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