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 arvados.errors import ApiError
  24 from .arvdocker import arv_docker_get_image
  25 from . import done
  26 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields
  27 from .fsaccess import CollectionFetcher
  28 from .pathmapper import NoFollowPathMapper, trim_listing
  29 from .perf import Perf
  30
  31 logger = logging.getLogger('arvados.cwl-runner')
  32 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  33
  34 class ArvadosContainer(object):
  35     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  36
  37     def __init__(self, runner):
  38         self.arvrunner = runner
  39         self.running = False
  40         self.uuid = None
  41
  42     def update_pipeline_component(self, r):
  43         pass
  44
  45     def run(self, dry_run=False, pull_image=True, **kwargs):
  46         # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
  47         # which calls makeJobRunner() to get a new ArvadosContainer
  48         # object.  The fields that define execution such as
  49         # command_line, environment, etc are set on the
  50         # ArvadosContainer object by CommandLineTool.job() before
  51         # run() is called.
  52
  53         container_request = {
  54             "command": self.command_line,
  55             "name": self.name,
  56             "output_path": self.outdir,
  57             "cwd": self.outdir,
  58             "priority": kwargs.get("priority"),
  59             "state": "Committed",
  60             "properties": {},
  61         }
  62         runtime_constraints = {}
  63
  64         if self.arvrunner.project_uuid:
  65             container_request["owner_uuid"] = self.arvrunner.project_uuid
  66
  67         if self.arvrunner.secret_store.has_secret(self.command_line):
  68             raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
  69
  70         if self.arvrunner.secret_store.has_secret(self.environment):
  71             raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
  72
  73         resources = self.builder.resources
  74         if resources is not None:
  75             runtime_constraints["vcpus"] = resources.get("cores", 1)
  76             runtime_constraints["ram"] = resources.get("ram") * 2**20
  77
  78         mounts = {
  79             self.outdir: {
  80                 "kind": "tmp",
  81                 "capacity": resources.get("outdirSize", 0) * 2**20
  82             },
  83             self.tmpdir: {
  84                 "kind": "tmp",
  85                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  86             }
  87         }
  88         secret_mounts = {}
  89         scheduling_parameters = {}
  90
  91         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  92         rf.sort(key=lambda k: k.resolved)
  93         prevdir = None
  94         for resolved, target, tp, stg in rf:
  95             if not stg:
  96                 continue
  97             if prevdir and target.startswith(prevdir):
  98                 continue
  99             if tp == "Directory":
 100                 targetdir = target
 101             else:
 102                 targetdir = os.path.dirname(target)
 103             sp = resolved.split("/", 1)
 104             pdh = sp[0][5:]   # remove "keep:"
 105             mounts[targetdir] = {
 106                 "kind": "collection",
 107                 "portable_data_hash": pdh
 108             }
 109             if len(sp) == 2:
 110                 if tp == "Directory":
 111                     path = sp[1]
 112                 else:
 113                     path = os.path.dirname(sp[1])
 114                 if path and path != "/":
 115                     mounts[targetdir]["path"] = path
 116             prevdir = targetdir + "/"
 117
 118         with Perf(metrics, "generatefiles %s" % self.name):
 119             if self.generatefiles["listing"]:
 120                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
 121                                                     keep_client=self.arvrunner.keep_client,
 122                                                     num_retries=self.arvrunner.num_retries)
 123                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
 124                                                     separateDirs=False)
 125
 126                 sorteditems = sorted(generatemapper.items(), None, key=lambda n: n[1].target)
 127
 128                 logger.debug("generatemapper is %s", sorteditems)
 129
 130                 with Perf(metrics, "createfiles %s" % self.name):
 131                     for f, p in sorteditems:
 132                         if not p.target:
 133                             pass
 134                         elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 135                             if p.resolved.startswith("_:"):
 136                                 vwd.mkdirs(p.target)
 137                             else:
 138                                 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 139                                 vwd.copy(path, p.target, source_collection=source)
 140                         elif p.type == "CreateFile":
 141                             if self.arvrunner.secret_store.has_secret(p.resolved):
 142                                 secret_mounts["%s/%s" % (self.outdir, p.target)] = {
 143                                     "kind": "text",
 144                                     "content": self.arvrunner.secret_store.retrieve(p.resolved)
 145                                 }
 146                             else:
 147                                 with vwd.open(p.target, "w") as n:
 148                                     n.write(p.resolved.encode("utf-8"))
 149
 150                 def keepemptydirs(p):
 151                     if isinstance(p, arvados.collection.RichCollectionBase):
 152                         if len(p) == 0:
 153                             p.open(".keep", "w").close()
 154                         else:
 155                             for c in p:
 156                                 keepemptydirs(p[c])
 157
 158                 keepemptydirs(vwd)
 159
 160                 trash_time = None
 161                 if self.arvrunner.intermediate_output_ttl > 0:
 162                     trash_time = datetime.datetime.now() + datetime.timedelta(seconds=self.arvrunner.intermediate_output_ttl)
 163
 164                 current_container_uuid = None
 165                 try:
 166                     current_container = self.arvrunner.api.containers().current().execute(num_retries=self.arvrunner.num_retries)
 167                     current_container_uuid = current_container['uuid']
 168                 except ApiError as e:
 169                     # Status code 404 just means we're not running in a container.
 170                     if e.resp.status != 404:
 171                         logger.info("Getting current container: %s", e)
 172                 props = {"type": "Intermediate",
 173                          "container": current_container_uuid}
 174
 175                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 176                     vwd.save_new(name="Intermediate collection",
 177                                  ensure_unique_name=True,
 178                                  trash_at=trash_time,
 179                                  properties=props)
 180
 181                 prev = None
 182                 for f, p in sorteditems:
 183                     if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
 184                         (prev is not None and p.target.startswith(prev))):
 185                         continue
 186                     mountpoint = "%s/%s" % (self.outdir, p.target)
 187                     mounts[mountpoint] = {"kind": "collection",
 188                                           "portable_data_hash": vwd.portable_data_hash(),
 189                                           "path": p.target}
 190                     if p.type.startswith("Writable"):
 191                         mounts[mountpoint]["writable"] = True
 192                     prev = p.target + "/"
 193
 194         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 195         if self.environment:
 196             container_request["environment"].update(self.environment)
 197
 198         if self.stdin:
 199             sp = self.stdin[6:].split("/", 1)
 200             mounts["stdin"] = {"kind": "collection",
 201                                 "portable_data_hash": sp[0],
 202                                 "path": sp[1]}
 203
 204         if self.stderr:
 205             mounts["stderr"] = {"kind": "file",
 206                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 207
 208         if self.stdout:
 209             mounts["stdout"] = {"kind": "file",
 210                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 211
 212         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 213         if not docker_req:
 214             docker_req = {"dockerImageId": "arvados/jobs"}
 215
 216         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 217                                                                      docker_req,
 218                                                                      pull_image,
 219                                                                      self.arvrunner.project_uuid)
 220
 221         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 222         if api_req:
 223             runtime_constraints["API"] = True
 224
 225         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 226         if runtime_req:
 227             if "keep_cache" in runtime_req:
 228                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 229             if "outputDirType" in runtime_req:
 230                 if runtime_req["outputDirType"] == "local_output_dir":
 231                     # Currently the default behavior.
 232                     pass
 233                 elif runtime_req["outputDirType"] == "keep_output_dir":
 234                     mounts[self.outdir]= {
 235                         "kind": "collection",
 236                         "writable": True
 237                     }
 238
 239         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 240         if partition_req:
 241             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 242
 243         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 244         if intermediate_output_req:
 245             self.output_ttl = intermediate_output_req["outputTTL"]
 246         else:
 247             self.output_ttl = self.arvrunner.intermediate_output_ttl
 248
 249         if self.output_ttl < 0:
 250             raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 251
 252         container_request["output_ttl"] = self.output_ttl
 253         container_request["mounts"] = mounts
 254         container_request["secret_mounts"] = secret_mounts
 255         container_request["runtime_constraints"] = runtime_constraints
 256         container_request["scheduling_parameters"] = scheduling_parameters
 257
 258         enable_reuse = kwargs.get("enable_reuse", True)
 259         if enable_reuse:
 260             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 261             if reuse_req:
 262                 enable_reuse = reuse_req["enableReuse"]
 263         container_request["use_existing"] = enable_reuse
 264
 265         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 266             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 267             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 268             if container_request["name"] == "main":
 269                 container_request["name"] = wfrecord["name"]
 270             container_request["properties"]["template_uuid"] = wfuuid
 271
 272         self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 273
 274         try:
 275             if kwargs.get("submit_request_uuid"):
 276                 response = self.arvrunner.api.container_requests().update(
 277                     uuid=kwargs["submit_request_uuid"],
 278                     body=container_request
 279                 ).execute(num_retries=self.arvrunner.num_retries)
 280             else:
 281                 response = self.arvrunner.api.container_requests().create(
 282                     body=container_request
 283                 ).execute(num_retries=self.arvrunner.num_retries)
 284
 285             self.uuid = response["uuid"]
 286             self.arvrunner.process_submitted(self)
 287
 288             if response["state"] == "Final":
 289                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 290             else:
 291                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 292         except Exception as e:
 293             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 294             self.output_callback({}, "permanentFail")
 295
 296     def done(self, record):
 297         outputs = {}
 298         try:
 299             container = self.arvrunner.api.containers().get(
 300                 uuid=record["container_uuid"]
 301             ).execute(num_retries=self.arvrunner.num_retries)
 302             if container["state"] == "Complete":
 303                 rcode = container["exit_code"]
 304                 if self.successCodes and rcode in self.successCodes:
 305                     processStatus = "success"
 306                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 307                     processStatus = "temporaryFail"
 308                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 309                     processStatus = "permanentFail"
 310                 elif rcode == 0:
 311                     processStatus = "success"
 312                 else:
 313                     processStatus = "permanentFail"
 314             else:
 315                 processStatus = "permanentFail"
 316
 317             if processStatus == "permanentFail":
 318                 logc = arvados.collection.CollectionReader(container["log"],
 319                                                            api_client=self.arvrunner.api,
 320                                                            keep_client=self.arvrunner.keep_client,
 321                                                            num_retries=self.arvrunner.num_retries)
 322                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 323
 324             if record["output_uuid"]:
 325                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 326                     # Compute the trash time to avoid requesting the collection record.
 327                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 328                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 329                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 330                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 331                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 332                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 333                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 334
 335             if container["output"]:
 336                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 337         except WorkflowException as e:
 338             logger.error("%s unable to collect output from %s:\n%s",
 339                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 340             processStatus = "permanentFail"
 341         except Exception as e:
 342             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 343             processStatus = "permanentFail"
 344         finally:
 345             self.output_callback(outputs, processStatus)
 346
 347
 348 class RunnerContainer(Runner):
 349     """Submit and manage a container that runs arvados-cwl-runner."""
 350
 351     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 352         """Create an Arvados container request for this workflow.
 353
 354         The returned dict can be used to create a container passed as
 355         the +body+ argument to container_requests().create().
 356         """
 357
 358         adjustDirObjs(self.job_order, trim_listing)
 359         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 360         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 361
 362         secret_mounts = {}
 363         for param in sorted(self.job_order.keys()):
 364             if self.secret_store.has_secret(self.job_order[param]):
 365                 mnt = "/secrets/s%d" % len(secret_mounts)
 366                 secret_mounts[mnt] = {
 367                     "kind": "text",
 368                     "content": self.secret_store.retrieve(self.job_order[param])
 369                 }
 370                 self.job_order[param] = {"$include": mnt}
 371
 372         container_req = {
 373             "name": self.name,
 374             "output_path": "/var/spool/cwl",
 375             "cwd": "/var/spool/cwl",
 376             "priority": self.priority,
 377             "state": "Committed",
 378             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 379             "mounts": {
 380                 "/var/lib/cwl/cwl.input.json": {
 381                     "kind": "json",
 382                     "content": self.job_order
 383                 },
 384                 "stdout": {
 385                     "kind": "file",
 386                     "path": "/var/spool/cwl/cwl.output.json"
 387                 },
 388                 "/var/spool/cwl": {
 389                     "kind": "collection",
 390                     "writable": True
 391                 }
 392             },
 393             "secret_mounts": secret_mounts,
 394             "runtime_constraints": {
 395                 "vcpus": 1,
 396                 "ram": 1024*1024 * self.submit_runner_ram,
 397                 "API": True
 398             },
 399             "use_existing": self.enable_reuse,
 400             "properties": {}
 401         }
 402
 403         if self.tool.tool.get("id", "").startswith("keep:"):
 404             sp = self.tool.tool["id"].split('/')
 405             workflowcollection = sp[0][5:]
 406             workflowname = "/".join(sp[1:])
 407             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 408             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 409                 "kind": "collection",
 410                 "portable_data_hash": "%s" % workflowcollection
 411             }
 412         else:
 413             packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
 414             workflowpath = "/var/lib/cwl/workflow.json#main"
 415             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 416                 "kind": "json",
 417                 "content": packed
 418             }
 419             if self.tool.tool.get("id", "").startswith("arvwf:"):
 420                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 421
 422
 423         # --local means execute the workflow instead of submitting a container request
 424         # --api=containers means use the containers API
 425         # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
 426         # --disable-validate because we already validated so don't need to do it again
 427         # --eval-timeout is the timeout for javascript invocation
 428         # --parallel-task-count is the number of threads to use for job submission
 429         # --enable/disable-reuse sets desired job reuse
 430         command = ["arvados-cwl-runner",
 431                    "--local",
 432                    "--api=containers",
 433                    "--no-log-timestamps",
 434                    "--disable-validate",
 435                    "--eval-timeout=%s" % self.arvrunner.eval_timeout,
 436                    "--thread-count=%s" % self.arvrunner.thread_count,
 437                    "--enable-reuse" if self.enable_reuse else "--disable-reuse"]
 438
 439         if self.output_name:
 440             command.append("--output-name=" + self.output_name)
 441             container_req["output_name"] = self.output_name
 442
 443         if self.output_tags:
 444             command.append("--output-tags=" + self.output_tags)
 445
 446         if kwargs.get("debug"):
 447             command.append("--debug")
 448
 449         if self.on_error:
 450             command.append("--on-error=" + self.on_error)
 451
 452         if self.intermediate_output_ttl:
 453             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 454
 455         if self.arvrunner.trash_intermediate:
 456             command.append("--trash-intermediate")
 457
 458         if self.arvrunner.project_uuid:
 459             command.append("--project-uuid="+self.arvrunner.project_uuid)
 460
 461         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 462
 463         container_req["command"] = command
 464
 465         return container_req
 466
 467
 468     def run(self, **kwargs):
 469         kwargs["keepprefix"] = "keep:"
 470         job_spec = self.arvados_job_spec(**kwargs)
 471         if self.arvrunner.project_uuid:
 472             job_spec["owner_uuid"] = self.arvrunner.project_uuid
 473
 474         if kwargs.get("submit_request_uuid"):
 475             response = self.arvrunner.api.container_requests().update(
 476                 uuid=kwargs["submit_request_uuid"],
 477                 body=job_spec
 478             ).execute(num_retries=self.arvrunner.num_retries)
 479         else:
 480             response = self.arvrunner.api.container_requests().create(
 481                 body=job_spec
 482             ).execute(num_retries=self.arvrunner.num_retries)
 483
 484         self.uuid = response["uuid"]
 485         self.arvrunner.process_submitted(self)
 486
 487         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 488
 489     def done(self, record):
 490         try:
 491             container = self.arvrunner.api.containers().get(
 492                 uuid=record["container_uuid"]
 493             ).execute(num_retries=self.arvrunner.num_retries)
 494         except Exception as e:
 495             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 496             self.arvrunner.output_callback({}, "permanentFail")
 497         else:
 498             super(RunnerContainer, self).done(container)