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