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