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 class ArvadosJob(object):
  28     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  29
  30     def __init__(self, runner):
  31         self.arvrunner = runner
  32         self.running = False
  33         self.uuid = None
  34
  35     def run(self, dry_run=False, pull_image=True, **kwargs):
  36         script_parameters = {
  37             "command": self.command_line
  38         }
  39         runtime_constraints = {}
  40
  41         with Perf(metrics, "generatefiles %s" % self.name):
  42             if self.generatefiles["listing"]:
  43                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  44                                                     keep_client=self.arvrunner.keep_client,
  45                                                     num_retries=self.arvrunner.num_retries)
  46                 script_parameters["task.vwd"] = {}
  47                 generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
  48                                                           separateDirs=False)
  49
  50                 with Perf(metrics, "createfiles %s" % self.name):
  51                     for f, p in generatemapper.items():
  52                         if p.type == "CreateFile":
  53                             with vwd.open(p.target, "w") as n:
  54                                 n.write(p.resolved.encode("utf-8"))
  55
  56                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
  57                     vwd.save_new()
  58
  59                 for f, p in generatemapper.items():
  60                     if p.type == "File":
  61                         script_parameters["task.vwd"][p.target] = p.resolved
  62                     if p.type == "CreateFile":
  63                         script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  64
  65         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  66         if self.environment:
  67             script_parameters["task.env"].update(self.environment)
  68
  69         if self.stdin:
  70             script_parameters["task.stdin"] = self.stdin
  71
  72         if self.stdout:
  73             script_parameters["task.stdout"] = self.stdout
  74
  75         if self.stderr:
  76             script_parameters["task.stderr"] = self.stderr
  77
  78         if self.successCodes:
  79             script_parameters["task.successCodes"] = self.successCodes
  80         if self.temporaryFailCodes:
  81             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  82         if self.permanentFailCodes:
  83             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  84
  85         with Perf(metrics, "arv_docker_get_image %s" % self.name):
  86             (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  87             if docker_req and kwargs.get("use_container") is not False:
  88                 if docker_req.get("dockerOutputDirectory"):
  89                     raise UnsupportedRequirement("Option 'dockerOutputDirectory' of DockerRequirement not supported.")
  90                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  91             else:
  92                 runtime_constraints["docker_image"] = arvados_jobs_image(self.arvrunner)
  93
  94         resources = self.builder.resources
  95         if resources is not None:
  96             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  97             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  98             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  99
 100         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 101         if runtime_req:
 102             if "keep_cache" in runtime_req:
 103                 runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
 104             if "outputDirType" in runtime_req:
 105                 if runtime_req["outputDirType"] == "local_output_dir":
 106                     script_parameters["task.keepTmpOutput"] = False
 107                 elif runtime_req["outputDirType"] == "keep_output_dir":
 108                     script_parameters["task.keepTmpOutput"] = True
 109
 110         filters = [["repository", "=", "arvados"],
 111                    ["script", "=", "crunchrunner"],
 112                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
 113         if not self.arvrunner.ignore_docker_for_reuse:
 114             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 115
 116         try:
 117             with Perf(metrics, "create %s" % self.name):
 118                 response = self.arvrunner.api.jobs().create(
 119                     body={
 120                         "owner_uuid": self.arvrunner.project_uuid,
 121                         "script": "crunchrunner",
 122                         "repository": "arvados",
 123                         "script_version": "master",
 124                         "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 125                         "script_parameters": {"tasks": [script_parameters]},
 126                         "runtime_constraints": runtime_constraints
 127                     },
 128                     filters=filters,
 129                     find_or_create=kwargs.get("enable_reuse", True)
 130                 ).execute(num_retries=self.arvrunner.num_retries)
 131
 132             self.arvrunner.processes[response["uuid"]] = self
 133
 134             self.update_pipeline_component(response)
 135
 136             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 137
 138             if response["state"] in ("Complete", "Failed", "Cancelled"):
 139                 with Perf(metrics, "done %s" % self.name):
 140                     self.done(response)
 141         except Exception as e:
 142             logger.exception("Job %s error" % (self.name))
 143             self.output_callback({}, "permanentFail")
 144
 145     def update_pipeline_component(self, record):
 146         if self.arvrunner.pipeline:
 147             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 148             with Perf(metrics, "update_pipeline_component %s" % self.name):
 149                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 150                                                                                  body={
 151                                                                                     "components": self.arvrunner.pipeline["components"]
 152                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 153         if self.arvrunner.uuid:
 154             try:
 155                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 156                 if job:
 157                     components = job["components"]
 158                     components[self.name] = record["uuid"]
 159                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 160                         body={
 161                             "components": components
 162                         }).execute(num_retries=self.arvrunner.num_retries)
 163             except Exception as e:
 164                 logger.info("Error adding to components: %s", e)
 165
 166     def done(self, record):
 167         try:
 168             self.update_pipeline_component(record)
 169         except:
 170             pass
 171
 172         try:
 173             if record["state"] == "Complete":
 174                 processStatus = "success"
 175             else:
 176                 processStatus = "permanentFail"
 177
 178             outputs = {}
 179             try:
 180                 if record["output"]:
 181                     with Perf(metrics, "inspect log %s" % self.name):
 182                         logc = arvados.collection.CollectionReader(record["log"],
 183                                                                    api_client=self.arvrunner.api,
 184                                                                    keep_client=self.arvrunner.keep_client,
 185                                                                    num_retries=self.arvrunner.num_retries)
 186                         log = logc.open(logc.keys()[0])
 187                         dirs = {}
 188                         tmpdir = None
 189                         outdir = None
 190                         keepdir = None
 191                         for l in log:
 192                             # Determine the tmpdir, outdir and keepdir paths from
 193                             # the job run.  Unfortunately, we can't take the first
 194                             # values we find (which are expected to be near the
 195                             # top) and stop scanning because if the node fails and
 196                             # the job restarts on a different node these values
 197                             # will different runs, and we need to know about the
 198                             # final run that actually produced output.
 199                             g = crunchrunner_re.match(l)
 200                             if g:
 201                                 dirs[g.group(1)] = g.group(2)
 202
 203                     with Perf(metrics, "output collection %s" % self.name):
 204                         outputs = done.done(self, record, dirs["tmpdir"],
 205                                             dirs["outdir"], dirs["keep"])
 206             except WorkflowException as e:
 207                 logger.error("Error while collecting output for job %s:\n%s", self.name, e, exc_info=(e if self.arvrunner.debug else False))
 208                 processStatus = "permanentFail"
 209             except Exception as e:
 210                 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
 211                 processStatus = "permanentFail"
 212
 213             # Note: Currently, on error output_callback is expecting an empty dict,
 214             # anything else will fail.
 215             if not isinstance(outputs, dict):
 216                 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
 217                 outputs = {}
 218                 processStatus = "permanentFail"
 219
 220             self.output_callback(outputs, processStatus)
 221         finally:
 222             del self.arvrunner.processes[record["uuid"]]
 223
 224
 225 class RunnerJob(Runner):
 226     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 227
 228     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 229         """Create an Arvados job specification for this workflow.
 230
 231         The returned dict can be used to create a job (i.e., passed as
 232         the +body+ argument to jobs().create()), or as a component in
 233         a pipeline template or pipeline instance.
 234         """
 235
 236         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 237
 238         # Need to filter this out, gets added by cwltool when providing
 239         # parameters on the command line, and arv-run-pipeline-instance doesn't
 240         # like it.
 241         if "job_order" in self.job_order:
 242             del self.job_order["job_order"]
 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         return {
 255             "script": "cwl-runner",
 256             "script_version": __version__,
 257             "repository": "arvados",
 258             "script_parameters": self.job_order,
 259             "runtime_constraints": {
 260                 "docker_image": arvados_jobs_image(self.arvrunner),
 261                 "min_ram_mb_per_node": self.submit_runner_ram
 262             }
 263         }
 264
 265     def run(self, *args, **kwargs):
 266         job_spec = self.arvados_job_spec(*args, **kwargs)
 267
 268         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 269
 270         job = self.arvrunner.api.jobs().create(
 271             body=job_spec,
 272             find_or_create=self.enable_reuse
 273         ).execute(num_retries=self.arvrunner.num_retries)
 274
 275         for k,v in job_spec["script_parameters"].items():
 276             if v is False or v is None or isinstance(v, dict):
 277                 job_spec["script_parameters"][k] = {"value": v}
 278
 279         del job_spec["owner_uuid"]
 280         job_spec["job"] = job
 281         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 282             body={
 283                 "owner_uuid": self.arvrunner.project_uuid,
 284                 "name": shortname(self.tool.tool["id"]),
 285                 "components": {"cwl-runner": job_spec },
 286                 "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries)
 287         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 288
 289         if kwargs.get("wait") is False:
 290             self.uuid = self.arvrunner.pipeline["uuid"]
 291             return
 292
 293         self.uuid = job["uuid"]
 294         self.arvrunner.processes[self.uuid] = self
 295
 296         if job["state"] in ("Complete", "Failed", "Cancelled"):
 297             self.done(job)
 298
 299
 300 class RunnerTemplate(object):
 301     """An Arvados pipeline template that invokes a CWL workflow."""
 302
 303     type_to_dataclass = {
 304         'boolean': 'boolean',
 305         'File': 'File',
 306         'Directory': 'Collection',
 307         'float': 'number',
 308         'int': 'number',
 309         'string': 'text',
 310     }
 311
 312     def __init__(self, runner, tool, job_order, enable_reuse, uuid, submit_runner_ram=0):
 313         self.runner = runner
 314         self.tool = tool
 315         self.job = RunnerJob(
 316             runner=runner,
 317             tool=tool,
 318             job_order=job_order,
 319             enable_reuse=enable_reuse,
 320             output_name=None,
 321             output_tags=None,
 322             submit_runner_ram=submit_runner_ram)
 323         self.uuid = uuid
 324
 325     def pipeline_component_spec(self):
 326         """Return a component that Workbench and a-r-p-i will understand.
 327
 328         Specifically, translate CWL input specs to Arvados pipeline
 329         format, like {"dataclass":"File","value":"xyz"}.
 330         """
 331
 332         spec = self.job.arvados_job_spec()
 333
 334         # Most of the component spec is exactly the same as the job
 335         # spec (script, script_version, etc.).
 336         # spec['script_parameters'] isn't right, though. A component
 337         # spec's script_parameters hash is a translation of
 338         # self.tool.tool['inputs'] with defaults/overrides taken from
 339         # the job order. So we move the job parameters out of the way
 340         # and build a new spec['script_parameters'].
 341         job_params = spec['script_parameters']
 342         spec['script_parameters'] = {}
 343
 344         for param in self.tool.tool['inputs']:
 345             param = copy.deepcopy(param)
 346
 347             # Data type and "required" flag...
 348             types = param['type']
 349             if not isinstance(types, list):
 350                 types = [types]
 351             param['required'] = 'null' not in types
 352             non_null_types = set(types) - set(['null'])
 353             if len(non_null_types) == 1:
 354                 the_type = [c for c in non_null_types][0]
 355                 dataclass = self.type_to_dataclass.get(the_type)
 356                 if dataclass:
 357                     param['dataclass'] = dataclass
 358             # Note: If we didn't figure out a single appropriate
 359             # dataclass, we just left that attribute out.  We leave
 360             # the "type" attribute there in any case, which might help
 361             # downstream.
 362
 363             # Title and description...
 364             title = param.pop('label', '')
 365             descr = param.pop('doc', '').rstrip('\n')
 366             if title:
 367                 param['title'] = title
 368             if descr:
 369                 param['description'] = descr
 370
 371             # Fill in the value from the current job order, if any.
 372             param_id = shortname(param.pop('id'))
 373             value = job_params.get(param_id)
 374             if value is None:
 375                 pass
 376             elif not isinstance(value, dict):
 377                 param['value'] = value
 378             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 379                 param['value'] = value['location'][5:]
 380
 381             spec['script_parameters'][param_id] = param
 382         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 383         return spec
 384
 385     def save(self):
 386         body = {
 387             "components": {
 388                 self.job.name: self.pipeline_component_spec(),
 389             },
 390             "name": self.job.name,
 391         }
 392         if self.runner.project_uuid:
 393             body["owner_uuid"] = self.runner.project_uuid
 394         if self.uuid:
 395             self.runner.api.pipeline_templates().update(
 396                 uuid=self.uuid, body=body).execute(
 397                     num_retries=self.runner.num_retries)
 398             logger.info("Updated template %s", self.uuid)
 399         else:
 400             self.uuid = self.runner.api.pipeline_templates().create(
 401                 body=body, ensure_unique_name=True).execute(
 402                     num_retries=self.runner.num_retries)['uuid']
 403             logger.info("Created template %s", self.uuid)