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