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