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
  13 import ruamel.yaml as yaml
  14
  15 from cwltool.errors import WorkflowException
  16 from cwltool.process import get_feature, UnsupportedRequirement, shortname
  17 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  18 from cwltool.utils import aslist
  19
  20 import arvados.collection
  21
  22 from .arvdocker import arv_docker_get_image
  23 from . import done
  24 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location, remove_redundant_fields
  25 from .fsaccess import CollectionFetcher
  26 from .pathmapper import NoFollowPathMapper, trim_listing
  27 from .perf import Perf
  28
  29 logger = logging.getLogger('arvados.cwl-runner')
  30 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  31
  32 class ArvadosContainer(object):
  33     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  34
  35     def __init__(self, runner):
  36         self.arvrunner = runner
  37         self.running = False
  38         self.uuid = None
  39
  40     def update_pipeline_component(self, r):
  41         pass
  42
  43     def run(self, dry_run=False, pull_image=True, **kwargs):
  44         container_request = {
  45             "command": self.command_line,
  46             "owner_uuid": self.arvrunner.project_uuid,
  47             "name": self.name,
  48             "output_path": self.outdir,
  49             "cwd": self.outdir,
  50             "priority": 1,
  51             "state": "Committed",
  52             "properties": {},
  53         }
  54         runtime_constraints = {}
  55
  56         resources = self.builder.resources
  57         if resources is not None:
  58             runtime_constraints["vcpus"] = resources.get("cores", 1)
  59             runtime_constraints["ram"] = resources.get("ram") * 2**20
  60
  61         mounts = {
  62             self.outdir: {
  63                 "kind": "tmp",
  64                 "capacity": resources.get("outdirSize", 0) * 2**20
  65             },
  66             self.tmpdir: {
  67                 "kind": "tmp",
  68                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  69             }
  70         }
  71         scheduling_parameters = {}
  72
  73         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  74         rf.sort(key=lambda k: k.resolved)
  75         prevdir = None
  76         for resolved, target, tp, stg in rf:
  77             if not stg:
  78                 continue
  79             if prevdir and target.startswith(prevdir):
  80                 continue
  81             if tp == "Directory":
  82                 targetdir = target
  83             else:
  84                 targetdir = os.path.dirname(target)
  85             sp = resolved.split("/", 1)
  86             pdh = sp[0][5:]   # remove "keep:"
  87             mounts[targetdir] = {
  88                 "kind": "collection",
  89                 "portable_data_hash": pdh
  90             }
  91             if len(sp) == 2:
  92                 if tp == "Directory":
  93                     path = sp[1]
  94                 else:
  95                     path = os.path.dirname(sp[1])
  96                 if path and path != "/":
  97                     mounts[targetdir]["path"] = path
  98             prevdir = targetdir + "/"
  99
 100         with Perf(metrics, "generatefiles %s" % self.name):
 101             if self.generatefiles["listing"]:
 102                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
 103                                                     keep_client=self.arvrunner.keep_client,
 104                                                     num_retries=self.arvrunner.num_retries)
 105                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
 106                                                     separateDirs=False)
 107
 108                 with Perf(metrics, "createfiles %s" % self.name):
 109                     for f, p in generatemapper.items():
 110                         if not p.target:
 111                             pass
 112                         elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 113                             source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 114                             vwd.copy(path, p.target, source_collection=source)
 115                         elif p.type == "CreateFile":
 116                             with vwd.open(p.target, "w") as n:
 117                                 n.write(p.resolved.encode("utf-8"))
 118
 119                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 120                     vwd.save_new()
 121
 122                 for f, p in generatemapper.items():
 123                     if not p.target:
 124                         continue
 125                     mountpoint = "%s/%s" % (self.outdir, p.target)
 126                     mounts[mountpoint] = {"kind": "collection",
 127                                           "portable_data_hash": vwd.portable_data_hash(),
 128                                           "path": p.target}
 129                     if p.type.startswith("Writable"):
 130                         mounts[mountpoint]["writable"] = True
 131
 132         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 133         if self.environment:
 134             container_request["environment"].update(self.environment)
 135
 136         if self.stdin:
 137             sp = self.stdin[6:].split("/", 1)
 138             mounts["stdin"] = {"kind": "collection",
 139                                 "portable_data_hash": sp[0],
 140                                 "path": sp[1]}
 141
 142         if self.stderr:
 143             mounts["stderr"] = {"kind": "file",
 144                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 145
 146         if self.stdout:
 147             mounts["stdout"] = {"kind": "file",
 148                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 149
 150         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 151         if not docker_req:
 152             docker_req = {"dockerImageId": "arvados/jobs"}
 153
 154         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 155                                                                      docker_req,
 156                                                                      pull_image,
 157                                                                      self.arvrunner.project_uuid)
 158
 159         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 160         if api_req:
 161             runtime_constraints["API"] = True
 162
 163         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 164         if runtime_req:
 165             if "keep_cache" in runtime_req:
 166                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 167             if "outputDirType" in runtime_req:
 168                 if runtime_req["outputDirType"] == "local_output_dir":
 169                     # Currently the default behavior.
 170                     pass
 171                 elif runtime_req["outputDirType"] == "keep_output_dir":
 172                     mounts[self.outdir]= {
 173                         "kind": "collection",
 174                         "writable": True
 175                     }
 176
 177         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 178         if partition_req:
 179             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 180
 181         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 182         if intermediate_output_req:
 183             self.output_ttl = intermediate_output_req["outputTTL"]
 184         else:
 185             self.output_ttl = self.arvrunner.intermediate_output_ttl
 186
 187         if self.output_ttl < 0:
 188             raise WorkflowError("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 189
 190         container_request["output_ttl"] = self.output_ttl
 191         container_request["mounts"] = mounts
 192         container_request["runtime_constraints"] = runtime_constraints
 193         container_request["scheduling_parameters"] = scheduling_parameters
 194
 195         enable_reuse = kwargs.get("enable_reuse", True)
 196         if enable_reuse:
 197             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 198             if reuse_req:
 199                 enable_reuse = reuse_req["enableReuse"]
 200         container_request["use_existing"] = enable_reuse
 201
 202         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 203             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 204             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 205             if container_request["name"] == "main":
 206                 container_request["name"] = wfrecord["name"]
 207             container_request["properties"]["template_uuid"] = wfuuid
 208
 209         try:
 210             response = self.arvrunner.api.container_requests().create(
 211                 body=container_request
 212             ).execute(num_retries=self.arvrunner.num_retries)
 213
 214             self.uuid = response["uuid"]
 215             self.arvrunner.processes[self.uuid] = self
 216
 217             if response["state"] == "Final":
 218                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 219                 self.done(response)
 220             else:
 221                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 222         except Exception as e:
 223             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 224             self.output_callback({}, "permanentFail")
 225
 226     def done(self, record):
 227         outputs = {}
 228         try:
 229             container = self.arvrunner.api.containers().get(
 230                 uuid=record["container_uuid"]
 231             ).execute(num_retries=self.arvrunner.num_retries)
 232             if container["state"] == "Complete":
 233                 rcode = container["exit_code"]
 234                 if self.successCodes and rcode in self.successCodes:
 235                     processStatus = "success"
 236                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 237                     processStatus = "temporaryFail"
 238                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 239                     processStatus = "permanentFail"
 240                 elif rcode == 0:
 241                     processStatus = "success"
 242                 else:
 243                     processStatus = "permanentFail"
 244             else:
 245                 processStatus = "permanentFail"
 246
 247             if processStatus == "permanentFail":
 248                 logc = arvados.collection.CollectionReader(container["log"],
 249                                                            api_client=self.arvrunner.api,
 250                                                            keep_client=self.arvrunner.keep_client,
 251                                                            num_retries=self.arvrunner.num_retries)
 252                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 253
 254             if record["output_uuid"]:
 255                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 256                     # Compute the trash time to avoid requesting the collection record.
 257                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 258                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 259                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 260                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 261                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 262                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 263                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 264
 265             if container["output"]:
 266                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 267         except WorkflowException as e:
 268             logger.error("%s unable to collect output from %s:\n%s",
 269                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 270             processStatus = "permanentFail"
 271         except Exception as e:
 272             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 273             processStatus = "permanentFail"
 274         finally:
 275             self.output_callback(outputs, processStatus)
 276             if record["uuid"] in self.arvrunner.processes:
 277                 del self.arvrunner.processes[record["uuid"]]
 278
 279
 280 class RunnerContainer(Runner):
 281     """Submit and manage a container that runs arvados-cwl-runner."""
 282
 283     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 284         """Create an Arvados container request for this workflow.
 285
 286         The returned dict can be used to create a container passed as
 287         the +body+ argument to container_requests().create().
 288         """
 289
 290         adjustDirObjs(self.job_order, trim_listing)
 291         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 292         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 293
 294         container_req = {
 295             "owner_uuid": self.arvrunner.project_uuid,
 296             "name": self.name,
 297             "output_path": "/var/spool/cwl",
 298             "cwd": "/var/spool/cwl",
 299             "priority": 1,
 300             "state": "Committed",
 301             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 302             "mounts": {
 303                 "/var/lib/cwl/cwl.input.json": {
 304                     "kind": "json",
 305                     "content": self.job_order
 306                 },
 307                 "stdout": {
 308                     "kind": "file",
 309                     "path": "/var/spool/cwl/cwl.output.json"
 310                 },
 311                 "/var/spool/cwl": {
 312                     "kind": "collection",
 313                     "writable": True
 314                 }
 315             },
 316             "runtime_constraints": {
 317                 "vcpus": 1,
 318                 "ram": 1024*1024 * self.submit_runner_ram,
 319                 "API": True
 320             },
 321             "use_existing": self.enable_reuse,
 322             "properties": {}
 323         }
 324
 325         if self.tool.tool.get("id", "").startswith("keep:"):
 326             sp = self.tool.tool["id"].split('/')
 327             workflowcollection = sp[0][5:]
 328             workflowname = "/".join(sp[1:])
 329             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 330             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 331                 "kind": "collection",
 332                 "portable_data_hash": "%s" % workflowcollection
 333             }
 334         else:
 335             packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
 336             workflowpath = "/var/lib/cwl/workflow.json#main"
 337             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 338                 "kind": "json",
 339                 "content": packed
 340             }
 341             if self.tool.tool.get("id", "").startswith("arvwf:"):
 342                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 343
 344
 345         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 346         if self.output_name:
 347             command.append("--output-name=" + self.output_name)
 348             container_req["output_name"] = self.output_name
 349
 350         if self.output_tags:
 351             command.append("--output-tags=" + self.output_tags)
 352
 353         if kwargs.get("debug"):
 354             command.append("--debug")
 355
 356         if self.enable_reuse:
 357             command.append("--enable-reuse")
 358         else:
 359             command.append("--disable-reuse")
 360
 361         if self.on_error:
 362             command.append("--on-error=" + self.on_error)
 363
 364         if self.intermediate_output_ttl:
 365             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 366
 367         if self.arvrunner.trash_intermediate:
 368             command.append("--trash-intermediate")
 369
 370         if self.arvrunner.project_uuid:
 371             command.append("--project-uuid="+self.arvrunner.project_uuid)
 372
 373         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 374
 375         container_req["command"] = command
 376
 377         return container_req
 378
 379
 380     def run(self, *args, **kwargs):
 381         kwargs["keepprefix"] = "keep:"
 382         job_spec = self.arvados_job_spec(*args, **kwargs)
 383         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 384
 385         response = self.arvrunner.api.container_requests().create(
 386             body=job_spec
 387         ).execute(num_retries=self.arvrunner.num_retries)
 388
 389         self.uuid = response["uuid"]
 390         self.arvrunner.processes[self.uuid] = self
 391
 392         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 393
 394         if response["state"] == "Final":
 395             self.done(response)
 396
 397     def done(self, record):
 398         try:
 399             container = self.arvrunner.api.containers().get(
 400                 uuid=record["container_uuid"]
 401             ).execute(num_retries=self.arvrunner.num_retries)
 402         except Exception as e:
 403             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 404             self.arvrunner.output_callback({}, "permanentFail")
 405         else:
 406             super(RunnerContainer, self).done(container)
 407         finally:
 408             if record["uuid"] in self.arvrunner.processes:
 409                 del self.arvrunner.processes[record["uuid"]]