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