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