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()
  44                 script_parameters["task.vwd"] = {}
  45                 generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
  46                                                           separateDirs=False)
  47
  48                 with Perf(metrics, "createfiles %s" % self.name):
  49                     for f, p in generatemapper.items():
  50                         if p.type == "CreateFile":
  51                             with vwd.open(p.target, "w") as n:
  52                                 n.write(p.resolved.encode("utf-8"))
  53
  54                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
  55                     vwd.save_new()
  56
  57                 for f, p in generatemapper.items():
  58                     if p.type == "File":
  59                         script_parameters["task.vwd"][p.target] = p.resolved
  60                     if p.type == "CreateFile":
  61                         script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  62
  63         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  64         if self.environment:
  65             script_parameters["task.env"].update(self.environment)
  66
  67         if self.stdin:
  68             script_parameters["task.stdin"] = self.stdin
  69
  70         if self.stdout:
  71             script_parameters["task.stdout"] = self.stdout
  72
  73         if self.stderr:
  74             script_parameters["task.stderr"] = self.stderr
  75
  76         if self.successCodes:
  77             script_parameters["task.successCodes"] = self.successCodes
  78         if self.temporaryFailCodes:
  79             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  80         if self.permanentFailCodes:
  81             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  82
  83         with Perf(metrics, "arv_docker_get_image %s" % self.name):
  84             (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  85             if docker_req and kwargs.get("use_container") is not False:
  86                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  87             else:
  88                 runtime_constraints["docker_image"] = "arvados/jobs"
  89
  90         resources = self.builder.resources
  91         if resources is not None:
  92             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  93             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  94             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  95
  96         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
  97         if runtime_req:
  98             runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
  99
 100         filters = [["repository", "=", "arvados"],
 101                    ["script", "=", "crunchrunner"],
 102                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
 103         if not self.arvrunner.ignore_docker_for_reuse:
 104             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 105
 106         try:
 107             with Perf(metrics, "create %s" % self.name):
 108                 response = self.arvrunner.api.jobs().create(
 109                     body={
 110                         "owner_uuid": self.arvrunner.project_uuid,
 111                         "script": "crunchrunner",
 112                         "repository": "arvados",
 113                         "script_version": "master",
 114                         "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 115                         "script_parameters": {"tasks": [script_parameters]},
 116                         "runtime_constraints": runtime_constraints
 117                     },
 118                     filters=filters,
 119                     find_or_create=kwargs.get("enable_reuse", True)
 120                 ).execute(num_retries=self.arvrunner.num_retries)
 121
 122             self.arvrunner.processes[response["uuid"]] = self
 123
 124             self.update_pipeline_component(response)
 125
 126             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 127
 128             if response["state"] in ("Complete", "Failed", "Cancelled"):
 129                 with Perf(metrics, "done %s" % self.name):
 130                     self.done(response)
 131         except Exception as e:
 132             logger.error("Got error %s" % str(e))
 133             self.output_callback({}, "permanentFail")
 134
 135     def update_pipeline_component(self, record):
 136         if self.arvrunner.pipeline:
 137             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 138             with Perf(metrics, "update_pipeline_component %s" % self.name):
 139                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 140                                                                                  body={
 141                                                                                     "components": self.arvrunner.pipeline["components"]
 142                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 143         if self.arvrunner.uuid:
 144             try:
 145                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 146                 if job:
 147                     components = job["components"]
 148                     components[self.name] = record["uuid"]
 149                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 150                         body={
 151                             "components": components
 152                         }).execute(num_retries=self.arvrunner.num_retries)
 153             except Exception as e:
 154                 logger.info("Error adding to components: %s", e)
 155
 156     def done(self, record):
 157         try:
 158             self.update_pipeline_component(record)
 159         except:
 160             pass
 161
 162         try:
 163             if record["state"] == "Complete":
 164                 processStatus = "success"
 165             else:
 166                 processStatus = "permanentFail"
 167
 168             outputs = {}
 169             try:
 170                 if record["output"]:
 171                     with Perf(metrics, "inspect log %s" % self.name):
 172                         logc = arvados.collection.Collection(record["log"])
 173                         log = logc.open(logc.keys()[0])
 174                         tmpdir = None
 175                         outdir = None
 176                         keepdir = None
 177                         for l in log:
 178                             # Determine the tmpdir, outdir and keepdir paths from
 179                             # the job run.  Unfortunately, we can't take the first
 180                             # values we find (which are expected to be near the
 181                             # top) and stop scanning because if the node fails and
 182                             # the job restarts on a different node these values
 183                             # will different runs, and we need to know about the
 184                             # final run that actually produced output.
 185
 186                             g = tmpdirre.match(l)
 187                             if g:
 188                                 tmpdir = g.group(1)
 189                             g = outdirre.match(l)
 190                             if g:
 191                                 outdir = g.group(1)
 192                             g = keepre.match(l)
 193                             if g:
 194                                 keepdir = g.group(1)
 195
 196                     with Perf(metrics, "output collection %s" % self.name):
 197                         outputs = done.done(self, record, tmpdir, outdir, keepdir)
 198             except WorkflowException as e:
 199                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 200                 processStatus = "permanentFail"
 201                 outputs = None
 202             except Exception as e:
 203                 logger.exception("Got unknown exception while collecting job outputs:")
 204                 processStatus = "permanentFail"
 205                 outputs = None
 206
 207             self.output_callback(outputs, processStatus)
 208         finally:
 209             del self.arvrunner.processes[record["uuid"]]
 210
 211
 212 class RunnerJob(Runner):
 213     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 214
 215     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 216         """Create an Arvados job specification for this workflow.
 217
 218         The returned dict can be used to create a job (i.e., passed as
 219         the +body+ argument to jobs().create()), or as a component in
 220         a pipeline template or pipeline instance.
 221         """
 222
 223         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 224
 225         self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"]).target[5:]
 226         return {
 227             "script": "cwl-runner",
 228             "script_version": "master",
 229             "repository": "arvados",
 230             "script_parameters": self.job_order,
 231             "runtime_constraints": {
 232                 "docker_image": "arvados/jobs"
 233             }
 234         }
 235
 236     def run(self, *args, **kwargs):
 237         job_spec = self.arvados_job_spec(*args, **kwargs)
 238         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 239
 240         response = self.arvrunner.api.jobs().create(
 241             body=job_spec,
 242             find_or_create=self.enable_reuse
 243         ).execute(num_retries=self.arvrunner.num_retries)
 244
 245         self.uuid = response["uuid"]
 246         self.arvrunner.processes[self.uuid] = self
 247
 248         logger.info("Submitted job %s", response["uuid"])
 249
 250         if kwargs.get("submit"):
 251             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 252                 body={
 253                     "owner_uuid": self.arvrunner.project_uuid,
 254                     "name": shortname(self.tool.tool["id"]),
 255                     "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } },
 256                     "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries)
 257
 258         if response["state"] in ("Complete", "Failed", "Cancelled"):
 259             self.done(response)
 260
 261
 262 class RunnerTemplate(object):
 263     """An Arvados pipeline template that invokes a CWL workflow."""
 264
 265     type_to_dataclass = {
 266         'boolean': 'boolean',
 267         'File': 'File',
 268         'Directory': 'Collection',
 269         'float': 'number',
 270         'int': 'number',
 271         'string': 'text',
 272     }
 273
 274     def __init__(self, runner, tool, job_order, enable_reuse):
 275         self.runner = runner
 276         self.tool = tool
 277         self.job = RunnerJob(
 278             runner=runner,
 279             tool=tool,
 280             job_order=job_order,
 281             enable_reuse=enable_reuse)
 282
 283     def pipeline_component_spec(self):
 284         """Return a component that Workbench and a-r-p-i will understand.
 285
 286         Specifically, translate CWL input specs to Arvados pipeline
 287         format, like {"dataclass":"File","value":"xyz"}.
 288         """
 289         spec = self.job.arvados_job_spec()
 290
 291         # Most of the component spec is exactly the same as the job
 292         # spec (script, script_version, etc.).
 293         # spec['script_parameters'] isn't right, though. A component
 294         # spec's script_parameters hash is a translation of
 295         # self.tool.tool['inputs'] with defaults/overrides taken from
 296         # the job order. So we move the job parameters out of the way
 297         # and build a new spec['script_parameters'].
 298         job_params = spec['script_parameters']
 299         spec['script_parameters'] = {}
 300
 301         for param in self.tool.tool['inputs']:
 302             param = copy.deepcopy(param)
 303
 304             # Data type and "required" flag...
 305             types = param['type']
 306             if not isinstance(types, list):
 307                 types = [types]
 308             param['required'] = 'null' not in types
 309             non_null_types = set(types) - set(['null'])
 310             if len(non_null_types) == 1:
 311                 the_type = [c for c in non_null_types][0]
 312                 dataclass = self.type_to_dataclass.get(the_type)
 313                 if dataclass:
 314                     param['dataclass'] = dataclass
 315             # Note: If we didn't figure out a single appropriate
 316             # dataclass, we just left that attribute out.  We leave
 317             # the "type" attribute there in any case, which might help
 318             # downstream.
 319
 320             # Title and description...
 321             title = param.pop('label', '')
 322             descr = param.pop('doc', '').rstrip('\n')
 323             if title:
 324                 param['title'] = title
 325             if descr:
 326                 param['description'] = descr
 327
 328             # Fill in the value from the current job order, if any.
 329             param_id = shortname(param.pop('id'))
 330             value = job_params.get(param_id)
 331             if value is None:
 332                 pass
 333             elif not isinstance(value, dict):
 334                 param['value'] = value
 335             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 336                 param['value'] = value['location'][5:]
 337
 338             spec['script_parameters'][param_id] = param
 339         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 340         return spec
 341
 342     def save(self):
 343         job_spec = self.pipeline_component_spec()
 344         response = self.runner.api.pipeline_templates().create(body={
 345             "components": {
 346                 self.job.name: job_spec,
 347             },
 348             "name": self.job.name,
 349             "owner_uuid": self.runner.project_uuid,
 350         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 351         self.uuid = response["uuid"]
 352         logger.info("Created template %s", self.uuid)