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