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 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
  26 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
  27 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
  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                 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.error("Got error %s" % str(e))
 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                         tmpdir = None
 188                         outdir = None
 189                         keepdir = None
 190                         for l in log:
 191                             # Determine the tmpdir, outdir and keepdir paths from
 192                             # the job run.  Unfortunately, we can't take the first
 193                             # values we find (which are expected to be near the
 194                             # top) and stop scanning because if the node fails and
 195                             # the job restarts on a different node these values
 196                             # will different runs, and we need to know about the
 197                             # final run that actually produced output.
 198
 199                             g = tmpdirre.match(l)
 200                             if g:
 201                                 tmpdir = g.group(1)
 202                             g = outdirre.match(l)
 203                             if g:
 204                                 outdir = g.group(1)
 205                             g = keepre.match(l)
 206                             if g:
 207                                 keepdir = g.group(1)
 208
 209                     with Perf(metrics, "output collection %s" % self.name):
 210                         outputs = done.done(self, record, tmpdir, outdir, keepdir)
 211             except WorkflowException as e:
 212                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 213                 processStatus = "permanentFail"
 214                 outputs = None
 215             except Exception as e:
 216                 logger.exception("Got unknown exception while collecting job outputs:")
 217                 processStatus = "permanentFail"
 218                 outputs = None
 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         if self.output_name:
 246             self.job_order["arv:output_name"] = self.output_name
 247         return {
 248             "script": "cwl-runner",
 249             "script_version": __version__,
 250             "repository": "arvados",
 251             "script_parameters": self.job_order,
 252             "runtime_constraints": {
 253                 "docker_image": arvados_jobs_image(self.arvrunner)
 254             }
 255         }
 256
 257     def run(self, *args, **kwargs):
 258         job_spec = self.arvados_job_spec(*args, **kwargs)
 259
 260         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 261
 262         job = self.arvrunner.api.jobs().create(
 263             body=job_spec,
 264             find_or_create=self.enable_reuse
 265         ).execute(num_retries=self.arvrunner.num_retries)
 266
 267         for k,v in job_spec["script_parameters"].items():
 268             if v is False or v is None or isinstance(v, dict):
 269                 job_spec["script_parameters"][k] = {"value": v}
 270
 271         del job_spec["owner_uuid"]
 272         job_spec["job"] = job
 273         self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 274             body={
 275                 "owner_uuid": self.arvrunner.project_uuid,
 276                 "name": shortname(self.tool.tool["id"]),
 277                 "components": {"cwl-runner": job_spec },
 278                 "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries)
 279         logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
 280
 281         if kwargs.get("wait") is False:
 282             self.uuid = self.arvrunner.pipeline["uuid"]
 283             return
 284
 285         self.uuid = job["uuid"]
 286         self.arvrunner.processes[self.uuid] = self
 287
 288         if job["state"] in ("Complete", "Failed", "Cancelled"):
 289             self.done(job)
 290
 291
 292 class RunnerTemplate(object):
 293     """An Arvados pipeline template that invokes a CWL workflow."""
 294
 295     type_to_dataclass = {
 296         'boolean': 'boolean',
 297         'File': 'File',
 298         'Directory': 'Collection',
 299         'float': 'number',
 300         'int': 'number',
 301         'string': 'text',
 302     }
 303
 304     def __init__(self, runner, tool, job_order, enable_reuse):
 305         self.runner = runner
 306         self.tool = tool
 307         self.job = RunnerJob(
 308             runner=runner,
 309             tool=tool,
 310             job_order=job_order,
 311             enable_reuse=enable_reuse,
 312             output_name=None)
 313
 314     def pipeline_component_spec(self):
 315         """Return a component that Workbench and a-r-p-i will understand.
 316
 317         Specifically, translate CWL input specs to Arvados pipeline
 318         format, like {"dataclass":"File","value":"xyz"}.
 319         """
 320         spec = self.job.arvados_job_spec()
 321
 322         # Most of the component spec is exactly the same as the job
 323         # spec (script, script_version, etc.).
 324         # spec['script_parameters'] isn't right, though. A component
 325         # spec's script_parameters hash is a translation of
 326         # self.tool.tool['inputs'] with defaults/overrides taken from
 327         # the job order. So we move the job parameters out of the way
 328         # and build a new spec['script_parameters'].
 329         job_params = spec['script_parameters']
 330         spec['script_parameters'] = {}
 331
 332         for param in self.tool.tool['inputs']:
 333             param = copy.deepcopy(param)
 334
 335             # Data type and "required" flag...
 336             types = param['type']
 337             if not isinstance(types, list):
 338                 types = [types]
 339             param['required'] = 'null' not in types
 340             non_null_types = set(types) - set(['null'])
 341             if len(non_null_types) == 1:
 342                 the_type = [c for c in non_null_types][0]
 343                 dataclass = self.type_to_dataclass.get(the_type)
 344                 if dataclass:
 345                     param['dataclass'] = dataclass
 346             # Note: If we didn't figure out a single appropriate
 347             # dataclass, we just left that attribute out.  We leave
 348             # the "type" attribute there in any case, which might help
 349             # downstream.
 350
 351             # Title and description...
 352             title = param.pop('label', '')
 353             descr = param.pop('doc', '').rstrip('\n')
 354             if title:
 355                 param['title'] = title
 356             if descr:
 357                 param['description'] = descr
 358
 359             # Fill in the value from the current job order, if any.
 360             param_id = shortname(param.pop('id'))
 361             value = job_params.get(param_id)
 362             if value is None:
 363                 pass
 364             elif not isinstance(value, dict):
 365                 param['value'] = value
 366             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 367                 param['value'] = value['location'][5:]
 368
 369             spec['script_parameters'][param_id] = param
 370         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 371         return spec
 372
 373     def save(self):
 374         job_spec = self.pipeline_component_spec()
 375         response = self.runner.api.pipeline_templates().create(body={
 376             "components": {
 377                 self.job.name: job_spec,
 378             },
 379             "name": self.job.name,
 380             "owner_uuid": self.runner.project_uuid,
 381         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 382         self.uuid = response["uuid"]
 383         logger.info("Created template %s", self.uuid)