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 get_feature, shortname, UnsupportedRequirement
  12 from cwltool.errors import WorkflowException
  13 from cwltool.draft2tool 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
  18 from schema_salad.sourceline import SourceLine
  19
  20 import ruamel.yaml as yaml
  21
  22 import arvados.collection
  23 from arvados.errors import ApiError
  24
  25 from .arvdocker import arv_docker_get_image
  26 from .runner import Runner, arvados_jobs_image, packed_workflow, upload_workflow_collection, trim_anonymous_location, remove_redundant_fields
  27 from .pathmapper import VwdPathMapper, trim_listing
  28 from .perf import Perf
  29 from . import done
  30 from ._version import __version__
  31
  32 logger = logging.getLogger('arvados.cwl-runner')
  33 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  34
  35 crunchrunner_re = re.compile(r"^.*crunchrunner: \$\(task\.(tmpdir|outdir|keep)\)=(.*)$")
  36
  37 crunchrunner_git_commit = 'a3f2cb186e437bfce0031b024b2157b73ed2717d'
  38
  39 class ArvadosJob(object):
  40     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  41
  42     def __init__(self, runner):
  43         self.arvrunner = runner
  44         self.running = False
  45         self.uuid = None
  46
  47     def run(self, dry_run=False, pull_image=True, **kwargs):
  48         script_parameters = {
  49             "command": self.command_line
  50         }
  51         runtime_constraints = {}
  52
  53         with Perf(metrics, "generatefiles %s" % self.name):
  54             if self.generatefiles["listing"]:
  55                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  56                                                     keep_client=self.arvrunner.keep_client,
  57                                                     num_retries=self.arvrunner.num_retries)
  58                 script_parameters["task.vwd"] = {}
  59                 generatemapper = VwdPathMapper([self.generatefiles], "", "",
  60                                                separateDirs=False)
  61
  62                 with Perf(metrics, "createfiles %s" % self.name):
  63                     for f, p in generatemapper.items():
  64                         if p.type == "CreateFile":
  65                             with vwd.open(p.target, "w") as n:
  66                                 n.write(p.resolved.encode("utf-8"))
  67
  68                 if vwd:
  69                     with Perf(metrics, "generatefiles.save_new %s" % self.name):
  70                         vwd.save_new()
  71
  72                 for f, p in generatemapper.items():
  73                     if p.type == "File":
  74                         script_parameters["task.vwd"][p.target] = p.resolved
  75                     if p.type == "CreateFile":
  76                         script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  77
  78         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  79         if self.environment:
  80             script_parameters["task.env"].update(self.environment)
  81
  82         if self.stdin:
  83             script_parameters["task.stdin"] = self.stdin
  84
  85         if self.stdout:
  86             script_parameters["task.stdout"] = self.stdout
  87
  88         if self.stderr:
  89             script_parameters["task.stderr"] = self.stderr
  90
  91         if self.successCodes:
  92             script_parameters["task.successCodes"] = self.successCodes
  93         if self.temporaryFailCodes:
  94             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  95         if self.permanentFailCodes:
  96             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  97
  98         with Perf(metrics, "arv_docker_get_image %s" % self.name):
  99             (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 100             if docker_req and kwargs.get("use_container") is not False:
 101                 if docker_req.get("dockerOutputDirectory"):
 102                     raise SourceLine(docker_req, "dockerOutputDirectory", UnsupportedRequirement).makeError(
 103                         "Option 'dockerOutputDirectory' of DockerRequirement not supported.")
 104                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
 105             else:
 106                 runtime_constraints["docker_image"] = "arvados/jobs"
 107
 108         resources = self.builder.resources
 109         if resources is not None:
 110             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
 111             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
 112             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
 113
 114         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 115         if runtime_req:
 116             if "keep_cache" in runtime_req:
 117                 runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
 118                 runtime_constraints["min_ram_mb_per_node"] += runtime_req["keep_cache"]
 119             if "outputDirType" in runtime_req:
 120                 if runtime_req["outputDirType"] == "local_output_dir":
 121                     script_parameters["task.keepTmpOutput"] = False
 122                 elif runtime_req["outputDirType"] == "keep_output_dir":
 123                     script_parameters["task.keepTmpOutput"] = True
 124
 125         filters = [["repository", "=", "arvados"],
 126                    ["script", "=", "crunchrunner"],
 127                    ["script_version", "in git", crunchrunner_git_commit]]
 128         if not self.arvrunner.ignore_docker_for_reuse:
 129             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 130
 131         enable_reuse = kwargs.get("enable_reuse", True)
 132         if enable_reuse:
 133             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 134             if reuse_req:
 135                 enable_reuse = reuse_req["enableReuse"]
 136
 137         try:
 138             with Perf(metrics, "create %s" % self.name):
 139                 response = self.arvrunner.api.jobs().create(
 140                     body={
 141                         "owner_uuid": self.arvrunner.project_uuid,
 142                         "script": "crunchrunner",
 143                         "repository": "arvados",
 144                         "script_version": "master",
 145                         "minimum_script_version": crunchrunner_git_commit,
 146                         "script_parameters": {"tasks": [script_parameters]},
 147                         "runtime_constraints": runtime_constraints
 148                     },
 149                     filters=filters,
 150                     find_or_create=enable_reuse
 151                 ).execute(num_retries=self.arvrunner.num_retries)
 152
 153             self.arvrunner.processes[response["uuid"]] = self
 154
 155             self.update_pipeline_component(response)
 156
 157             if response["state"] == "Complete":
 158                 logger.info("%s reused job %s", self.arvrunner.label(self), response["uuid"])
 159                 # Give read permission to the desired project on reused jobs
 160                 if response["owner_uuid"] != self.arvrunner.project_uuid:
 161                     try:
 162                         self.arvrunner.api.links().create(body={
 163                             'link_class': 'permission',
 164                             'name': 'can_read',
 165                             'tail_uuid': self.arvrunner.project_uuid,
 166                             'head_uuid': response["uuid"],
 167                             }).execute(num_retries=self.arvrunner.num_retries)
 168                     except ApiError as e:
 169                         # The user might not have "manage" access on the job: log
 170                         # a message and continue.
 171                         logger.info("Creating read permission on job %s: %s",
 172                                     response["uuid"],
 173                                     e)
 174
 175                 with Perf(metrics, "done %s" % self.name):
 176                     self.done(response)
 177             else:
 178                 logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 179         except Exception as e:
 180             logger.exception("%s error" % (self.arvrunner.label(self)))
 181             self.output_callback({}, "permanentFail")
 182
 183     def update_pipeline_component(self, record):
 184         if self.arvrunner.pipeline:
 185             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 186             with Perf(metrics, "update_pipeline_component %s" % self.name):
 187                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(
 188                     uuid=self.arvrunner.pipeline["uuid"],
 189                     body={
 190                         "components": self.arvrunner.pipeline["components"]
 191                     }).execute(num_retries=self.arvrunner.num_retries)
 192         if self.arvrunner.uuid:
 193             try:
 194                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 195                 if job:
 196                     components = job["components"]
 197                     components[self.name] = record["uuid"]
 198                     self.arvrunner.api.jobs().update(
 199                         uuid=self.arvrunner.uuid,
 200                         body={
 201                             "components": components
 202                         }).execute(num_retries=self.arvrunner.num_retries)
 203             except Exception as e:
 204                 logger.info("Error adding to components: %s", e)
 205
 206     def done(self, record):
 207         try:
 208             self.update_pipeline_component(record)
 209         except:
 210             pass
 211
 212         try:
 213             if record["state"] == "Complete":
 214                 processStatus = "success"
 215             else:
 216                 processStatus = "permanentFail"
 217
 218             outputs = {}
 219             try:
 220                 if record["output"]:
 221                     with Perf(metrics, "inspect log %s" % self.name):
 222                         logc = arvados.collection.CollectionReader(record["log"],
 223                                                                    api_client=self.arvrunner.api,
 224                                                                    keep_client=self.arvrunner.keep_client,
 225                                                                    num_retries=self.arvrunner.num_retries)
 226                         log = logc.open(logc.keys()[0])
 227                         dirs = {
 228                             "tmpdir": "/tmpdir",
 229                             "outdir": "/outdir",
 230                             "keep": "/keep"
 231                         }
 232                         for l in log:
 233                             # Determine the tmpdir, outdir and keep paths from
 234                             # the job run.  Unfortunately, we can't take the first
 235                             # values we find (which are expected to be near the
 236                             # top) and stop scanning because if the node fails and
 237                             # the job restarts on a different node these values
 238                             # will different runs, and we need to know about the
 239                             # final run that actually produced output.
 240                             g = crunchrunner_re.match(l)
 241                             if g:
 242                                 dirs[g.group(1)] = g.group(2)
 243
 244                     if processStatus == "permanentFail":
 245                         done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 246
 247                     with Perf(metrics, "output collection %s" % self.name):
 248                         outputs = done.done(self, record, dirs["tmpdir"],
 249                                             dirs["outdir"], dirs["keep"])
 250             except WorkflowException as e:
 251                 logger.error("%s unable to collect output from %s:\n%s",
 252                              self.arvrunner.label(self), record["output"], e, exc_info=(e if self.arvrunner.debug else False))
 253                 processStatus = "permanentFail"
 254             except Exception as e:
 255                 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
 256                 processStatus = "permanentFail"
 257
 258             # Note: Currently, on error output_callback is expecting an empty dict,
 259             # anything else will fail.
 260             if not isinstance(outputs, dict):
 261                 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
 262                 outputs = {}
 263                 processStatus = "permanentFail"
 264         finally:
 265             self.output_callback(outputs, processStatus)
 266             if record["uuid"] in self.arvrunner.processes:
 267                 del self.arvrunner.processes[record["uuid"]]
 268
 269 class RunnerJob(Runner):
 270     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 271
 272     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 273         """Create an Arvados job specification for this workflow.
 274
 275         The returned dict can be used to create a job (i.e., passed as
 276         the +body+ argument to jobs().create()), or as a component in
 277         a pipeline template or pipeline instance.
 278         """
 279
 280         if self.tool.tool["id"].startswith("keep:"):
 281             self.job_order["cwl:tool"] = self.tool.tool["id"][5:]
 282         else:
 283             packed = packed_workflow(self.arvrunner, self.tool)
 284             wf_pdh = upload_workflow_collection(self.arvrunner, self.name, packed)
 285             self.job_order["cwl:tool"] = "%s/workflow.cwl#main" % wf_pdh
 286
 287         adjustDirObjs(self.job_order, trim_listing)
 288         visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location)
 289         visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields)
 290
 291         if self.output_name:
 292             self.job_order["arv:output_name"] = self.output_name
 293
 294         if self.output_tags:
 295             self.job_order["arv:output_tags"] = self.output_tags
 296
 297         self.job_order["arv:enable_reuse"] = self.enable_reuse
 298
 299         if self.on_error:
 300             self.job_order["arv:on_error"] = self.on_error
 301
 302         if kwargs.get("debug"):
 303             self.job_order["arv:debug"] = True
 304
 305         return {
 306             "script": "cwl-runner",
 307             "script_version": "master",
 308             "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9",
 309             "repository": "arvados",
 310             "script_parameters": self.job_order,
 311             "runtime_constraints": {
 312                 "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 313                 "min_ram_mb_per_node": self.submit_runner_ram
 314             }
 315         }
 316
 317     def run(self, *args, **kwargs):
 318         job_spec = self.arvados_job_spec(*args, **kwargs)
 319
 320         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 321
 322         job = self.arvrunner.api.jobs().create(
 323             body=job_spec,
 324             find_or_create=self.enable_reuse
 325         ).execute(num_retries=self.arvrunner.num_retries)
 326
 327         for k,v in job_spec["script_parameters"].items():
 328             if v is False or v is None or isinstance(v, dict):
 329                 job_spec["script_parameters"][k] = {"value": v}
 330
 331         del job_spec["owner_uuid"]
 332         job_spec["job"] = job
 333
 334         instance_spec = {
 335             "owner_uuid": self.arvrunner.project_uuid,
 336             "name": self.name,
 337             "components": {
 338                 "cwl-runner": job_spec,
 339             },
 340             "state": "RunningOnServer",
 341         }
 342         if not self.enable_reuse:
 343             instance_spec["properties"] = {"run_options": {"enable_job_reuse": False}}
 344
 345         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 346             body=instance_spec).execute(num_retries=self.arvrunner.num_retries)
 347         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 348
 349         if kwargs.get("wait") is False:
 350             self.uuid = self.arvrunner.pipeline["uuid"]
 351             return
 352
 353         self.uuid = job["uuid"]
 354         self.arvrunner.processes[self.uuid] = self
 355
 356         if job["state"] in ("Complete", "Failed", "Cancelled"):
 357             self.done(job)
 358
 359
 360 class RunnerTemplate(object):
 361     """An Arvados pipeline template that invokes a CWL workflow."""
 362
 363     type_to_dataclass = {
 364         'boolean': 'boolean',
 365         'File': 'File',
 366         'Directory': 'Collection',
 367         'float': 'number',
 368         'int': 'number',
 369         'string': 'text',
 370     }
 371
 372     def __init__(self, runner, tool, job_order, enable_reuse, uuid,
 373                  submit_runner_ram=0, name=None):
 374         self.runner = runner
 375         self.tool = tool
 376         self.job = RunnerJob(
 377             runner=runner,
 378             tool=tool,
 379             job_order=job_order,
 380             enable_reuse=enable_reuse,
 381             output_name=None,
 382             output_tags=None,
 383             submit_runner_ram=submit_runner_ram,
 384             name=name)
 385         self.uuid = uuid
 386
 387     def pipeline_component_spec(self):
 388         """Return a component that Workbench and a-r-p-i will understand.
 389
 390         Specifically, translate CWL input specs to Arvados pipeline
 391         format, like {"dataclass":"File","value":"xyz"}.
 392         """
 393
 394         spec = self.job.arvados_job_spec()
 395
 396         # Most of the component spec is exactly the same as the job
 397         # spec (script, script_version, etc.).
 398         # spec['script_parameters'] isn't right, though. A component
 399         # spec's script_parameters hash is a translation of
 400         # self.tool.tool['inputs'] with defaults/overrides taken from
 401         # the job order. So we move the job parameters out of the way
 402         # and build a new spec['script_parameters'].
 403         job_params = spec['script_parameters']
 404         spec['script_parameters'] = {}
 405
 406         for param in self.tool.tool['inputs']:
 407             param = copy.deepcopy(param)
 408
 409             # Data type and "required" flag...
 410             types = param['type']
 411             if not isinstance(types, list):
 412                 types = [types]
 413             param['required'] = 'null' not in types
 414             non_null_types = [t for t in types if t != "null"]
 415             if len(non_null_types) == 1:
 416                 the_type = [c for c in non_null_types][0]
 417                 dataclass = None
 418                 if isinstance(the_type, basestring):
 419                     dataclass = self.type_to_dataclass.get(the_type)
 420                 if dataclass:
 421                     param['dataclass'] = dataclass
 422             # Note: If we didn't figure out a single appropriate
 423             # dataclass, we just left that attribute out.  We leave
 424             # the "type" attribute there in any case, which might help
 425             # downstream.
 426
 427             # Title and description...
 428             title = param.pop('label', '')
 429             descr = param.pop('doc', '').rstrip('\n')
 430             if title:
 431                 param['title'] = title
 432             if descr:
 433                 param['description'] = descr
 434
 435             # Fill in the value from the current job order, if any.
 436             param_id = shortname(param.pop('id'))
 437             value = job_params.get(param_id)
 438             if value is None:
 439                 pass
 440             elif not isinstance(value, dict):
 441                 param['value'] = value
 442             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 443                 param['value'] = value['location'][5:]
 444
 445             spec['script_parameters'][param_id] = param
 446         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 447         return spec
 448
 449     def save(self):
 450         body = {
 451             "components": {
 452                 self.job.name: self.pipeline_component_spec(),
 453             },
 454             "name": self.job.name,
 455         }
 456         if self.runner.project_uuid:
 457             body["owner_uuid"] = self.runner.project_uuid
 458         if self.uuid:
 459             self.runner.api.pipeline_templates().update(
 460                 uuid=self.uuid, body=body).execute(
 461                     num_retries=self.runner.num_retries)
 462             logger.info("Updated template %s", self.uuid)
 463         else:
 464             self.uuid = self.runner.api.pipeline_templates().create(
 465                 body=body, ensure_unique_name=True).execute(
 466                     num_retries=self.runner.num_retries)['uuid']
 467             logger.info("Created template %s", self.uuid)