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