sdk/cwl/arvados_cwl/arvjob.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 import logging
   6 import re
   7 import copy
   8 import json
   9 import datetime
  10 import time
  11
  12 from cwltool.process import shortname, UnsupportedRequirement
  13 from cwltool.errors import WorkflowException
  14 from cwltool.command_line_tool import revmap_file, CommandLineTool
  15 from cwltool.load_tool import fetch_document
  16 from cwltool.builder import Builder
  17 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  18 from cwltool.job import JobBase
  19
  20 from schema_salad.sourceline import SourceLine
  21
  22 import ruamel.yaml as yaml
  23
  24 import arvados.collection
  25 from arvados.errors import ApiError
  26
  27 from .arvdocker import arv_docker_get_image
  28 from .runner import Runner, arvados_jobs_image, packed_workflow, upload_workflow_collection, trim_anonymous_location, remove_redundant_fields
  29 from .pathmapper import VwdPathMapper, trim_listing
  30 from .perf import Perf
  31 from . import done
  32 from ._version import __version__
  33
  34 logger = logging.getLogger('arvados.cwl-runner')
  35 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  36
  37 crunchrunner_re = re.compile(r"^.*crunchrunner: \$\(task\.(tmpdir|outdir|keep)\)=(.*)$")
  38
  39 crunchrunner_git_commit = 'a3f2cb186e437bfce0031b024b2157b73ed2717d'
  40
  41 class ArvadosJob(JobBase):
  42     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  43
  44     def __init__(self, runner,
  45                  builder,   # type: Builder
  46                  joborder,  # type: Dict[Text, Union[Dict[Text, Any], List, Text]]
  47                  make_path_mapper,  # type: Callable[..., PathMapper]
  48                  requirements,      # type: List[Dict[Text, Text]]
  49                  hints,     # type: List[Dict[Text, Text]]
  50                  name       # type: Text
  51     ):
  52         super(ArvadosJob, self).__init__(builder, joborder, make_path_mapper, requirements, hints, name)
  53         self.arvrunner = runner
  54         self.running = False
  55         self.uuid = None
  56
  57     def run(self, runtimeContext):
  58         script_parameters = {
  59             "command": self.command_line
  60         }
  61         runtime_constraints = {}
  62
  63         with Perf(metrics, "generatefiles %s" % self.name):
  64             if self.generatefiles["listing"]:
  65                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  66                                                     keep_client=self.arvrunner.keep_client,
  67                                                     num_retries=self.arvrunner.num_retries)
  68                 script_parameters["task.vwd"] = {}
  69                 generatemapper = VwdPathMapper([self.generatefiles], "", "",
  70                                                separateDirs=False)
  71
  72                 with Perf(metrics, "createfiles %s" % self.name):
  73                     for f, p in generatemapper.items():
  74                         if p.type == "CreateFile":
  75                             with vwd.open(p.target, "w") as n:
  76                                 n.write(p.resolved.encode("utf-8"))
  77
  78                 if vwd:
  79                     with Perf(metrics, "generatefiles.save_new %s" % self.name):
  80                         info = self._get_intermediate_collection_info()
  81                         vwd.save_new(name=info["name"],
  82                                      ensure_unique_name=True,
  83                                      trash_at=info["trash_at"],
  84                                      properties=info["properties"])
  85
  86                 for f, p in generatemapper.items():
  87                     if p.type == "File":
  88                         script_parameters["task.vwd"][p.target] = p.resolved
  89                     if p.type == "CreateFile":
  90                         script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  91
  92         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  93         if self.environment:
  94             script_parameters["task.env"].update(self.environment)
  95
  96         if self.stdin:
  97             script_parameters["task.stdin"] = self.stdin
  98
  99         if self.stdout:
 100             script_parameters["task.stdout"] = self.stdout
 101
 102         if self.stderr:
 103             script_parameters["task.stderr"] = self.stderr
 104
 105         if self.successCodes:
 106             script_parameters["task.successCodes"] = self.successCodes
 107         if self.temporaryFailCodes:
 108             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
 109         if self.permanentFailCodes:
 110             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
 111
 112         with Perf(metrics, "arv_docker_get_image %s" % self.name):
 113             (docker_req, docker_is_req) = self.get_requirement("DockerRequirement")
 114             if docker_req and runtimeContext.use_container is not False:
 115                 if docker_req.get("dockerOutputDirectory"):
 116                     raise SourceLine(docker_req, "dockerOutputDirectory", UnsupportedRequirement).makeError(
 117                         "Option 'dockerOutputDirectory' of DockerRequirement not supported.")
 118                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api,
 119                                                                            docker_req,
 120                                                                            runtimeContext.pull_image,
 121                                                                            self.arvrunner.project_uuid)
 122             else:
 123                 runtime_constraints["docker_image"] = "arvados/jobs"
 124
 125         resources = self.builder.resources
 126         if resources is not None:
 127             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
 128             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
 129             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
 130
 131         runtime_req, _ = self.get_requirement("http://arvados.org/cwl#RuntimeConstraints")
 132         if runtime_req:
 133             if "keep_cache" in runtime_req:
 134                 runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
 135                 runtime_constraints["min_ram_mb_per_node"] += runtime_req["keep_cache"]
 136             if "outputDirType" in runtime_req:
 137                 if runtime_req["outputDirType"] == "local_output_dir":
 138                     script_parameters["task.keepTmpOutput"] = False
 139                 elif runtime_req["outputDirType"] == "keep_output_dir":
 140                     script_parameters["task.keepTmpOutput"] = True
 141
 142         filters = [["repository", "=", "arvados"],
 143                    ["script", "=", "crunchrunner"],
 144                    ["script_version", "in git", crunchrunner_git_commit]]
 145         if not self.arvrunner.ignore_docker_for_reuse:
 146             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 147
 148         enable_reuse = runtimeContext.enable_reuse
 149         if enable_reuse:
 150             reuse_req, _ = self.get_requirement("http://arvados.org/cwl#ReuseRequirement")
 151             if reuse_req:
 152                 enable_reuse = reuse_req["enableReuse"]
 153
 154         self.output_callback = self.arvrunner.get_wrapped_callback(self.output_callback)
 155
 156         try:
 157             with Perf(metrics, "create %s" % self.name):
 158                 response = self.arvrunner.api.jobs().create(
 159                     body={
 160                         "owner_uuid": self.arvrunner.project_uuid,
 161                         "script": "crunchrunner",
 162                         "repository": "arvados",
 163                         "script_version": "master",
 164                         "minimum_script_version": crunchrunner_git_commit,
 165                         "script_parameters": {"tasks": [script_parameters]},
 166                         "runtime_constraints": runtime_constraints
 167                     },
 168                     filters=filters,
 169                     find_or_create=enable_reuse
 170                 ).execute(num_retries=self.arvrunner.num_retries)
 171
 172             self.uuid = response["uuid"]
 173             self.arvrunner.process_submitted(self)
 174
 175             self.update_pipeline_component(response)
 176
 177             if response["state"] == "Complete":
 178                 logger.info("%s reused job %s", self.arvrunner.label(self), response["uuid"])
 179                 # Give read permission to the desired project on reused jobs
 180                 if response["owner_uuid"] != self.arvrunner.project_uuid:
 181                     try:
 182                         self.arvrunner.api.links().create(body={
 183                             'link_class': 'permission',
 184                             'name': 'can_read',
 185                             'tail_uuid': self.arvrunner.project_uuid,
 186                             'head_uuid': response["uuid"],
 187                             }).execute(num_retries=self.arvrunner.num_retries)
 188                     except ApiError as e:
 189                         # The user might not have "manage" access on the job: log
 190                         # a message and continue.
 191                         logger.info("Creating read permission on job %s: %s",
 192                                     response["uuid"],
 193                                     e)
 194             else:
 195                 logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 196         except Exception as e:
 197             logger.exception("%s error" % (self.arvrunner.label(self)))
 198             self.output_callback({}, "permanentFail")
 199
 200     def update_pipeline_component(self, record):
 201         with self.arvrunner.workflow_eval_lock:
 202             if self.arvrunner.pipeline:
 203                 self.arvrunner.pipeline["components"][self.name] = {"job": record}
 204                 with Perf(metrics, "update_pipeline_component %s" % self.name):
 205                     self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(
 206                         uuid=self.arvrunner.pipeline["uuid"],
 207                         body={
 208                             "components": self.arvrunner.pipeline["components"]
 209                         }).execute(num_retries=self.arvrunner.num_retries)
 210             if self.arvrunner.uuid:
 211                 try:
 212                     job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 213                     if job:
 214                         components = job["components"]
 215                         components[self.name] = record["uuid"]
 216                         self.arvrunner.api.jobs().update(
 217                             uuid=self.arvrunner.uuid,
 218                             body={
 219                                 "components": components
 220                             }).execute(num_retries=self.arvrunner.num_retries)
 221                 except Exception as e:
 222                     logger.info("Error adding to components: %s", e)
 223
 224     def done(self, record):
 225         try:
 226             self.update_pipeline_component(record)
 227         except:
 228             pass
 229
 230         try:
 231             if record["state"] == "Complete":
 232                 processStatus = "success"
 233             else:
 234                 processStatus = "permanentFail"
 235
 236             outputs = {}
 237             try:
 238                 if record["output"]:
 239                     with Perf(metrics, "inspect log %s" % self.name):
 240                         logc = arvados.collection.CollectionReader(record["log"],
 241                                                                    api_client=self.arvrunner.api,
 242                                                                    keep_client=self.arvrunner.keep_client,
 243                                                                    num_retries=self.arvrunner.num_retries)
 244                         log = logc.open(logc.keys()[0])
 245                         dirs = {
 246                             "tmpdir": "/tmpdir",
 247                             "outdir": "/outdir",
 248                             "keep": "/keep"
 249                         }
 250                         for l in log:
 251                             # Determine the tmpdir, outdir and keep paths from
 252                             # the job run.  Unfortunately, we can't take the first
 253                             # values we find (which are expected to be near the
 254                             # top) and stop scanning because if the node fails and
 255                             # the job restarts on a different node these values
 256                             # will different runs, and we need to know about the
 257                             # final run that actually produced output.
 258                             g = crunchrunner_re.match(l)
 259                             if g:
 260                                 dirs[g.group(1)] = g.group(2)
 261
 262                     if processStatus == "permanentFail":
 263                         done.logtail(logc, logger.error, "%s (%s) error log:" % (self.arvrunner.label(self), record["uuid"]), maxlen=40)
 264
 265                     with Perf(metrics, "output collection %s" % self.name):
 266                         outputs = done.done(self, record, dirs["tmpdir"],
 267                                             dirs["outdir"], dirs["keep"])
 268             except WorkflowException as e:
 269                 logger.error("%s unable to collect output from %s:\n%s",
 270                              self.arvrunner.label(self), record["output"], e, exc_info=(e if self.arvrunner.debug else False))
 271                 processStatus = "permanentFail"
 272             except Exception as e:
 273                 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
 274                 processStatus = "permanentFail"
 275
 276             # Note: Currently, on error output_callback is expecting an empty dict,
 277             # anything else will fail.
 278             if not isinstance(outputs, dict):
 279                 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
 280                 outputs = {}
 281                 processStatus = "permanentFail"
 282         finally:
 283             self.output_callback(outputs, processStatus)
 284
 285     def _get_intermediate_collection_info(self):
 286             trash_time = None
 287             if self.arvrunner.intermediate_output_ttl > 0:
 288                 trash_time = datetime.datetime.now() + datetime.timedelta(seconds=self.arvrunner.intermediate_output_ttl)
 289
 290             current_container_uuid = None
 291             try:
 292                 current_container = self.arvrunner.api.containers().current().execute(num_retries=self.arvrunner.num_retries)
 293                 current_container_uuid = current_container['uuid']
 294             except ApiError as e:
 295                 # Status code 404 just means we're not running in a container.
 296                 if e.resp.status != 404:
 297                     logger.info("Getting current container: %s", e)
 298             props = {"type": "Intermediate",
 299                           "container": current_container_uuid}
 300
 301             return {"name" : "Intermediate collection",
 302                     "trash_at" : trash_time,
 303                     "properties" : props}
 304
 305
 306 class RunnerJob(Runner):
 307     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 308
 309     def arvados_job_spec(self, debug=False):
 310         """Create an Arvados job specification for this workflow.
 311
 312         The returned dict can be used to create a job (i.e., passed as
 313         the +body+ argument to jobs().create()), or as a component in
 314         a pipeline template or pipeline instance.
 315         """
 316
 317         if self.tool.tool["id"].startswith("keep:"):
 318             self.job_order["cwl:tool"] = self.tool.tool["id"][5:]
 319         else:
 320             packed = packed_workflow(self.arvrunner, self.tool, self.merged_map)
 321             wf_pdh = upload_workflow_collection(self.arvrunner, self.name, packed)
 322             self.job_order["cwl:tool"] = "%s/workflow.cwl#main" % wf_pdh
 323
 324         adjustDirObjs(self.job_order, trim_listing)
 325         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 326         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 327
 328         if self.output_name:
 329             self.job_order["arv:output_name"] = self.output_name
 330
 331         if self.output_tags:
 332             self.job_order["arv:output_tags"] = self.output_tags
 333
 334         self.job_order["arv:enable_reuse"] = self.enable_reuse
 335
 336         if self.on_error:
 337             self.job_order["arv:on_error"] = self.on_error
 338
 339         if debug:
 340             self.job_order["arv:debug"] = True
 341
 342         return {
 343             "script": "cwl-runner",
 344             "script_version": "master",
 345             "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9",
 346             "repository": "arvados",
 347             "script_parameters": self.job_order,
 348             "runtime_constraints": {
 349                 "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 350                 "min_ram_mb_per_node": self.submit_runner_ram
 351             }
 352         }
 353
 354     def run(self, runtimeContext):
 355         job_spec = self.arvados_job_spec(runtimeContext.debug)
 356
 357         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 358
 359         job = self.arvrunner.api.jobs().create(
 360             body=job_spec,
 361             find_or_create=self.enable_reuse
 362         ).execute(num_retries=self.arvrunner.num_retries)
 363
 364         for k,v in job_spec["script_parameters"].items():
 365             if v is False or v is None or isinstance(v, dict):
 366                 job_spec["script_parameters"][k] = {"value": v}
 367
 368         del job_spec["owner_uuid"]
 369         job_spec["job"] = job
 370
 371         instance_spec = {
 372             "owner_uuid": self.arvrunner.project_uuid,
 373             "name": self.name,
 374             "components": {
 375                 "cwl-runner": job_spec,
 376             },
 377             "state": "RunningOnServer",
 378         }
 379         if not self.enable_reuse:
 380             instance_spec["properties"] = {"run_options": {"enable_job_reuse": False}}
 381
 382         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 383             body=instance_spec).execute(num_retries=self.arvrunner.num_retries)
 384         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 385
 386         if runtimeContext.wait is False:
 387             self.uuid = self.arvrunner.pipeline["uuid"]
 388             return
 389
 390         self.uuid = job["uuid"]
 391         self.arvrunner.process_submitted(self)
 392
 393
 394 class RunnerTemplate(object):
 395     """An Arvados pipeline template that invokes a CWL workflow."""
 396
 397     type_to_dataclass = {
 398         'boolean': 'boolean',
 399         'File': 'File',
 400         'Directory': 'Collection',
 401         'float': 'number',
 402         'int': 'number',
 403         'string': 'text',
 404     }
 405
 406     def __init__(self, runner, tool, job_order, enable_reuse, uuid,
 407                  submit_runner_ram=0, name=None, merged_map=None):
 408         self.runner = runner
 409         self.tool = tool
 410         self.job = RunnerJob(
 411             runner=runner,
 412             tool=tool,
 413             job_order=job_order,
 414             enable_reuse=enable_reuse,
 415             output_name=None,
 416             output_tags=None,
 417             submit_runner_ram=submit_runner_ram,
 418             name=name,
 419             merged_map=merged_map)
 420         self.uuid = uuid
 421
 422     def pipeline_component_spec(self):
 423         """Return a component that Workbench and a-r-p-i will understand.
 424
 425         Specifically, translate CWL input specs to Arvados pipeline
 426         format, like {"dataclass":"File","value":"xyz"}.
 427         """
 428
 429         spec = self.job.arvados_job_spec()
 430
 431         # Most of the component spec is exactly the same as the job
 432         # spec (script, script_version, etc.).
 433         # spec['script_parameters'] isn't right, though. A component
 434         # spec's script_parameters hash is a translation of
 435         # self.tool.tool['inputs'] with defaults/overrides taken from
 436         # the job order. So we move the job parameters out of the way
 437         # and build a new spec['script_parameters'].
 438         job_params = spec['script_parameters']
 439         spec['script_parameters'] = {}
 440
 441         for param in self.tool.tool['inputs']:
 442             param = copy.deepcopy(param)
 443
 444             # Data type and "required" flag...
 445             types = param['type']
 446             if not isinstance(types, list):
 447                 types = [types]
 448             param['required'] = 'null' not in types
 449             non_null_types = [t for t in types if t != "null"]
 450             if len(non_null_types) == 1:
 451                 the_type = [c for c in non_null_types][0]
 452                 dataclass = None
 453                 if isinstance(the_type, basestring):
 454                     dataclass = self.type_to_dataclass.get(the_type)
 455                 if dataclass:
 456                     param['dataclass'] = dataclass
 457             # Note: If we didn't figure out a single appropriate
 458             # dataclass, we just left that attribute out.  We leave
 459             # the "type" attribute there in any case, which might help
 460             # downstream.
 461
 462             # Title and description...
 463             title = param.pop('label', '')
 464             descr = param.pop('doc', '').rstrip('\n')
 465             if title:
 466                 param['title'] = title
 467             if descr:
 468                 param['description'] = descr
 469
 470             # Fill in the value from the current job order, if any.
 471             param_id = shortname(param.pop('id'))
 472             value = job_params.get(param_id)
 473             if value is None:
 474                 pass
 475             elif not isinstance(value, dict):
 476                 param['value'] = value
 477             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 478                 param['value'] = value['location'][5:]
 479
 480             spec['script_parameters'][param_id] = param
 481         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 482         return spec
 483
 484     def save(self):
 485         body = {
 486             "components": {
 487                 self.job.name: self.pipeline_component_spec(),
 488             },
 489             "name": self.job.name,
 490         }
 491         if self.runner.project_uuid:
 492             body["owner_uuid"] = self.runner.project_uuid
 493         if self.uuid:
 494             self.runner.api.pipeline_templates().update(
 495                 uuid=self.uuid, body=body).execute(
 496                     num_retries=self.runner.num_retries)
 497             logger.info("Updated template %s", self.uuid)
 498         else:
 499             self.uuid = self.runner.api.pipeline_templates().create(
 500                 body=body, ensure_unique_name=True).execute(
 501                     num_retries=self.runner.num_retries)['uuid']
 502             logger.info("Created template %s", self.uuid)