "content": self.job_order
 364                 },
 365                 "stdout": {
 366                     "kind": "file",
 367                     "path": "/var/spool/cwl/cwl.output.json"
 368                 },
 369                 "/var/spool/cwl": {
 370                     "kind": "collection",
 371                     "writable": True
 372                 }
 373             },
 374             "secret_mounts": secret_mounts,
 375             "runtime_constraints": {
 376                 "vcpus": 1,
 377                 "ram": 1024*1024 * self.submit_runner_ram,
 378                 "API": True
 379             },
 380             "use_existing": self.enable_reuse,
 381             "properties": {}
 382         }
 383
 384         if self.tool.tool.get("id", "").startswith("keep:"):
 385             sp = self.tool.tool["id"].split('/')
 386             workflowcollection = sp[0][5:]
 387             workflowname = "/".join(sp[1:])
 388             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 389             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 390                 "kind": "collection",
 391                 "portable_data_hash": "%s" % workflowcollection
 392             }
 393         else:
 394             packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
 395             workflowpath = "/var/lib/cwl/workflow.json#main"
 396             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 397                 "kind": "json",
 398                 "content": packed
 399             }
 400             if self.tool.tool.get("id", "").startswith("arvwf:"):
 401                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 402
 403
 404         # --local means execute the workflow instead of submitting a container request
 405         # --api=containers means use the containers API
 406         # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
 407         # --disable-validate because we already validated so don't need to do it again
 408         # --eval-timeout is the timeout for javascript invocation
 409         # --parallel-task-count is the number of threads to use for job submission
 410         # --enable/disable-reuse sets desired job reuse
 411         command = ["arvados-cwl-runner",
 412                    "--local",
 413                    "--api=containers",
 414                    "--no-log-timestamps",
 415                    "--disable-validate",
 416                    "--eval-timeout=%s" % self.arvrunner.eval_timeout,
 417                    "--thread-count=%s" % self.arvrunner.thread_count,
 418                    "--enable-reuse" if self.enable_reuse else "--disable-reuse"]
 419
 420         if self.output_name:
 421             command.append("--output-name=" + self.output_name)
 422             container_req["output_name"] = self.output_name
 423
 424         if self.output_tags:
 425             command.append("--output-tags=" + self.output_tags)
 426
 427         if kwargs.get("debug"):
 428             command.append("--debug")
 429
 430         if kwargs.get("storage_classes") and kwargs.get("storage_classes") != self.default_storage_classes:
 431             command.append("--storage-classes=" + kwargs.get("storage_classes"))
 432
 433         if self.on_error:
 434             command.append("--on-error=" + self.on_error)
 435
 436         if self.intermediate_output_ttl:
 437             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 438
 439         if self.arvrunner.trash_intermediate:
 440             command.append("--trash-intermediate")
 441
 442         if self.arvrunner.project_uuid:
 443             command.append("--project-uuid="+self.arvrunner.project_uuid)
 444
 445         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 446
 447         container_req["command"] = command
 448
 449         return container_req
 450
 451
 452     def run(self, **kwargs):
 453         kwargs["keepprefix"] = "keep:"
 454         job_spec = self.arvados_job_spec(**kwargs)
 455         if self.arvrunner.project_uuid:
 456             job_spec["owner_uuid"] = self.arvrunner.project_uuid
 457
 458         if kwargs.get("submit_request_uuid"):
 459             response = self.arvrunner.api.container_requests().update(
 460                 uuid=kwargs["submit_request_uuid"],
 461                 body=job_spec
 462             ).execute(num_retries=self.arvrunner.num_retries)
 463         else:
 464             response = self.arvrunner.api.container_requests().create(
 465                 body=job_spec
 466             ).execute(num_retries=self.arvrunner.num_retries)
 467
 468         self.uuid = response["uuid"]
 469         self.arvrunner.process_submitted(self)
 470
 471         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 472
 473     def done(self, record):
 474         try:
 475             container = self.arvrunner.api.containers().get(
 476                 uuid=record["container_uuid"]
 477             ).execute(num_retries=self.arvrunner.num_retries)
 478         except Exception as e:
 479             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 480             self.arvrunner.output_callback({}, "permanentFail")
 481         else:
 482             super(RunnerContainer, self).done(container)