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
  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
  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"):
 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
 130         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 131         if self.environment:
 132             container_request["environment"].update(self.environment)
 133
 134         if self.stdin:
 135             sp = self.stdin[6:].split("/", 1)
 136             mounts["stdin"] = {"kind": "collection",
 137                                 "portable_data_hash": sp[0],
 138                                 "path": sp[1]}
 139
 140         if self.stderr:
 141             mounts["stderr"] = {"kind": "file",
 142                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 143
 144         if self.stdout:
 145             mounts["stdout"] = {"kind": "file",
 146                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 147
 148         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 149         if not docker_req:
 150             docker_req = {"dockerImageId": "arvados/jobs"}
 151
 152         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 153                                                                      docker_req,
 154                                                                      pull_image,
 155                                                                      self.arvrunner.project_uuid)
 156
 157         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 158         if api_req:
 159             runtime_constraints["API"] = True
 160
 161         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 162         if runtime_req:
 163             if "keep_cache" in runtime_req:
 164                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 165             if "outputDirType" in runtime_req:
 166                 if runtime_req["outputDirType"] == "local_output_dir":
 167                     # Currently the default behavior.
 168                     pass
 169                 elif runtime_req["outputDirType"] == "keep_output_dir":
 170                     mounts[self.outdir]= {
 171                         "kind": "collection",
 172                         "writable": True
 173                     }
 174
 175         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 176         if partition_req:
 177             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 178
 179         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 180         if intermediate_output_req:
 181             self.output_ttl = intermediate_output_req["outputTTL"]
 182         else:
 183             self.output_ttl = self.arvrunner.intermediate_output_ttl
 184
 185         if self.output_ttl < 0:
 186             raise WorkflowError("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 187
 188         container_request["output_ttl"] = self.output_ttl
 189         container_request["mounts"] = mounts
 190         container_request["runtime_constraints"] = runtime_constraints
 191         container_request["scheduling_parameters"] = scheduling_parameters
 192
 193         enable_reuse = kwargs.get("enable_reuse", True)
 194         if enable_reuse:
 195             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 196             if reuse_req:
 197                 enable_reuse = reuse_req["enableReuse"]
 198         container_request["use_existing"] = enable_reuse
 199
 200         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 201             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 202             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 203             if container_request["name"] == "main":
 204                 container_request["name"] = wfrecord["name"]
 205             container_request["properties"]["template_uuid"] = wfuuid
 206
 207         try:
 208             response = self.arvrunner.api.container_requests().create(
 209                 body=container_request
 210             ).execute(num_retries=self.arvrunner.num_retries)
 211
 212             self.uuid = response["uuid"]
 213             self.arvrunner.processes[self.uuid] = self
 214
 215             if response["state"] == "Final":
 216                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 217                 self.done(response)
 218             else:
 219                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 220         except Exception as e:
 221             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 222             self.output_callback({}, "permanentFail")
 223
 224     def done(self, record):
 225         outputs = {}
 226         try:
 227             container = self.arvrunner.api.containers().get(
 228                 uuid=record["container_uuid"]
 229             ).execute(num_retries=self.arvrunner.num_retries)
 230             if container["state"] == "Complete":
 231                 rcode = container["exit_code"]
 232                 if self.successCodes and rcode in self.successCodes:
 233                     processStatus = "success"
 234                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 235                     processStatus = "temporaryFail"
 236                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 237                     processStatus = "permanentFail"
 238                 elif rcode == 0:
 239                     processStatus = "success"
 240                 else:
 241                     processStatus = "permanentFail"
 242             else:
 243                 processStatus = "permanentFail"
 244
 245             if processStatus == "permanentFail":
 246                 logc = arvados.collection.CollectionReader(container["log"],
 247                                                            api_client=self.arvrunner.api,
 248                                                            keep_client=self.arvrunner.keep_client,
 249                                                            num_retries=self.arvrunner.num_retries)
 250                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 251
 252             if record["output_uuid"]:
 253                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 254                     # Compute the trash time to avoid requesting the collection record.
 255                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 256                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 257                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 258                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 259                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 260                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 261                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 262
 263             if container["output"]:
 264                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 265         except WorkflowException as e:
 266             logger.error("%s unable to collect output from %s:\n%s",
 267                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 268             processStatus = "permanentFail"
 269         except Exception as e:
 270             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 271             processStatus = "permanentFail"
 272         finally:
 273             self.output_callback(outputs, processStatus)
 274             if record["uuid"] in self.arvrunner.processes:
 275                 del self.arvrunner.processes[record["uuid"]]
 276
 277
 278 class RunnerContainer(Runner):
 279     """Submit and manage a container that runs arvados-cwl-runner."""
 280
 281     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 282         """Create an Arvados container request for this workflow.
 283
 284         The returned dict can be used to create a container passed as
 285         the +body+ argument to container_requests().create().
 286         """
 287
 288         adjustDirObjs(self.job_order, trim_listing)
 289         adjustFileObjs(self.job_order, trim_anonymous_location)
 290         adjustDirObjs(self.job_order, trim_anonymous_location)
 291
 292         container_req = {
 293             "owner_uuid": self.arvrunner.project_uuid,
 294             "name": self.name,
 295             "output_path": "/var/spool/cwl",
 296             "cwd": "/var/spool/cwl",
 297             "priority": 1,
 298             "state": "Committed",
 299             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 300             "mounts": {
 301                 "/var/lib/cwl/cwl.input.json": {
 302                     "kind": "json",
 303                     "content": self.job_order
 304                 },
 305                 "stdout": {
 306                     "kind": "file",
 307                     "path": "/var/spool/cwl/cwl.output.json"
 308                 },
 309                 "/var/spool/cwl": {
 310                     "kind": "collection",
 311                     "writable": True
 312                 }
 313             },
 314             "runtime_constraints": {
 315                 "vcpus": 1,
 316                 "ram": 1024*1024 * self.submit_runner_ram,
 317                 "API": True
 318             },
 319             "properties": {}
 320         }
 321
 322         if self.tool.tool.get("id", "").startswith("keep:"):
 323             sp = self.tool.tool["id"].split('/')
 324             workflowcollection = sp[0][5:]
 325             workflowname = "/".join(sp[1:])
 326             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 327             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 328                 "kind": "collection",
 329                 "portable_data_hash": "%s" % workflowcollection
 330             }
 331         else:
 332             packed = packed_workflow(self.arvrunner, self.tool)
 333             workflowpath = "/var/lib/cwl/workflow.json#main"
 334             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 335                 "kind": "json",
 336                 "content": packed
 337             }
 338             if self.tool.tool.get("id", "").startswith("arvwf:"):
 339                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 340
 341         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 342         if self.output_name:
 343             command.append("--output-name=" + self.output_name)
 344             container_req["output_name"] = self.output_name
 345
 346         if self.output_tags:
 347             command.append("--output-tags=" + self.output_tags)
 348
 349         if kwargs.get("debug"):
 350             command.append("--debug")
 351
 352         if self.enable_reuse:
 353             command.append("--enable-reuse")
 354         else:
 355             command.append("--disable-reuse")
 356
 357         if self.on_error:
 358             command.append("--on-error=" + self.on_error)
 359
 360         if self.intermediate_output_ttl:
 361             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 362
 363         if self.arvrunner.trash_intermediate:
 364             command.append("--trash-intermediate")
 365
 366         if self.arvrunner.project_uuid:
 367             command.append("--project-uuid="+self.arvrunner.project_uuid)
 368
 369         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 370
 371         container_req["command"] = command
 372
 373         return container_req
 374
 375
 376     def run(self, *args, **kwargs):
 377         kwargs["keepprefix"] = "keep:"
 378         job_spec = self.arvados_job_spec(*args, **kwargs)
 379         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 380
 381         response = self.arvrunner.api.container_requests().create(
 382             body=job_spec
 383         ).execute(num_retries=self.arvrunner.num_retries)
 384
 385         self.uuid = response["uuid"]
 386         self.arvrunner.processes[self.uuid] = self
 387
 388         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 389
 390         if response["state"] == "Final":
 391             self.done(response)
 392
 393     def done(self, record):
 394         try:
 395             container = self.arvrunner.api.containers().get(
 396                 uuid=record["container_uuid"]
 397             ).execute(num_retries=self.arvrunner.num_retries)
 398         except Exception as e:
 399             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 400             self.arvrunner.output_callback({}, "permanentFail")
 401         else:
 402             super(RunnerContainer, self).done(container)
 403         finally:
 404             if record["uuid"] in self.arvrunner.processes:
 405                 del self.arvrunner.processes[record["uuid"]]