sdk/cwl/arvados_cwl/arvcontainer.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 import logging
   6 import json
   7 import os
   8 import urllib
   9 import time
  10 import datetime
  11 import ciso8601
  12 import uuid
  13
  14 import ruamel.yaml as yaml
  15
  16 from cwltool.errors import WorkflowException
  17 from cwltool.process import get_feature, UnsupportedRequirement, shortname
  18 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  19 from cwltool.utils import aslist
  20
  21 import arvados.collection
  22
  23 from .arvdocker import arv_docker_get_image
  24 from . import done
  25 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields
  26 from .fsaccess import CollectionFetcher
  27 from .pathmapper import NoFollowPathMapper, trim_listing
  28 from .perf import Perf
  29
  30 logger = logging.getLogger('arvados.cwl-runner')
  31 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  32
  33 class ArvadosContainer(object):
  34     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  35
  36     def __init__(self, runner):
  37         self.arvrunner = runner
  38         self.running = False
  39         self.uuid = None
  40
  41     def update_pipeline_component(self, r):
  42         pass
  43
  44     def run(self, dry_run=False, pull_image=True, **kwargs):
  45         # ArvadosCommandTool subclasses from cwltool.CommandLineTool,
  46         # which calls makeJobRunner() to get a new ArvadosContainer
  47         # object.  The fields that define execution such as
  48         # command_line, environment, etc are set on the
  49         # ArvadosContainer object by CommandLineTool.job() before
  50         # run() is called.
  51
  52         container_request = {
  53             "command": self.command_line,
  54             "owner_uuid": self.arvrunner.project_uuid,
  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.secret_store.has_secret(self.command_line):
  65             raise WorkflowException("Secret material leaked on command line, only file literals may contain secrets")
  66
  67         if self.arvrunner.secret_store.has_secret(self.environment):
  68             raise WorkflowException("Secret material leaked in environment, only file literals may contain secrets")
  69
  70         resources = self.builder.resources
  71         if resources is not None:
  72             runtime_constraints["vcpus"] = resources.get("cores", 1)
  73             runtime_constraints["ram"] = resources.get("ram") * 2**20
  74
  75         mounts = {
  76             self.outdir: {
  77                 "kind": "tmp",
  78                 "capacity": resources.get("outdirSize", 0) * 2**20
  79             },
  80             self.tmpdir: {
  81                 "kind": "tmp",
  82                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  83             }
  84         }
  85         secret_mounts = {}
  86         scheduling_parameters = {}
  87
  88         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  89         rf.sort(key=lambda k: k.resolved)
  90         prevdir = None
  91         for resolved, target, tp, stg in rf:
  92             if not stg:
  93                 continue
  94             if prevdir and target.startswith(prevdir):
  95                 continue
  96             if tp == "Directory":
  97                 targetdir = target
  98             else:
  99                 targetdir = os.path.dirname(target)
 100             sp = resolved.split("/", 1)
 101             pdh = sp[0][5:]   # remove "keep:"
 102             mounts[targetdir] = {
 103                 "kind": "collection",
 104                 "portable_data_hash": pdh
 105             }
 106             if len(sp) == 2:
 107                 if tp == "Directory":
 108                     path = sp[1]
 109                 else:
 110                     path = os.path.dirname(sp[1])
 111                 if path and path != "/":
 112                     mounts[targetdir]["path"] = path
 113             prevdir = targetdir + "/"
 114
 115         with Perf(metrics, "generatefiles %s" % self.name):
 116             if self.generatefiles["listing"]:
 117                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
 118                                                     keep_client=self.arvrunner.keep_client,
 119                                                     num_retries=self.arvrunner.num_retries)
 120                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
 121                                                     separateDirs=False)
 122
 123                 sorteditems = sorted(generatemapper.items(), None, key=lambda n: n[1].target)
 124
 125                 logger.debug("generatemapper is %s", sorteditems)
 126
 127                 with Perf(metrics, "createfiles %s" % self.name):
 128                     for f, p in sorteditems:
 129                         if not p.target:
 130                             pass
 131                         elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 132                             if p.resolved.startswith("_:"):
 133                                 vwd.mkdirs(p.target)
 134                             else:
 135                                 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 136                                 vwd.copy(path, p.target, source_collection=source)
 137                         elif p.type == "CreateFile":
 138                             if self.arvrunner.secret_store.has_secret(p.resolved):
 139                                 secret_mounts["%s/%s" % (self.outdir, p.target)] = {
 140                                     "kind": "text",
 141                                     "content": self.arvrunner.secret_store.retrieve(p.resolved)
 142                                 }
 143                             else:
 144                                 with vwd.open(p.target, "w") as n:
 145                                     n.write(p.resolved.encode("utf-8"))
 146
 147                 def keepemptydirs(p):
 148                     if isinstance(p, arvados.collection.RichCollectionBase):
 149                         if len(p) == 0:
 150                             p.open(".keep", "w").close()
 151                         else:
 152                             for c in p:
 153                                 keepemptydirs(p[c])
 154
 155                 keepemptydirs(vwd)
 156
 157                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 158                     vwd.save_new()
 159
 160                 prev = None
 161                 for f, p in sorteditems:
 162                     if (not p.target or self.arvrunner.secret_store.has_secret(p.resolved) or
 163                         (prev is not None and p.target.startswith(prev))):
 164                         continue
 165                     mountpoint = "%s/%s" % (self.outdir, p.target)
 166                     mounts[mountpoint] = {"kind": "collection",
 167                                           "portable_data_hash": vwd.portable_data_hash(),
 168                                           "path": p.target}
 169                     if p.type.startswith("Writable"):
 170                         mounts[mountpoint]["writable"] = True
 171                     prev = p.target + "/"
 172
 173         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 174         if self.environment:
 175             container_request["environment"].update(self.environment)
 176
 177         if self.stdin:
 178             sp = self.stdin[6:].split("/", 1)
 179             mounts["stdin"] = {"kind": "collection",
 180                                 "portable_data_hash": sp[0],
 181                                 "path": sp[1]}
 182
 183         if self.stderr:
 184             mounts["stderr"] = {"kind": "file",
 185                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 186
 187         if self.stdout:
 188             mounts["stdout"] = {"kind": "file",
 189                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 190
 191         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 192         if not docker_req:
 193             docker_req = {"dockerImageId": "arvados/jobs"}
 194
 195         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 196                                                                      docker_req,
 197                                                                      pull_image,
 198                                                                      self.arvrunner.project_uuid)
 199
 200         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 201         if api_req:
 202             runtime_constraints["API"] = True
 203
 204         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 205         if runtime_req:
 206             if "keep_cache" in runtime_req:
 207                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 208             if "outputDirType" in runtime_req:
 209                 if runtime_req["outputDirType"] == "local_output_dir":
 210                     # Currently the default behavior.
 211                     pass
 212                 elif runtime_req["outputDirType"] == "keep_output_dir":
 213                     mounts[self.outdir]= {
 214                         "kind": "collection",
 215                         "writable": True
 216                     }
 217
 218         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 219         if partition_req:
 220             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 221
 222         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 223         if intermediate_output_req:
 224             self.output_ttl = intermediate_output_req["outputTTL"]
 225         else:
 226             self.output_ttl = self.arvrunner.intermediate_output_ttl
 227
 228         if self.output_ttl < 0:
 229             raise WorkflowException("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 230
 231         container_request["output_ttl"] = self.output_ttl
 232         container_request["mounts"] = mounts
 233         container_request["secret_mounts"] = secret_mounts
 234         container_request["runtime_constraints"] = runtime_constraints
 235         container_request["scheduling_parameters"] = scheduling_parameters
 236
 237         enable_reuse = kwargs.get("enable_reuse", True)
 238         if enable_reuse:
 239             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 240             if reuse_req:
 241                 enable_reuse = reuse_req["enableReuse"]
 242         container_request["use_existing"] = enable_reuse
 243
 244         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 245             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 246             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 247             if container_request["name"] == "main":
 248                 container_request["name"] = wfrecord["name"]
 249             container_request["properties"]["template_uuid"] = wfuuid
 250
 251         self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 252
 253         try:
 254             response = self.arvrunner.api.container_requests().create(
 255                 body=container_request
 256             ).execute(num_retries=self.arvrunner.num_retries)
 257
 258             self.uuid = response["uuid"]
 259             self.arvrunner.process_submitted(self)
 260
 261             if response["state"] == "Final":
 262                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 263                 self.done(response)
 264             else:
 265                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 266         except Exception as e:
 267             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 268             self.output_callback({}, "permanentFail")
 269
 270     def done(self, record):
 271         outputs = {}
 272         try:
 273             container = self.arvrunner.api.containers().get(
 274                 uuid=record["container_uuid"]
 275             ).execute(num_retries=self.arvrunner.num_retries)
 276             if container["state"] == "Complete":
 277                 rcode = container["exit_code"]
 278                 if self.successCodes and rcode in self.successCodes:
 279                     processStatus = "success"
 280                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 281                     processStatus = "temporaryFail"
 282                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 283                     processStatus = "permanentFail"
 284                 elif rcode == 0:
 285                     processStatus = "success"
 286                 else:
 287                     processStatus = "permanentFail"
 288             else:
 289                 processStatus = "permanentFail"
 290
 291             if processStatus == "permanentFail":
 292                 logc = arvados.collection.CollectionReader(container["log"],
 293                                                            api_client=self.arvrunner.api,
 294                                                            keep_client=self.arvrunner.keep_client,
 295                                                            num_retries=self.arvrunner.num_retries)
 296                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 297
 298             if record["output_uuid"]:
 299                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 300                     # Compute the trash time to avoid requesting the collection record.
 301                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 302                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 303                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 304                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 305                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 306                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 307                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 308
 309             if container["output"]:
 310                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 311         except WorkflowException as e:
 312             logger.error("%s unable to collect output from %s:\n%s",
 313                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 314             processStatus = "permanentFail"
 315         except Exception as e:
 316             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 317             processStatus = "permanentFail"
 318         finally:
 319             self.output_callback(outputs, processStatus)
 320             self.arvrunner.process_done(record["uuid"])
 321
 322
 323 class RunnerContainer(Runner):
 324     """Submit and manage a container that runs arvados-cwl-runner."""
 325
 326     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 327         """Create an Arvados container request for this workflow.
 328
 329         The returned dict can be used to create a container passed as
 330         the +body+ argument to container_requests().create().
 331         """
 332
 333         adjustDirObjs(self.job_order, trim_listing)
 334         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 335         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 336
 337         secret_mounts = {}
 338         for param in sorted(self.job_order.keys()):
 339             if self.secret_store.has_secret(self.job_order[param]):
 340                 mnt = "/secrets/s%d" % len(secret_mounts)
 341                 secret_mounts[mnt] = {
 342                     "kind": "text",
 343                     "content": self.secret_store.retrieve(self.job_order[param])
 344                 }
 345                 self.job_order[param] = {"$include": mnt}
 346
 347         container_req = {
 348             "owner_uuid": self.arvrunner.project_uuid,
 349             "name": self.name,
 350             "output_path": "/var/spool/cwl",
 351             "cwd": "/var/spool/cwl",
 352             "priority": self.priority,
 353             "state": "Committed",
 354             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 355             "mounts": {
 356                 "/var/lib/cwl/cwl.input.json": {
 357                     "kind": "json",
 358                     "content": self.job_order
 359                 },
 360                 "stdout": {
 361                     "kind": "file",
 362                     "path": "/var/spool/cwl/cwl.output.json"
 363                 },
 364                 "/var/spool/cwl": {
 365                     "kind": "collection",
 366                     "writable": True
 367                 }
 368             },
 369             "secret_mounts": secret_mounts,
 370             "runtime_constraints": {
 371                 "vcpus": 1,
 372                 "ram": 1024*1024 * self.submit_runner_ram,
 373                 "API": True
 374             },
 375             "use_existing": self.enable_reuse,
 376             "properties": {}
 377         }
 378
 379         if self.tool.tool.get("id", "").startswith("keep:"):
 380             sp = self.tool.tool["id"].split('/')
 381             workflowcollection = sp[0][5:]
 382             workflowname = "/".join(sp[1:])
 383             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 384             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 385                 "kind": "collection",
 386                 "portable_data_hash": "%s" % workflowcollection
 387             }
 388         else:
 389             packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
 390             workflowpath = "/var/lib/cwl/workflow.json#main"
 391             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 392                 "kind": "json",
 393                 "content": packed
 394             }
 395             if self.tool.tool.get("id", "").startswith("arvwf:"):
 396                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 397
 398
 399         # --local means execute the workflow instead of submitting a container request
 400         # --api=containers means use the containers API
 401         # --no-log-timestamps means don't add timestamps (the logging infrastructure does this)
 402         # --disable-validate because we already validated so don't need to do it again
 403         # --eval-timeout is the timeout for javascript invocation
 404         # --parallel-task-count is the number of threads to use for job submission
 405         # --enable/disable-reuse sets desired job reuse
 406         command = ["arvados-cwl-runner",
 407                    "--local",
 408                    "--api=containers",
 409                    "--no-log-timestamps",
 410                    "--disable-validate",
 411                    "--eval-timeout=%s" % self.arvrunner.eval_timeout,
 412                    "--thread-count=%s" % self.arvrunner.thread_count,
 413                    "--enable-reuse" if self.enable_reuse else "--disable-reuse"]
 414
 415         if self.output_name:
 416             command.append("--output-name=" + self.output_name)
 417             container_req["output_name"] = self.output_name
 418
 419         if self.output_tags:
 420             command.append("--output-tags=" + self.output_tags)
 421
 422         if kwargs.get("debug"):
 423             command.append("--debug")
 424
 425         if self.on_error:
 426             command.append("--on-error=" + self.on_error)
 427
 428         if self.intermediate_output_ttl:
 429             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 430
 431         if self.arvrunner.trash_intermediate:
 432             command.append("--trash-intermediate")
 433
 434         if self.arvrunner.project_uuid:
 435             command.append("--project-uuid="+self.arvrunner.project_uuid)
 436
 437         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 438
 439         container_req["command"] = command
 440
 441         return container_req
 442
 443
 444     def run(self, **kwargs):
 445         kwargs["keepprefix"] = "keep:"
 446         job_spec = self.arvados_job_spec(**kwargs)
 447         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 448
 449         response = self.arvrunner.api.container_requests().create(
 450             body=job_spec
 451         ).execute(num_retries=self.arvrunner.num_retries)
 452
 453         self.uuid = response["uuid"]
 454         self.arvrunner.process_submitted(self)
 455
 456         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 457
 458         if response["state"] == "Final":
 459             self.done(response)
 460
 461     def done(self, record):
 462         try:
 463             container = self.arvrunner.api.containers().get(
 464                 uuid=record["container_uuid"]
 465             ).execute(num_retries=self.arvrunner.num_retries)
 466         except Exception as e:
 467             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 468             self.arvrunner.output_callback({}, "permanentFail")
 469         else:
 470             super(RunnerContainer, self).done(container)
 471         finally:
 472             self.arvrunner.process_done(record["uuid"])