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                 logger.debug("generatemapper is %s", generatemapper._pathmap)
 109
 110                 with Perf(metrics, "createfiles %s" % self.name):
 111                     for f, p in generatemapper.items():
 112                         if not p.target:
 113                             pass
 114                         elif p.type in ("File", "Directory", "WritableFile", "WritableDirectory"):
 115                             if p.resolved.startswith("_:"):
 116                                 vwd.mkdirs(p.target)
 117                             else:
 118                                 source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 119                                 vwd.copy(path, p.target, source_collection=source)
 120                         elif p.type == "CreateFile":
 121                             with vwd.open(p.target, "w") as n:
 122                                 n.write(p.resolved.encode("utf-8"))
 123
 124                 def keepemptydirs(p):
 125                     if isinstance(p, arvados.collection.RichCollectionBase):
 126                         if len(p) == 0:
 127                             p.open(".keep", "w").close()
 128                         else:
 129                             for c in p:
 130                                 keepemptydirs(p[c])
 131
 132                 keepemptydirs(vwd)
 133
 134                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 135                     vwd.save_new()
 136
 137                 for f, p in generatemapper.items():
 138                     if not p.target:
 139                         continue
 140                     mountpoint = "%s/%s" % (self.outdir, p.target)
 141                     mounts[mountpoint] = {"kind": "collection",
 142                                           "portable_data_hash": vwd.portable_data_hash(),
 143                                           "path": p.target}
 144                     if p.type.startswith("Writable"):
 145                         mounts[mountpoint]["writable"] = True
 146
 147         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 148         if self.environment:
 149             container_request["environment"].update(self.environment)
 150
 151         if self.stdin:
 152             sp = self.stdin[6:].split("/", 1)
 153             mounts["stdin"] = {"kind": "collection",
 154                                 "portable_data_hash": sp[0],
 155                                 "path": sp[1]}
 156
 157         if self.stderr:
 158             mounts["stderr"] = {"kind": "file",
 159                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 160
 161         if self.stdout:
 162             mounts["stdout"] = {"kind": "file",
 163                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 164
 165         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 166         if not docker_req:
 167             docker_req = {"dockerImageId": "arvados/jobs"}
 168
 169         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 170                                                                      docker_req,
 171                                                                      pull_image,
 172                                                                      self.arvrunner.project_uuid)
 173
 174         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 175         if api_req:
 176             runtime_constraints["API"] = True
 177
 178         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 179         if runtime_req:
 180             if "keep_cache" in runtime_req:
 181                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 182             if "outputDirType" in runtime_req:
 183                 if runtime_req["outputDirType"] == "local_output_dir":
 184                     # Currently the default behavior.
 185                     pass
 186                 elif runtime_req["outputDirType"] == "keep_output_dir":
 187                     mounts[self.outdir]= {
 188                         "kind": "collection",
 189                         "writable": True
 190                     }
 191
 192         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 193         if partition_req:
 194             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 195
 196         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 197         if intermediate_output_req:
 198             self.output_ttl = intermediate_output_req["outputTTL"]
 199         else:
 200             self.output_ttl = self.arvrunner.intermediate_output_ttl
 201
 202         if self.output_ttl < 0:
 203             raise WorkflowError("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 204
 205         container_request["output_ttl"] = self.output_ttl
 206         container_request["mounts"] = mounts
 207         container_request["runtime_constraints"] = runtime_constraints
 208         container_request["scheduling_parameters"] = scheduling_parameters
 209
 210         enable_reuse = kwargs.get("enable_reuse", True)
 211         if enable_reuse:
 212             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 213             if reuse_req:
 214                 enable_reuse = reuse_req["enableReuse"]
 215         container_request["use_existing"] = enable_reuse
 216
 217         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 218             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 219             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 220             if container_request["name"] == "main":
 221                 container_request["name"] = wfrecord["name"]
 222             container_request["properties"]["template_uuid"] = wfuuid
 223
 224         try:
 225             response = self.arvrunner.api.container_requests().create(
 226                 body=container_request
 227             ).execute(num_retries=self.arvrunner.num_retries)
 228
 229             self.uuid = response["uuid"]
 230             self.arvrunner.processes[self.uuid] = self
 231
 232             if response["state"] == "Final":
 233                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 234                 self.done(response)
 235             else:
 236                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 237         except Exception as e:
 238             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 239             self.output_callback({}, "permanentFail")
 240
 241     def done(self, record):
 242         outputs = {}
 243         try:
 244             container = self.arvrunner.api.containers().get(
 245                 uuid=record["container_uuid"]
 246             ).execute(num_retries=self.arvrunner.num_retries)
 247             if container["state"] == "Complete":
 248                 rcode = container["exit_code"]
 249                 if self.successCodes and rcode in self.successCodes:
 250                     processStatus = "success"
 251                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 252                     processStatus = "temporaryFail"
 253                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 254                     processStatus = "permanentFail"
 255                 elif rcode == 0:
 256                     processStatus = "success"
 257                 else:
 258                     processStatus = "permanentFail"
 259             else:
 260                 processStatus = "permanentFail"
 261
 262             if processStatus == "permanentFail":
 263                 logc = arvados.collection.CollectionReader(container["log"],
 264                                                            api_client=self.arvrunner.api,
 265                                                            keep_client=self.arvrunner.keep_client,
 266                                                            num_retries=self.arvrunner.num_retries)
 267                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 268
 269             if record["output_uuid"]:
 270                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 271                     # Compute the trash time to avoid requesting the collection record.
 272                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 273                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 274                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 275                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 276                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 277                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 278                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 279
 280             if container["output"]:
 281                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 282         except WorkflowException as e:
 283             logger.error("%s unable to collect output from %s:\n%s",
 284                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 285             processStatus = "permanentFail"
 286         except Exception as e:
 287             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 288             processStatus = "permanentFail"
 289         finally:
 290             self.output_callback(outputs, processStatus)
 291             if record["uuid"] in self.arvrunner.processes:
 292                 del self.arvrunner.processes[record["uuid"]]
 293
 294
 295 class RunnerContainer(Runner):
 296     """Submit and manage a container that runs arvados-cwl-runner."""
 297
 298     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 299         """Create an Arvados container request for this workflow.
 300
 301         The returned dict can be used to create a container passed as
 302         the +body+ argument to container_requests().create().
 303         """
 304
 305         adjustDirObjs(self.job_order, trim_listing)
 306         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 307         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 308
 309         container_req = {
 310             "owner_uuid": self.arvrunner.project_uuid,
 311             "name": self.name,
 312             "output_path": "/var/spool/cwl",
 313             "cwd": "/var/spool/cwl",
 314             "priority": 1,
 315             "state": "Committed",
 316             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 317             "mounts": {
 318                 "/var/lib/cwl/cwl.input.json": {
 319                     "kind": "json",
 320                     "content": self.job_order
 321                 },
 322                 "stdout": {
 323                     "kind": "file",
 324                     "path": "/var/spool/cwl/cwl.output.json"
 325                 },
 326                 "/var/spool/cwl": {
 327                     "kind": "collection",
 328                     "writable": True
 329                 }
 330             },
 331             "runtime_constraints": {
 332                 "vcpus": 1,
 333                 "ram": 1024*1024 * self.submit_runner_ram,
 334                 "API": True
 335             },
 336             "use_existing": self.enable_reuse,
 337             "properties": {}
 338         }
 339
 340         if self.tool.tool.get("id", "").startswith("keep:"):
 341             sp = self.tool.tool["id"].split('/')
 342             workflowcollection = sp[0][5:]
 343             workflowname = "/".join(sp[1:])
 344             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 345             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 346                 "kind": "collection",
 347                 "portable_data_hash": "%s" % workflowcollection
 348             }
 349         else:
 350             packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
 351             workflowpath = "/var/lib/cwl/workflow.json#main"
 352             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 353                 "kind": "json",
 354                 "content": packed
 355             }
 356             if self.tool.tool.get("id", "").startswith("arvwf:"):
 357                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 358
 359
 360         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 361         if self.output_name:
 362             command.append("--output-name=" + self.output_name)
 363             container_req["output_name"] = self.output_name
 364
 365         if self.output_tags:
 366             command.append("--output-tags=" + self.output_tags)
 367
 368         if kwargs.get("debug"):
 369             command.append("--debug")
 370
 371         if self.enable_reuse:
 372             command.append("--enable-reuse")
 373         else:
 374             command.append("--disable-reuse")
 375
 376         if self.on_error:
 377             command.append("--on-error=" + self.on_error)
 378
 379         if self.intermediate_output_ttl:
 380             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 381
 382         if self.arvrunner.trash_intermediate:
 383             command.append("--trash-intermediate")
 384
 385         if self.arvrunner.project_uuid:
 386             command.append("--project-uuid="+self.arvrunner.project_uuid)
 387
 388         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 389
 390         container_req["command"] = command
 391
 392         return container_req
 393
 394
 395     def run(self, *args, **kwargs):
 396         kwargs["keepprefix"] = "keep:"
 397         job_spec = self.arvados_job_spec(*args, **kwargs)
 398         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 399
 400         response = self.arvrunner.api.container_requests().create(
 401             body=job_spec
 402         ).execute(num_retries=self.arvrunner.num_retries)
 403
 404         self.uuid = response["uuid"]
 405         self.arvrunner.processes[self.uuid] = self
 406
 407         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 408
 409         if response["state"] == "Final":
 410             self.done(response)
 411
 412     def done(self, record):
 413         try:
 414             container = self.arvrunner.api.containers().get(
 415                 uuid=record["container_uuid"]
 416             ).execute(num_retries=self.arvrunner.num_retries)
 417         except Exception as e:
 418             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 419             self.arvrunner.output_callback({}, "permanentFail")
 420         else:
 421             super(RunnerContainer, self).done(container)
 422         finally:
 423             if record["uuid"] in self.arvrunner.processes:
 424                 del self.arvrunner.processes[record["uuid"]]