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