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                 with Perf(metrics, "done %s" % self.name):
 149                     self.done(response)
 150             else:
 151                 logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 152         except Exception as e:
 153             logger.exception("%s error" % (self.arvrunner.label(self)))
 154             self.output_callback({}, "permanentFail")
 155
 156     def update_pipeline_component(self, record):
 157         if self.arvrunner.pipeline:
 158             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 159             with Perf(metrics, "update_pipeline_component %s" % self.name):
 160                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 161                                                                                  body={
 162                                                                                     "components": self.arvrunner.pipeline["components"]
 163                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 164         if self.arvrunner.uuid:
 165             try:
 166                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 167                 if job:
 168                     components = job["components"]
 169                     components[self.name] = record["uuid"]
 170                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 171                         body={
 172                             "components": components
 173                         }).execute(num_retries=self.arvrunner.num_retries)
 174             except Exception as e:
 175                 logger.info("Error adding to components: %s", e)
 176
 177     def done(self, record):
 178         try:
 179             self.update_pipeline_component(record)
 180         except:
 181             pass
 182
 183         try:
 184             if record["state"] == "Complete":
 185                 processStatus = "success"
 186             else:
 187                 processStatus = "permanentFail"
 188
 189             outputs = {}
 190             try:
 191                 if record["output"]:
 192                     with Perf(metrics, "inspect log %s" % self.name):
 193                         logc = arvados.collection.CollectionReader(record["log"],
 194                                                                    api_client=self.arvrunner.api,
 195                                                                    keep_client=self.arvrunner.keep_client,
 196                                                                    num_retries=self.arvrunner.num_retries)
 197                         log = logc.open(logc.keys()[0])
 198                         dirs = {}
 199                         tmpdir = None
 200                         outdir = None
 201                         keepdir = None
 202                         for l in log:
 203                             # Determine the tmpdir, outdir and keepdir paths from
 204                             # the job run.  Unfortunately, we can't take the first
 205                             # values we find (which are expected to be near the
 206                             # top) and stop scanning because if the node fails and
 207                             # the job restarts on a different node these values
 208                             # will different runs, and we need to know about the
 209                             # final run that actually produced output.
 210                             g = crunchrunner_re.match(l)
 211                             if g:
 212                                 dirs[g.group(1)] = g.group(2)
 213
 214                     if processStatus == "permanentFail":
 215                         done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 216
 217                     with Perf(metrics, "output collection %s" % self.name):
 218                         outputs = done.done(self, record, dirs["tmpdir"],
 219                                             dirs["outdir"], dirs["keep"])
 220             except WorkflowException as e:
 221                 logger.error("%s unable to collect output from %s:\n%s",
 222                              self.arvrunner.label(self), record["output"], e, exc_info=(e if self.arvrunner.debug else False))
 223                 processStatus = "permanentFail"
 224             except Exception as e:
 225                 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
 226                 processStatus = "permanentFail"
 227
 228             # Note: Currently, on error output_callback is expecting an empty dict,
 229             # anything else will fail.
 230             if not isinstance(outputs, dict):
 231                 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
 232                 outputs = {}
 233                 processStatus = "permanentFail"
 234         finally:
 235             self.output_callback(outputs, processStatus)
 236             if record["uuid"] in self.arvrunner.processes:
 237                 del self.arvrunner.processes[record["uuid"]]
 238
 239 class RunnerJob(Runner):
 240     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 241
 242     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 243         """Create an Arvados job specification for this workflow.
 244
 245         The returned dict can be used to create a job (i.e., passed as
 246         the +body+ argument to jobs().create()), or as a component in
 247         a pipeline template or pipeline instance.
 248         """
 249
 250         if self.tool.tool["id"].startswith("keep:"):
 251             self.job_order["cwl:tool"] = self.tool.tool["id"][5:]
 252         else:
 253             packed = packed_workflow(self.arvrunner, self.tool)
 254             wf_pdh = upload_workflow_collection(self.arvrunner, self.name, packed)
 255             self.job_order["cwl:tool"] = "%s/workflow.cwl#main" % wf_pdh
 256
 257         adjustDirObjs(self.job_order, trim_listing)
 258         adjustFileObjs(self.job_order, trim_anonymous_location)
 259         adjustDirObjs(self.job_order, trim_anonymous_location)
 260
 261         if self.output_name:
 262             self.job_order["arv:output_name"] = self.output_name
 263
 264         if self.output_tags:
 265             self.job_order["arv:output_tags"] = self.output_tags
 266
 267         self.job_order["arv:enable_reuse"] = self.enable_reuse
 268
 269         if self.on_error:
 270             self.job_order["arv:on_error"] = self.on_error
 271
 272         return {
 273             "script": "cwl-runner",
 274             "script_version": "master",
 275             "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9",
 276             "repository": "arvados",
 277             "script_parameters": self.job_order,
 278             "runtime_constraints": {
 279                 "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 280                 "min_ram_mb_per_node": self.submit_runner_ram
 281             }
 282         }
 283
 284     def run(self, *args, **kwargs):
 285         job_spec = self.arvados_job_spec(*args, **kwargs)
 286
 287         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 288
 289         job = self.arvrunner.api.jobs().create(
 290             body=job_spec,
 291             find_or_create=self.enable_reuse
 292         ).execute(num_retries=self.arvrunner.num_retries)
 293
 294         if self.enable_reuse:
 295             # When reusing jobs, copy its output/log collection to the desired project
 296             reused_collections = [('Output', job.get('output', None)),
 297                                   ('Log', job.get('log', None))]
 298             for col_type, pdh in [(n, p) for n, p in reused_collections if p]:
 299                 c = arvados.collection.Collection(pdh,
 300                                                   api_client=self.arvrunner.api,
 301                                                   keep_client=self.arvrunner.keep_client,
 302                                                   num_retries=self.arvrunner.num_retries)
 303                 c.save_new(name="{} of {}".format(col_type, self.name),
 304                            owner_uuid=self.arvrunner.project_uuid,
 305                            ensure_unique_name=True,
 306                            num_retries=self.arvrunner.num_retries)
 307                 logger.info("Copied reused job's %s to collection %s",
 308                             col_type.lower(),
 309                             c.manifest_locator())
 310             # Give read permission to the desired project on reused jobs
 311             for job_name, job_uuid in job.get('components', {}).items():
 312                 self.arvrunner.api.links().create(body={
 313                     'link_class': 'can_read',
 314                     'tail_uuid': self.arvrunner.project_uuid,
 315                     'head_uuid': job_uuid,
 316                     }).execute(num_retries=self.arvrunner.num_retries)
 317
 318         for k,v in job_spec["script_parameters"].items():
 319             if v is False or v is None or isinstance(v, dict):
 320                 job_spec["script_parameters"][k] = {"value": v}
 321
 322         del job_spec["owner_uuid"]
 323         job_spec["job"] = job
 324         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 325             body={
 326                 "owner_uuid": self.arvrunner.project_uuid,
 327                 "name": self.name,
 328                 "components": {"cwl-runner": job_spec },
 329                 "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries)
 330         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 331
 332         if kwargs.get("wait") is False:
 333             self.uuid = self.arvrunner.pipeline["uuid"]
 334             return
 335
 336         self.uuid = job["uuid"]
 337         self.arvrunner.processes[self.uuid] = self
 338
 339         if job["state"] in ("Complete", "Failed", "Cancelled"):
 340             self.done(job)
 341
 342
 343 class RunnerTemplate(object):
 344     """An Arvados pipeline template that invokes a CWL workflow."""
 345
 346     type_to_dataclass = {
 347         'boolean': 'boolean',
 348         'File': 'File',
 349         'Directory': 'Collection',
 350         'float': 'number',
 351         'int': 'number',
 352         'string': 'text',
 353     }
 354
 355     def __init__(self, runner, tool, job_order, enable_reuse, uuid,
 356                  submit_runner_ram=0, name=None):
 357         self.runner = runner
 358         self.tool = tool
 359         self.job = RunnerJob(
 360             runner=runner,
 361             tool=tool,
 362             job_order=job_order,
 363             enable_reuse=enable_reuse,
 364             output_name=None,
 365             output_tags=None,
 366             submit_runner_ram=submit_runner_ram,
 367             name=name)
 368         self.uuid = uuid
 369
 370     def pipeline_component_spec(self):
 371         """Return a component that Workbench and a-r-p-i will understand.
 372
 373         Specifically, translate CWL input specs to Arvados pipeline
 374         format, like {"dataclass":"File","value":"xyz"}.
 375         """
 376
 377         spec = self.job.arvados_job_spec()
 378
 379         # Most of the component spec is exactly the same as the job
 380         # spec (script, script_version, etc.).
 381         # spec['script_parameters'] isn't right, though. A component
 382         # spec's script_parameters hash is a translation of
 383         # self.tool.tool['inputs'] with defaults/overrides taken from
 384         # the job order. So we move the job parameters out of the way
 385         # and build a new spec['script_parameters'].
 386         job_params = spec['script_parameters']
 387         spec['script_parameters'] = {}
 388
 389         for param in self.tool.tool['inputs']:
 390             param = copy.deepcopy(param)
 391
 392             # Data type and "required" flag...
 393             types = param['type']
 394             if not isinstance(types, list):
 395                 types = [types]
 396             param['required'] = 'null' not in types
 397             non_null_types = [t for t in types if t != "null"]
 398             if len(non_null_types) == 1:
 399                 the_type = [c for c in non_null_types][0]
 400                 dataclass = None
 401                 if isinstance(the_type, basestring):
 402                     dataclass = self.type_to_dataclass.get(the_type)
 403                 if dataclass:
 404                     param['dataclass'] = dataclass
 405             # Note: If we didn't figure out a single appropriate
 406             # dataclass, we just left that attribute out.  We leave
 407             # the "type" attribute there in any case, which might help
 408             # downstream.
 409
 410             # Title and description...
 411             title = param.pop('label', '')
 412             descr = param.pop('doc', '').rstrip('\n')
 413             if title:
 414                 param['title'] = title
 415             if descr:
 416                 param['description'] = descr
 417
 418             # Fill in the value from the current job order, if any.
 419             param_id = shortname(param.pop('id'))
 420             value = job_params.get(param_id)
 421             if value is None:
 422                 pass
 423             elif not isinstance(value, dict):
 424                 param['value'] = value
 425             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 426                 param['value'] = value['location'][5:]
 427
 428             spec['script_parameters'][param_id] = param
 429         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 430         return spec
 431
 432     def save(self):
 433         body = {
 434             "components": {
 435                 self.job.name: self.pipeline_component_spec(),
 436             },
 437             "name": self.job.name,
 438         }
 439         if self.runner.project_uuid:
 440             body["owner_uuid"] = self.runner.project_uuid
 441         if self.uuid:
 442             self.runner.api.pipeline_templates().update(
 443                 uuid=self.uuid, body=body).execute(
 444                     num_retries=self.runner.num_retries)
 445             logger.info("Updated template %s", self.uuid)
 446         else:
 447             self.uuid = self.runner.api.pipeline_templates().create(
 448                 body=body, ensure_unique_name=True).execute(
 449                     num_retries=self.runner.num_retries)['uuid']
 450             logger.info("Created template %s", self.uuid)