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