sdk/cwl/arvados_cwl/arvjob.py

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