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         enable_reuse = kwargs.get("enable_reuse", True)
 128         if enable_reuse:
 129             reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement")
 130             if reuse_req:
 131                 enable_reuse = reuse_req["enableReuse"]
 132
 133         try:
 134             with Perf(metrics, "create %s" % self.name):
 135                 response = self.arvrunner.api.jobs().create(
 136                     body={
 137                         "owner_uuid": self.arvrunner.project_uuid,
 138                         "script": "crunchrunner",
 139                         "repository": "arvados",
 140                         "script_version": "master",
 141                         "minimum_script_version": crunchrunner_git_commit,
 142                         "script_parameters": {"tasks": [script_parameters]},
 143                         "runtime_constraints": runtime_constraints
 144                     },
 145                     filters=filters,
 146                     find_or_create=enable_reuse
 147                 ).execute(num_retries=self.arvrunner.num_retries)
 148
 149             self.arvrunner.processes[response["uuid"]] = self
 150
 151             self.update_pipeline_component(response)
 152
 153             if response["state"] == "Complete":
 154                 logger.info("%s reused job %s", self.arvrunner.label(self), response["uuid"])
 155                 # Give read permission to the desired project on reused jobs
 156                 if response["owner_uuid"] != self.arvrunner.project_uuid:
 157                     try:
 158                         self.arvrunner.api.links().create(body={
 159                             'link_class': 'permission',
 160                             'name': 'can_read',
 161                             'tail_uuid': self.arvrunner.project_uuid,
 162                             'head_uuid': response["uuid"],
 163                             }).execute(num_retries=self.arvrunner.num_retries)
 164                     except ApiError as e:
 165                         # The user might not have "manage" access on the job: log
 166                         # a message and continue.
 167                         logger.info("Creating read permission on job %s: %s",
 168                                     response["uuid"],
 169                                     e)
 170
 171                 with Perf(metrics, "done %s" % self.name):
 172                     self.done(response)
 173             else:
 174                 logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 175         except Exception as e:
 176             logger.exception("%s error" % (self.arvrunner.label(self)))
 177             self.output_callback({}, "permanentFail")
 178
 179     def update_pipeline_component(self, record):
 180         if self.arvrunner.pipeline:
 181             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 182             with Perf(metrics, "update_pipeline_component %s" % self.name):
 183                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 184                                                                                  body={
 185                                                                                     "components": self.arvrunner.pipeline["components"]
 186                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 187         if self.arvrunner.uuid:
 188             try:
 189                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 190                 if job:
 191                     components = job["components"]
 192                     components[self.name] = record["uuid"]
 193                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 194                         body={
 195                             "components": components
 196                         }).execute(num_retries=self.arvrunner.num_retries)
 197             except Exception as e:
 198                 logger.info("Error adding to components: %s", e)
 199
 200     def done(self, record):
 201         try:
 202             self.update_pipeline_component(record)
 203         except:
 204             pass
 205
 206         try:
 207             if record["state"] == "Complete":
 208                 processStatus = "success"
 209             else:
 210                 processStatus = "permanentFail"
 211
 212             outputs = {}
 213             try:
 214                 if record["output"]:
 215                     with Perf(metrics, "inspect log %s" % self.name):
 216                         logc = arvados.collection.CollectionReader(record["log"],
 217                                                                    api_client=self.arvrunner.api,
 218                                                                    keep_client=self.arvrunner.keep_client,
 219                                                                    num_retries=self.arvrunner.num_retries)
 220                         log = logc.open(logc.keys()[0])
 221                         dirs = {}
 222                         tmpdir = None
 223                         outdir = None
 224                         keepdir = None
 225                         for l in log:
 226                             # Determine the tmpdir, outdir and keepdir paths from
 227                             # the job run.  Unfortunately, we can't take the first
 228                             # values we find (which are expected to be near the
 229                             # top) and stop scanning because if the node fails and
 230                             # the job restarts on a different node these values
 231                             # will different runs, and we need to know about the
 232                             # final run that actually produced output.
 233                             g = crunchrunner_re.match(l)
 234                             if g:
 235                                 dirs[g.group(1)] = g.group(2)
 236
 237                     if processStatus == "permanentFail":
 238                         done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 239
 240                     with Perf(metrics, "output collection %s" % self.name):
 241                         outputs = done.done(self, record, dirs["tmpdir"],
 242                                             dirs["outdir"], dirs["keep"])
 243             except WorkflowException as e:
 244                 logger.error("%s unable to collect output from %s:\n%s",
 245                              self.arvrunner.label(self), record["output"], e, exc_info=(e if self.arvrunner.debug else False))
 246                 processStatus = "permanentFail"
 247             except Exception as e:
 248                 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
 249                 processStatus = "permanentFail"
 250
 251             # Note: Currently, on error output_callback is expecting an empty dict,
 252             # anything else will fail.
 253             if not isinstance(outputs, dict):
 254                 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
 255                 outputs = {}
 256                 processStatus = "permanentFail"
 257         finally:
 258             self.output_callback(outputs, processStatus)
 259             if record["uuid"] in self.arvrunner.processes:
 260                 del self.arvrunner.processes[record["uuid"]]
 261
 262 class RunnerJob(Runner):
 263     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 264
 265     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 266         """Create an Arvados job specification for this workflow.
 267
 268         The returned dict can be used to create a job (i.e., passed as
 269         the +body+ argument to jobs().create()), or as a component in
 270         a pipeline template or pipeline instance.
 271         """
 272
 273         if self.tool.tool["id"].startswith("keep:"):
 274             self.job_order["cwl:tool"] = self.tool.tool["id"][5:]
 275         else:
 276             packed = packed_workflow(self.arvrunner, self.tool)
 277             wf_pdh = upload_workflow_collection(self.arvrunner, self.name, packed)
 278             self.job_order["cwl:tool"] = "%s/workflow.cwl#main" % wf_pdh
 279
 280         adjustDirObjs(self.job_order, trim_listing)
 281         adjustFileObjs(self.job_order, trim_anonymous_location)
 282         adjustDirObjs(self.job_order, trim_anonymous_location)
 283
 284         if self.output_name:
 285             self.job_order["arv:output_name"] = self.output_name
 286
 287         if self.output_tags:
 288             self.job_order["arv:output_tags"] = self.output_tags
 289
 290         self.job_order["arv:enable_reuse"] = self.enable_reuse
 291
 292         if self.on_error:
 293             self.job_order["arv:on_error"] = self.on_error
 294
 295         return {
 296             "script": "cwl-runner",
 297             "script_version": "master",
 298             "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9",
 299             "repository": "arvados",
 300             "script_parameters": self.job_order,
 301             "runtime_constraints": {
 302                 "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 303                 "min_ram_mb_per_node": self.submit_runner_ram
 304             }
 305         }
 306
 307     def run(self, *args, **kwargs):
 308         job_spec = self.arvados_job_spec(*args, **kwargs)
 309
 310         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 311
 312         job = self.arvrunner.api.jobs().create(
 313             body=job_spec,
 314             find_or_create=self.enable_reuse
 315         ).execute(num_retries=self.arvrunner.num_retries)
 316
 317         for k,v in job_spec["script_parameters"].items():
 318             if v is False or v is None or isinstance(v, dict):
 319                 job_spec["script_parameters"][k] = {"value": v}
 320
 321         del job_spec["owner_uuid"]
 322         job_spec["job"] = job
 323         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 324             body={
 325                 "owner_uuid": self.arvrunner.project_uuid,
 326                 "name": self.name,
 327                 "components": {"cwl-runner": job_spec },
 328                 "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries)
 329         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 330
 331         if kwargs.get("wait") is False:
 332             self.uuid = self.arvrunner.pipeline["uuid"]
 333             return
 334
 335         self.uuid = job["uuid"]
 336         self.arvrunner.processes[self.uuid] = self
 337
 338         if job["state"] in ("Complete", "Failed", "Cancelled"):
 339             self.done(job)
 340
 341
 342 class RunnerTemplate(object):
 343     """An Arvados pipeline template that invokes a CWL workflow."""
 344
 345     type_to_dataclass = {
 346         'boolean': 'boolean',
 347         'File': 'File',
 348         'Directory': 'Collection',
 349         'float': 'number',
 350         'int': 'number',
 351         'string': 'text',
 352     }
 353
 354     def __init__(self, runner, tool, job_order, enable_reuse, uuid,
 355                  submit_runner_ram=0, name=None):
 356         self.runner = runner
 357         self.tool = tool
 358         self.job = RunnerJob(
 359             runner=runner,
 360             tool=tool,
 361             job_order=job_order,
 362             enable_reuse=enable_reuse,
 363             output_name=None,
 364             output_tags=None,
 365             submit_runner_ram=submit_runner_ram,
 366             name=name)
 367         self.uuid = uuid
 368
 369     def pipeline_component_spec(self):
 370         """Return a component that Workbench and a-r-p-i will understand.
 371
 372         Specifically, translate CWL input specs to Arvados pipeline
 373         format, like {"dataclass":"File","value":"xyz"}.
 374         """
 375
 376         spec = self.job.arvados_job_spec()
 377
 378         # Most of the component spec is exactly the same as the job
 379         # spec (script, script_version, etc.).
 380         # spec['script_parameters'] isn't right, though. A component
 381         # spec's script_parameters hash is a translation of
 382         # self.tool.tool['inputs'] with defaults/overrides taken from
 383         # the job order. So we move the job parameters out of the way
 384         # and build a new spec['script_parameters'].
 385         job_params = spec['script_parameters']
 386         spec['script_parameters'] = {}
 387
 388         for param in self.tool.tool['inputs']:
 389             param = copy.deepcopy(param)
 390
 391             # Data type and "required" flag...
 392             types = param['type']
 393             if not isinstance(types, list):
 394                 types = [types]
 395             param['required'] = 'null' not in types
 396             non_null_types = [t for t in types if t != "null"]
 397             if len(non_null_types) == 1:
 398                 the_type = [c for c in non_null_types][0]
 399                 dataclass = None
 400                 if isinstance(the_type, basestring):
 401                     dataclass = self.type_to_dataclass.get(the_type)
 402                 if dataclass:
 403                     param['dataclass'] = dataclass
 404             # Note: If we didn't figure out a single appropriate
 405             # dataclass, we just left that attribute out.  We leave
 406             # the "type" attribute there in any case, which might help
 407             # downstream.
 408
 409             # Title and description...
 410             title = param.pop('label', '')
 411             descr = param.pop('doc', '').rstrip('\n')
 412             if title:
 413                 param['title'] = title
 414             if descr:
 415                 param['description'] = descr
 416
 417             # Fill in the value from the current job order, if any.
 418             param_id = shortname(param.pop('id'))
 419             value = job_params.get(param_id)
 420             if value is None:
 421                 pass
 422             elif not isinstance(value, dict):
 423                 param['value'] = value
 424             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 425                 param['value'] = value['location'][5:]
 426
 427             spec['script_parameters'][param_id] = param
 428         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 429         return spec
 430
 431     def save(self):
 432         body = {
 433             "components": {
 434                 self.job.name: self.pipeline_component_spec(),
 435             },
 436             "name": self.job.name,
 437         }
 438         if self.runner.project_uuid:
 439             body["owner_uuid"] = self.runner.project_uuid
 440         if self.uuid:
 441             self.runner.api.pipeline_templates().update(
 442                 uuid=self.uuid, body=body).execute(
 443                     num_retries=self.runner.num_retries)
 444             logger.info("Updated template %s", self.uuid)
 445         else:
 446             self.uuid = self.runner.api.pipeline_templates().create(
 447                 body=body, ensure_unique_name=True).execute(
 448                     num_retries=self.runner.num_retries)['uuid']
 449             logger.info("Created template %s", self.uuid)