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