sdk/cwl/arvados_cwl/arvcontainer.py

   1 import logging
   2 import json
   3 import os
   4 import urllib
   5 import time
   6 import datetime
   7 import ciso8601
   8
   9 import ruamel.yaml as yaml
  10
  11 from cwltool.errors import WorkflowException
  12 from cwltool.process import get_feature, UnsupportedRequirement, shortname
  13 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
  14 from cwltool.utils import aslist
  15
  16 import arvados.collection
  17
  18 from .arvdocker import arv_docker_get_image
  19 from . import done
  20 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location
  21 from .fsaccess import CollectionFetcher
  22 from .pathmapper import NoFollowPathMapper, trim_listing
  23 from .perf import Perf
  24
  25 logger = logging.getLogger('arvados.cwl-runner')
  26 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  27
  28 class ArvadosContainer(object):
  29     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  30
  31     def __init__(self, runner):
  32         self.arvrunner = runner
  33         self.running = False
  34         self.uuid = None
  35
  36     def update_pipeline_component(self, r):
  37         pass
  38
  39     def run(self, dry_run=False, pull_image=True, **kwargs):
  40         container_request = {
  41             "command": self.command_line,
  42             "owner_uuid": self.arvrunner.project_uuid,
  43             "name": self.name,
  44             "output_path": self.outdir,
  45             "cwd": self.outdir,
  46             "priority": 1,
  47             "state": "Committed",
  48             "properties": {},
  49         }
  50         runtime_constraints = {}
  51
  52         resources = self.builder.resources
  53         if resources is not None:
  54             runtime_constraints["vcpus"] = resources.get("cores", 1)
  55             runtime_constraints["ram"] = resources.get("ram") * 2**20
  56
  57         mounts = {
  58             self.outdir: {
  59                 "kind": "tmp",
  60                 "capacity": resources.get("outdirSize", 0) * 2**20
  61             },
  62             self.tmpdir: {
  63                 "kind": "tmp",
  64                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  65             }
  66         }
  67         scheduling_parameters = {}
  68
  69         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  70         rf.sort(key=lambda k: k.resolved)
  71         prevdir = None
  72         for resolved, target, tp, stg in rf:
  73             if not stg:
  74                 continue
  75             if prevdir and target.startswith(prevdir):
  76                 continue
  77             if tp == "Directory":
  78                 targetdir = target
  79             else:
  80                 targetdir = os.path.dirname(target)
  81             sp = resolved.split("/", 1)
  82             pdh = sp[0][5:]   # remove "keep:"
  83             mounts[targetdir] = {
  84                 "kind": "collection",
  85                 "portable_data_hash": pdh
  86             }
  87             if len(sp) == 2:
  88                 if tp == "Directory":
  89                     path = sp[1]
  90                 else:
  91                     path = os.path.dirname(sp[1])
  92                 if path and path != "/":
  93                     mounts[targetdir]["path"] = path
  94             prevdir = targetdir + "/"
  95
  96         with Perf(metrics, "generatefiles %s" % self.name):
  97             if self.generatefiles["listing"]:
  98                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  99                                                     keep_client=self.arvrunner.keep_client,
 100                                                     num_retries=self.arvrunner.num_retries)
 101                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
 102                                                     separateDirs=False)
 103
 104                 with Perf(metrics, "createfiles %s" % self.name):
 105                     for f, p in generatemapper.items():
 106                         if not p.target:
 107                             pass
 108                         elif p.type in ("File", "Directory"):
 109                             source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 110                             vwd.copy(path, p.target, source_collection=source)
 111                         elif p.type == "CreateFile":
 112                             with vwd.open(p.target, "w") as n:
 113                                 n.write(p.resolved.encode("utf-8"))
 114
 115                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 116                     vwd.save_new()
 117
 118                 for f, p in generatemapper.items():
 119                     if not p.target:
 120                         continue
 121                     mountpoint = "%s/%s" % (self.outdir, p.target)
 122                     mounts[mountpoint] = {"kind": "collection",
 123                                           "portable_data_hash": vwd.portable_data_hash(),
 124                                           "path": p.target}
 125
 126         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 127         if self.environment:
 128             container_request["environment"].update(self.environment)
 129
 130         if self.stdin:
 131             sp = self.stdin[6:].split("/", 1)
 132             mounts["stdin"] = {"kind": "collection",
 133                                 "portable_data_hash": sp[0],
 134                                 "path": sp[1]}
 135
 136         if self.stderr:
 137             mounts["stderr"] = {"kind": "file",
 138                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 139
 140         if self.stdout:
 141             mounts["stdout"] = {"kind": "file",
 142                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 143
 144         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 145         if not docker_req:
 146             docker_req = {"dockerImageId": "arvados/jobs"}
 147
 148         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 149                                                                      docker_req,
 150                                                                      pull_image,
 151                                                                      self.arvrunner.project_uuid)
 152
 153         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 154         if api_req:
 155             runtime_constraints["API"] = True
 156
 157         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 158         if runtime_req:
 159             if "keep_cache" in runtime_req:
 160                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 161             if "outputDirType" in runtime_req:
 162                 if runtime_req["outputDirType"] == "local_output_dir":
 163                     # Currently the default behavior.
 164                     pass
 165                 elif runtime_req["outputDirType"] == "keep_output_dir":
 166                     mounts[self.outdir]= {
 167                         "kind": "collection",
 168                         "writable": True
 169                     }
 170
 171         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 172         if partition_req:
 173             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 174
 175         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 176         if intermediate_output_req:
 177             self.output_ttl = intermediate_output_req["outputTTL"]
 178         else:
 179             self.output_ttl = self.arvrunner.intermediate_output_ttl
 180
 181         if self.output_ttl < 0:
 182             raise WorkflowError("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 183
 184         container_request["output_ttl"] = self.output_ttl
 185         container_request["mounts"] = mounts
 186         container_request["runtime_constraints"] = runtime_constraints
 187         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 188         container_request["scheduling_parameters"] = scheduling_parameters
 189
 190         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 191             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 192             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 193             if container_request["name"] == "main":
 194                 container_request["name"] = wfrecord["name"]
 195             container_request["properties"]["template_uuid"] = wfuuid
 196
 197         try:
 198             response = self.arvrunner.api.container_requests().create(
 199                 body=container_request
 200             ).execute(num_retries=self.arvrunner.num_retries)
 201
 202             self.uuid = response["uuid"]
 203             self.arvrunner.processes[self.uuid] = self
 204
 205             if response["state"] == "Final":
 206                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 207                 self.done(response)
 208             else:
 209                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 210         except Exception as e:
 211             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 212             self.output_callback({}, "permanentFail")
 213
 214     def done(self, record):
 215         outputs = {}
 216         try:
 217             container = self.arvrunner.api.containers().get(
 218                 uuid=record["container_uuid"]
 219             ).execute(num_retries=self.arvrunner.num_retries)
 220             if container["state"] == "Complete":
 221                 rcode = container["exit_code"]
 222                 if self.successCodes and rcode in self.successCodes:
 223                     processStatus = "success"
 224                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 225                     processStatus = "temporaryFail"
 226                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 227                     processStatus = "permanentFail"
 228                 elif rcode == 0:
 229                     processStatus = "success"
 230                 else:
 231                     processStatus = "permanentFail"
 232             else:
 233                 processStatus = "permanentFail"
 234
 235             if processStatus == "permanentFail":
 236                 logc = arvados.collection.CollectionReader(container["log"],
 237                                                            api_client=self.arvrunner.api,
 238                                                            keep_client=self.arvrunner.keep_client,
 239                                                            num_retries=self.arvrunner.num_retries)
 240                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 241
 242             if record["output_uuid"]:
 243                 if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl:
 244                     # Compute the trash time to avoid requesting the collection record.
 245                     trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl)
 246                     aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else ""
 247                     orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else ""
 248                     oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else ""
 249                     logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % (
 250                         self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete))
 251                 self.arvrunner.add_intermediate_output(record["output_uuid"])
 252
 253             if container["output"]:
 254                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 255         except WorkflowException as e:
 256             logger.error("%s unable to collect output from %s:\n%s",
 257                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 258             processStatus = "permanentFail"
 259         except Exception as e:
 260             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 261             processStatus = "permanentFail"
 262         finally:
 263             self.output_callback(outputs, processStatus)
 264             if record["uuid"] in self.arvrunner.processes:
 265                 del self.arvrunner.processes[record["uuid"]]
 266
 267
 268 class RunnerContainer(Runner):
 269     """Submit and manage a container that runs arvados-cwl-runner."""
 270
 271     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 272         """Create an Arvados container request for this workflow.
 273
 274         The returned dict can be used to create a container passed as
 275         the +body+ argument to container_requests().create().
 276         """
 277
 278         adjustDirObjs(self.job_order, trim_listing)
 279         adjustFileObjs(self.job_order, trim_anonymous_location)
 280         adjustDirObjs(self.job_order, trim_anonymous_location)
 281
 282         container_req = {
 283             "owner_uuid": self.arvrunner.project_uuid,
 284             "name": self.name,
 285             "output_path": "/var/spool/cwl",
 286             "cwd": "/var/spool/cwl",
 287             "priority": 1,
 288             "state": "Committed",
 289             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 290             "mounts": {
 291                 "/var/lib/cwl/cwl.input.json": {
 292                     "kind": "json",
 293                     "content": self.job_order
 294                 },
 295                 "stdout": {
 296                     "kind": "file",
 297                     "path": "/var/spool/cwl/cwl.output.json"
 298                 },
 299                 "/var/spool/cwl": {
 300                     "kind": "collection",
 301                     "writable": True
 302                 }
 303             },
 304             "runtime_constraints": {
 305                 "vcpus": 1,
 306                 "ram": 1024*1024 * self.submit_runner_ram,
 307                 "API": True
 308             },
 309             "properties": {}
 310         }
 311
 312         if self.tool.tool.get("id", "").startswith("keep:"):
 313             sp = self.tool.tool["id"].split('/')
 314             workflowcollection = sp[0][5:]
 315             workflowname = "/".join(sp[1:])
 316             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 317             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 318                 "kind": "collection",
 319                 "portable_data_hash": "%s" % workflowcollection
 320             }
 321         else:
 322             packed = packed_workflow(self.arvrunner, self.tool)
 323             workflowpath = "/var/lib/cwl/workflow.json#main"
 324             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 325                 "kind": "json",
 326                 "content": packed
 327             }
 328             if self.tool.tool.get("id", "").startswith("arvwf:"):
 329                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 330
 331         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 332         if self.output_name:
 333             command.append("--output-name=" + self.output_name)
 334             container_req["output_name"] = self.output_name
 335
 336         if self.output_tags:
 337             command.append("--output-tags=" + self.output_tags)
 338
 339         if kwargs.get("debug"):
 340             command.append("--debug")
 341
 342         if self.enable_reuse:
 343             command.append("--enable-reuse")
 344         else:
 345             command.append("--disable-reuse")
 346
 347         if self.on_error:
 348             command.append("--on-error=" + self.on_error)
 349
 350         if self.intermediate_output_ttl:
 351             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 352
 353         if self.arvrunner.trash_intermediate:
 354             command.append("--trash-intermediate")
 355
 356         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 357
 358         container_req["command"] = command
 359
 360         return container_req
 361
 362
 363     def run(self, *args, **kwargs):
 364         kwargs["keepprefix"] = "keep:"
 365         job_spec = self.arvados_job_spec(*args, **kwargs)
 366         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 367
 368         response = self.arvrunner.api.container_requests().create(
 369             body=job_spec
 370         ).execute(num_retries=self.arvrunner.num_retries)
 371
 372         self.uuid = response["uuid"]
 373         self.arvrunner.processes[self.uuid] = self
 374
 375         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 376
 377         if response["state"] == "Final":
 378             self.done(response)
 379
 380     def done(self, record):
 381         try:
 382             container = self.arvrunner.api.containers().get(
 383                 uuid=record["container_uuid"]
 384             ).execute(num_retries=self.arvrunner.num_retries)
 385         except Exception as e:
 386             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 387             self.arvrunner.output_callback({}, "permanentFail")
 388         else:
 389             super(RunnerContainer, self).done(container)
 390         finally:
 391             if record["uuid"] in self.arvrunner.processes:
 392                 del self.arvrunner.processes[record["uuid"]]