sdk/cwl/arvados_cwl/arvjob.py

   1 import logging
   2 import re
   3 import copy
   4 import json
   5
   6 from cwltool.process import get_feature, shortname
   7 from cwltool.errors import WorkflowException
   8 from cwltool.draft2tool import revmap_file, CommandLineTool
   9 from cwltool.load_tool import fetch_document
  10 from cwltool.builder import Builder
  11
  12 import arvados.collection
  13
  14 from .arvdocker import arv_docker_get_image
  15 from .runner import Runner
  16 from .pathmapper import InitialWorkDirPathMapper
  17 from .perf import Perf
  18 from . import done
  19
  20 logger = logging.getLogger('arvados.cwl-runner')
  21 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  22
  23 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
  24 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
  25 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
  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                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  89             else:
  90                 runtime_constraints["docker_image"] = "arvados/jobs"
  91
  92         resources = self.builder.resources
  93         if resources is not None:
  94             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  95             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  96             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  97
  98         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
  99         if runtime_req:
 100             runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
 101
 102         filters = [["repository", "=", "arvados"],
 103                    ["script", "=", "crunchrunner"],
 104                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
 105         if not self.arvrunner.ignore_docker_for_reuse:
 106             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 107
 108         try:
 109             with Perf(metrics, "create %s" % self.name):
 110                 response = self.arvrunner.api.jobs().create(
 111                     body={
 112                         "owner_uuid": self.arvrunner.project_uuid,
 113                         "script": "crunchrunner",
 114                         "repository": "arvados",
 115                         "script_version": "master",
 116                         "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 117                         "script_parameters": {"tasks": [script_parameters]},
 118                         "runtime_constraints": runtime_constraints
 119                     },
 120                     filters=filters,
 121                     find_or_create=kwargs.get("enable_reuse", True)
 122                 ).execute(num_retries=self.arvrunner.num_retries)
 123
 124             self.arvrunner.processes[response["uuid"]] = self
 125
 126             self.update_pipeline_component(response)
 127
 128             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 129
 130             if response["state"] in ("Complete", "Failed", "Cancelled"):
 131                 with Perf(metrics, "done %s" % self.name):
 132                     self.done(response)
 133         except Exception as e:
 134             logger.error("Got error %s" % str(e))
 135             self.output_callback({}, "permanentFail")
 136
 137     def update_pipeline_component(self, record):
 138         if self.arvrunner.pipeline:
 139             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 140             with Perf(metrics, "update_pipeline_component %s" % self.name):
 141                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 142                                                                                  body={
 143                                                                                     "components": self.arvrunner.pipeline["components"]
 144                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 145         if self.arvrunner.uuid:
 146             try:
 147                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 148                 if job:
 149                     components = job["components"]
 150                     components[self.name] = record["uuid"]
 151                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 152                         body={
 153                             "components": components
 154                         }).execute(num_retries=self.arvrunner.num_retries)
 155             except Exception as e:
 156                 logger.info("Error adding to components: %s", e)
 157
 158     def done(self, record):
 159         try:
 160             self.update_pipeline_component(record)
 161         except:
 162             pass
 163
 164         try:
 165             if record["state"] == "Complete":
 166                 processStatus = "success"
 167             else:
 168                 processStatus = "permanentFail"
 169
 170             outputs = {}
 171             try:
 172                 if record["output"]:
 173                     with Perf(metrics, "inspect log %s" % self.name):
 174                         logc = arvados.collection.CollectionReader(record["log"],
 175                                                                    api_client=self.arvrunner.api,
 176                                                                    keep_client=self.arvrunner.keep_client,
 177                                                                    num_retries=self.arvrunner.num_retries)
 178                         log = logc.open(logc.keys()[0])
 179                         tmpdir = None
 180                         outdir = None
 181                         keepdir = None
 182                         for l in log:
 183                             # Determine the tmpdir, outdir and keepdir paths from
 184                             # the job run.  Unfortunately, we can't take the first
 185                             # values we find (which are expected to be near the
 186                             # top) and stop scanning because if the node fails and
 187                             # the job restarts on a different node these values
 188                             # will different runs, and we need to know about the
 189                             # final run that actually produced output.
 190
 191                             g = tmpdirre.match(l)
 192                             if g:
 193                                 tmpdir = g.group(1)
 194                             g = outdirre.match(l)
 195                             if g:
 196                                 outdir = g.group(1)
 197                             g = keepre.match(l)
 198                             if g:
 199                                 keepdir = g.group(1)
 200
 201                     with Perf(metrics, "output collection %s" % self.name):
 202                         outputs = done.done(self, record, tmpdir, outdir, keepdir)
 203             except WorkflowException as e:
 204                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 205                 processStatus = "permanentFail"
 206                 outputs = None
 207             except Exception as e:
 208                 logger.exception("Got unknown exception while collecting job outputs:")
 209                 processStatus = "permanentFail"
 210                 outputs = None
 211
 212             self.output_callback(outputs, processStatus)
 213         finally:
 214             del self.arvrunner.processes[record["uuid"]]
 215
 216
 217 class RunnerJob(Runner):
 218     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 219
 220     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 221         """Create an Arvados job specification for this workflow.
 222
 223         The returned dict can be used to create a job (i.e., passed as
 224         the +body+ argument to jobs().create()), or as a component in
 225         a pipeline template or pipeline instance.
 226         """
 227
 228         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 229
 230         self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"]).target[5:]
 231         if self.output_name:
 232             self.job_order["arv:output_name"] = self.output_name
 233         return {
 234             "script": "cwl-runner",
 235             "script_version": "master",
 236             "repository": "arvados",
 237             "script_parameters": self.job_order,
 238             "runtime_constraints": {
 239                 "docker_image": "arvados/jobs"
 240             }
 241         }
 242
 243     def run(self, *args, **kwargs):
 244         job_spec = self.arvados_job_spec(*args, **kwargs)
 245         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 246
 247         response = self.arvrunner.api.jobs().create(
 248             body=job_spec,
 249             find_or_create=self.enable_reuse
 250         ).execute(num_retries=self.arvrunner.num_retries)
 251
 252         self.uuid = response["uuid"]
 253         self.arvrunner.processes[self.uuid] = self
 254
 255         logger.info("Submitted job %s", response["uuid"])
 256
 257         if kwargs.get("submit"):
 258             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 259                 body={
 260                     "owner_uuid": self.arvrunner.project_uuid,
 261                     "name": shortname(self.tool.tool["id"]),
 262                     "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } },
 263                     "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries)
 264
 265         if response["state"] in ("Complete", "Failed", "Cancelled"):
 266             self.done(response)
 267
 268
 269 class RunnerTemplate(object):
 270     """An Arvados pipeline template that invokes a CWL workflow."""
 271
 272     type_to_dataclass = {
 273         'boolean': 'boolean',
 274         'File': 'File',
 275         'Directory': 'Collection',
 276         'float': 'number',
 277         'int': 'number',
 278         'string': 'text',
 279     }
 280
 281     def __init__(self, runner, tool, job_order, enable_reuse):
 282         self.runner = runner
 283         self.tool = tool
 284         self.job = RunnerJob(
 285             runner=runner,
 286             tool=tool,
 287             job_order=job_order,
 288             enable_reuse=enable_reuse,
 289             output_name=None)
 290
 291     def pipeline_component_spec(self):
 292         """Return a component that Workbench and a-r-p-i will understand.
 293
 294         Specifically, translate CWL input specs to Arvados pipeline
 295         format, like {"dataclass":"File","value":"xyz"}.
 296         """
 297         spec = self.job.arvados_job_spec()
 298
 299         # Most of the component spec is exactly the same as the job
 300         # spec (script, script_version, etc.).
 301         # spec['script_parameters'] isn't right, though. A component
 302         # spec's script_parameters hash is a translation of
 303         # self.tool.tool['inputs'] with defaults/overrides taken from
 304         # the job order. So we move the job parameters out of the way
 305         # and build a new spec['script_parameters'].
 306         job_params = spec['script_parameters']
 307         spec['script_parameters'] = {}
 308
 309         for param in self.tool.tool['inputs']:
 310             param = copy.deepcopy(param)
 311
 312             # Data type and "required" flag...
 313             types = param['type']
 314             if not isinstance(types, list):
 315                 types = [types]
 316             param['required'] = 'null' not in types
 317             non_null_types = set(types) - set(['null'])
 318             if len(non_null_types) == 1:
 319                 the_type = [c for c in non_null_types][0]
 320                 dataclass = self.type_to_dataclass.get(the_type)
 321                 if dataclass:
 322                     param['dataclass'] = dataclass
 323             # Note: If we didn't figure out a single appropriate
 324             # dataclass, we just left that attribute out.  We leave
 325             # the "type" attribute there in any case, which might help
 326             # downstream.
 327
 328             # Title and description...
 329             title = param.pop('label', '')
 330             descr = param.pop('doc', '').rstrip('\n')
 331             if title:
 332                 param['title'] = title
 333             if descr:
 334                 param['description'] = descr
 335
 336             # Fill in the value from the current job order, if any.
 337             param_id = shortname(param.pop('id'))
 338             value = job_params.get(param_id)
 339             if value is None:
 340                 pass
 341             elif not isinstance(value, dict):
 342                 param['value'] = value
 343             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 344                 param['value'] = value['location'][5:]
 345
 346             spec['script_parameters'][param_id] = param
 347         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 348         return spec
 349
 350     def save(self):
 351         job_spec = self.pipeline_component_spec()
 352         response = self.runner.api.pipeline_templates().create(body={
 353             "components": {
 354                 self.job.name: job_spec,
 355             },
 356             "name": self.job.name,
 357             "owner_uuid": self.runner.project_uuid,
 358         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 359         self.uuid = response["uuid"]
 360         logger.info("Created template %s", self.uuid)