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