sdk/cwl/arvados_cwl/arvjob.py

   1 import logging
   2 import re
   3 import copy
   4 import json
   5 import time
   6
   7 from cwltool.process import get_feature, shortname, UnsupportedRequirement
   8 from cwltool.errors import WorkflowException
   9 from cwltool.draft2tool import revmap_file, CommandLineTool
  10 from cwltool.load_tool import fetch_document
  11 from cwltool.builder import Builder
  12 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
  13
  14 from schema_salad.sourceline import SourceLine
  15
  16 import ruamel.yaml as yaml
  17
  18 import arvados.collection
  19 from arvados.errors import ApiError
  20
  21 from .arvdocker import arv_docker_get_image
  22 from .runner import Runner, arvados_jobs_image, packed_workflow, upload_workflow_collection, trim_anonymous_location
  23 from .pathmapper import VwdPathMapper, trim_listing
  24 from .perf import Perf
  25 from . import done
  26 from ._version import __version__
  27
  28 logger = logging.getLogger('arvados.cwl-runner')
  29 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  30
  31 crunchrunner_re = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.(tmpdir|outdir|keep)\)=(.*)")
  32
  33 crunchrunner_git_commit = 'a3f2cb186e437bfce0031b024b2157b73ed2717d'
  34
  35 class ArvadosJob(object):
  36     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  37
  38     def __init__(self, runner):
  39         self.arvrunner = runner
  40         self.running = False
  41         self.uuid = None
  42
  43     def run(self, dry_run=False, pull_image=True, **kwargs):
  44         script_parameters = {
  45             "command": self.command_line
  46         }
  47         runtime_constraints = {}
  48
  49         with Perf(metrics, "generatefiles %s" % self.name):
  50             if self.generatefiles["listing"]:
  51                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  52                                                     keep_client=self.arvrunner.keep_client,
  53                                                     num_retries=self.arvrunner.num_retries)
  54                 script_parameters["task.vwd"] = {}
  55                 generatemapper = VwdPathMapper([self.generatefiles], "", "",
  56                                                separateDirs=False)
  57
  58                 with Perf(metrics, "createfiles %s" % self.name):
  59                     for f, p in generatemapper.items():
  60                         if p.type == "CreateFile":
  61                             with vwd.open(p.target, "w") as n:
  62                                 n.write(p.resolved.encode("utf-8"))
  63
  64                 if vwd:
  65                     with Perf(metrics, "generatefiles.save_new %s" % self.name):
  66                         vwd.save_new()
  67
  68                 for f, p in generatemapper.items():
  69                     if p.type == "File":
  70                         script_parameters["task.vwd"][p.target] = p.resolved
  71                     if p.type == "CreateFile":
  72                         script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  73
  74         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  75         if self.environment:
  76             script_parameters["task.env"].update(self.environment)
  77
  78         if self.stdin:
  79             script_parameters["task.stdin"] = self.stdin
  80
  81         if self.stdout:
  82             script_parameters["task.stdout"] = self.stdout
  83
  84         if self.stderr:
  85             script_parameters["task.stderr"] = self.stderr
  86
  87         if self.successCodes:
  88             script_parameters["task.successCodes"] = self.successCodes
  89         if self.temporaryFailCodes:
  90             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  91         if self.permanentFailCodes:
  92             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  93
  94         with Perf(metrics, "arv_docker_get_image %s" % self.name):
  95             (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  96             if docker_req and kwargs.get("use_container") is not False:
  97                 if docker_req.get("dockerOutputDirectory"):
  98                     raise SourceLine(docker_req, "dockerOutputDirectory", UnsupportedRequirement).makeError(
  99                         "Option 'dockerOutputDirectory' of DockerRequirement not supported.")
 100                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
 101             else:
 102                 runtime_constraints["docker_image"] = "arvados/jobs"
 103
 104         resources = self.builder.resources
 105         if resources is not None:
 106             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
 107             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
 108             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
 109
 110         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 111         if runtime_req:
 112             if "keep_cache" in runtime_req:
 113                 runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
 114                 runtime_constraints["min_ram_mb_per_node"] += runtime_req["keep_cache"]
 115             if "outputDirType" in runtime_req:
 116                 if runtime_req["outputDirType"] == "local_output_dir":
 117                     script_parameters["task.keepTmpOutput"] = False
 118                 elif runtime_req["outputDirType"] == "keep_output_dir":
 119                     script_parameters["task.keepTmpOutput"] = True
 120
 121         filters = [["repository", "=", "arvados"],
 122                    ["script", "=", "crunchrunner"],
 123                    ["script_version", "in git", crunchrunner_git_commit]]
 124         if not self.arvrunner.ignore_docker_for_reuse:
 125             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 126
 127         try:
 128             with Perf(metrics, "create %s" % self.name):
 129                 response = self.arvrunner.api.jobs().create(
 130                     body={
 131                         "owner_uuid": self.arvrunner.project_uuid,
 132                         "script": "crunchrunner",
 133                         "repository": "arvados",
 134                         "script_version": "master",
 135                         "minimum_script_version": crunchrunner_git_commit,
 136                         "script_parameters": {"tasks": [script_parameters]},
 137                         "runtime_constraints": runtime_constraints
 138                     },
 139                     filters=filters,
 140                     find_or_create=kwargs.get("enable_reuse", True)
 141                 ).execute(num_retries=self.arvrunner.num_retries)
 142
 143             self.arvrunner.processes[response["uuid"]] = self
 144
 145             self.update_pipeline_component(response)
 146
 147             if response["state"] == "Complete":
 148                 logger.info("%s reused job %s", self.arvrunner.label(self), response["uuid"])
 149                 # Give read permission to the desired project on reused jobs
 150                 if response["owner_uuid"] != self.arvrunner.project_uuid:
 151                     try:
 152                         self.arvrunner.api.links().create(body={
 153                             'link_class': 'permission',
 154                             'name': 'can_read',
 155                             'tail_uuid': self.arvrunner.project_uuid,
 156                             'head_uuid': response["uuid"],
 157                             }).execute(num_retries=self.arvrunner.num_retries)
 158                     except ApiError as e:
 159                         # The user might not have "manage" access on the job: log
 160                         # a message and continue.
 161                         logger.info("Creating read permission on job %s: %s",
 162                                     response["uuid"],
 163                                     e)
 164
 165                 with Perf(metrics, "done %s" % self.name):
 166                     self.done(response)
 167             else:
 168                 logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 169         except Exception as e:
 170             logger.exception("%s error" % (self.arvrunner.label(self)))
 171             self.output_callback({}, "permanentFail")
 172
 173     def update_pipeline_component(self, record):
 174         if self.arvrunner.pipeline:
 175             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 176             with Perf(metrics, "update_pipeline_component %s" % self.name):
 177                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 178                                                                                  body={
 179                                                                                     "components": self.arvrunner.pipeline["components"]
 180                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 181         if self.arvrunner.uuid:
 182             try:
 183                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 184                 if job:
 185                     components = job["components"]
 186                     components[self.name] = record["uuid"]
 187                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 188                         body={
 189                             "components": components
 190                         }).execute(num_retries=self.arvrunner.num_retries)
 191             except Exception as e:
 192                 logger.info("Error adding to components: %s", e)
 193
 194     def done(self, record):
 195         try:
 196             self.update_pipeline_component(record)
 197         except:
 198             pass
 199
 200         try:
 201             if record["state"] == "Complete":
 202                 processStatus = "success"
 203             else:
 204                 processStatus = "permanentFail"
 205
 206             outputs = {}
 207             try:
 208                 if record["output"]:
 209                     with Perf(metrics, "inspect log %s" % self.name):
 210                         logc = arvados.collection.CollectionReader(record["log"],
 211                                                                    api_client=self.arvrunner.api,
 212                                                                    keep_client=self.arvrunner.keep_client,
 213                                                                    num_retries=self.arvrunner.num_retries)
 214                         log = logc.open(logc.keys()[0])
 215                         dirs = {}
 216                         tmpdir = None
 217                         outdir = None
 218                         keepdir = None
 219                         for l in log:
 220                             # Determine the tmpdir, outdir and keepdir paths from
 221                             # the job run.  Unfortunately, we can't take the first
 222                             # values we find (which are expected to be near the
 223                             # top) and stop scanning because if the node fails and
 224                             # the job restarts on a different node these values
 225                             # will different runs, and we need to know about the
 226                             # final run that actually produced output.
 227                             g = crunchrunner_re.match(l)
 228                             if g:
 229                                 dirs[g.group(1)] = g.group(2)
 230
 231                     if processStatus == "permanentFail":
 232                         done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 233
 234                     with Perf(metrics, "output collection %s" % self.name):
 235                         outputs = done.done(self, record, dirs["tmpdir"],
 236                                             dirs["outdir"], dirs["keep"])
 237             except WorkflowException as e:
 238                 logger.error("%s unable to collect output from %s:\n%s",
 239                              self.arvrunner.label(self), record["output"], e, exc_info=(e if self.arvrunner.debug else False))
 240                 processStatus = "permanentFail"
 241             except Exception as e:
 242                 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
 243                 processStatus = "permanentFail"
 244
 245             # Note: Currently, on error output_callback is expecting an empty dict,
 246             # anything else will fail.
 247             if not isinstance(outputs, dict):
 248                 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
 249                 outputs = {}
 250                 processStatus = "permanentFail"
 251         finally:
 252             self.output_callback(outputs, processStatus)
 253             if record["uuid"] in self.arvrunner.processes:
 254                 del self.arvrunner.processes[record["uuid"]]
 255
 256 class RunnerJob(Runner):
 257     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 258
 259     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 260         """Create an Arvados job specification for this workflow.
 261
 262         The returned dict can be used to create a job (i.e., passed as
 263         the +body+ argument to jobs().create()), or as a component in
 264         a pipeline template or pipeline instance.
 265         """
 266
 267         if self.tool.tool["id"].startswith("keep:"):
 268             self.job_order["cwl:tool"] = self.tool.tool["id"][5:]
 269         else:
 270             packed = packed_workflow(self.arvrunner, self.tool)
 271             wf_pdh = upload_workflow_collection(self.arvrunner, self.name, packed)
 272             self.job_order["cwl:tool"] = "%s/workflow.cwl#main" % wf_pdh
 273
 274         adjustDirObjs(self.job_order, trim_listing)
 275         adjustFileObjs(self.job_order, trim_anonymous_location)
 276         adjustDirObjs(self.job_order, trim_anonymous_location)
 277
 278         if self.output_name:
 279             self.job_order["arv:output_name"] = self.output_name
 280
 281         if self.output_tags:
 282             self.job_order["arv:output_tags"] = self.output_tags
 283
 284         self.job_order["arv:enable_reuse"] = self.enable_reuse
 285
 286         if self.on_error:
 287             self.job_order["arv:on_error"] = self.on_error
 288
 289         return {
 290             "script": "cwl-runner",
 291             "script_version": "master",
 292             "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9",
 293             "repository": "arvados",
 294             "script_parameters": self.job_order,
 295             "runtime_constraints": {
 296                 "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 297                 "min_ram_mb_per_node": self.submit_runner_ram
 298             }
 299         }
 300
 301     def run(self, *args, **kwargs):
 302         job_spec = self.arvados_job_spec(*args, **kwargs)
 303
 304         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 305
 306         job = self.arvrunner.api.jobs().create(
 307             body=job_spec,
 308             find_or_create=self.enable_reuse
 309         ).execute(num_retries=self.arvrunner.num_retries)
 310
 311         for k,v in job_spec["script_parameters"].items():
 312             if v is False or v is None or isinstance(v, dict):
 313                 job_spec["script_parameters"][k] = {"value": v}
 314
 315         del job_spec["owner_uuid"]
 316         job_spec["job"] = job
 317         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 318             body={
 319                 "owner_uuid": self.arvrunner.project_uuid,
 320                 "name": self.name,
 321                 "components": {"cwl-runner": job_spec },
 322                 "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries)
 323         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 324
 325         if kwargs.get("wait") is False:
 326             self.uuid = self.arvrunner.pipeline["uuid"]
 327             return
 328
 329         self.uuid = job["uuid"]
 330         self.arvrunner.processes[self.uuid] = self
 331
 332         if job["state"] in ("Complete", "Failed", "Cancelled"):
 333             self.done(job)
 334
 335
 336 class RunnerTemplate(object):
 337     """An Arvados pipeline template that invokes a CWL workflow."""
 338
 339     type_to_dataclass = {
 340         'boolean': 'boolean',
 341         'File': 'File',
 342         'Directory': 'Collection',
 343         'float': 'number',
 344         'int': 'number',
 345         'string': 'text',
 346     }
 347
 348     def __init__(self, runner, tool, job_order, enable_reuse, uuid,
 349                  submit_runner_ram=0, name=None):
 350         self.runner = runner
 351         self.tool = tool
 352         self.job = RunnerJob(
 353             runner=runner,
 354             tool=tool,
 355             job_order=job_order,
 356             enable_reuse=enable_reuse,
 357             output_name=None,
 358             output_tags=None,
 359             submit_runner_ram=submit_runner_ram,
 360             name=name)
 361         self.uuid = uuid
 362
 363     def pipeline_component_spec(self):
 364         """Return a component that Workbench and a-r-p-i will understand.
 365
 366         Specifically, translate CWL input specs to Arvados pipeline
 367         format, like {"dataclass":"File","value":"xyz"}.
 368         """
 369
 370         spec = self.job.arvados_job_spec()
 371
 372         # Most of the component spec is exactly the same as the job
 373         # spec (script, script_version, etc.).
 374         # spec['script_parameters'] isn't right, though. A component
 375         # spec's script_parameters hash is a translation of
 376         # self.tool.tool['inputs'] with defaults/overrides taken from
 377         # the job order. So we move the job parameters out of the way
 378         # and build a new spec['script_parameters'].
 379         job_params = spec['script_parameters']
 380         spec['script_parameters'] = {}
 381
 382         for param in self.tool.tool['inputs']:
 383             param = copy.deepcopy(param)
 384
 385             # Data type and "required" flag...
 386             types = param['type']
 387             if not isinstance(types, list):
 388                 types = [types]
 389             param['required'] = 'null' not in types
 390             non_null_types = [t for t in types if t != "null"]
 391             if len(non_null_types) == 1:
 392                 the_type = [c for c in non_null_types][0]
 393                 dataclass = None
 394                 if isinstance(the_type, basestring):
 395                     dataclass = self.type_to_dataclass.get(the_type)
 396                 if dataclass:
 397                     param['dataclass'] = dataclass
 398             # Note: If we didn't figure out a single appropriate
 399             # dataclass, we just left that attribute out.  We leave
 400             # the "type" attribute there in any case, which might help
 401             # downstream.
 402
 403             # Title and description...
 404             title = param.pop('label', '')
 405             descr = param.pop('doc', '').rstrip('\n')
 406             if title:
 407                 param['title'] = title
 408             if descr:
 409                 param['description'] = descr
 410
 411             # Fill in the value from the current job order, if any.
 412             param_id = shortname(param.pop('id'))
 413             value = job_params.get(param_id)
 414             if value is None:
 415                 pass
 416             elif not isinstance(value, dict):
 417                 param['value'] = value
 418             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 419                 param['value'] = value['location'][5:]
 420
 421             spec['script_parameters'][param_id] = param
 422         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 423         return spec
 424
 425     def save(self):
 426         body = {
 427             "components": {
 428                 self.job.name: self.pipeline_component_spec(),
 429             },
 430             "name": self.job.name,
 431         }
 432         if self.runner.project_uuid:
 433             body["owner_uuid"] = self.runner.project_uuid
 434         if self.uuid:
 435             self.runner.api.pipeline_templates().update(
 436                 uuid=self.uuid, body=body).execute(
 437                     num_retries=self.runner.num_retries)
 438             logger.info("Updated template %s", self.uuid)
 439         else:
 440             self.uuid = self.runner.api.pipeline_templates().create(
 441                 body=body, ensure_unique_name=True).execute(
 442                     num_retries=self.runner.num_retries)['uuid']
 443             logger.info("Created template %s", self.uuid)