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
  22 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
  23 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
  24 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
  25
  26 class ArvadosJob(object):
  27     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  28
  29     def __init__(self, runner):
  30         self.arvrunner = runner
  31         self.running = False
  32         self.uuid = None
  33
  34     def run(self, dry_run=False, pull_image=True, **kwargs):
  35         script_parameters = {
  36             "command": self.command_line
  37         }
  38         runtime_constraints = {}
  39
  40         if self.generatefiles["listing"]:
  41             vwd = arvados.collection.Collection()
  42             script_parameters["task.vwd"] = {}
  43             generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
  44                                         separateDirs=False)
  45             for f, p in generatemapper.items():
  46                 if p.type == "CreateFile":
  47                     with vwd.open(p.target, "w") as n:
  48                         n.write(p.resolved.encode("utf-8"))
  49             vwd.save_new()
  50             for f, p in generatemapper.items():
  51                 if p.type == "File":
  52                     script_parameters["task.vwd"][p.target] = p.resolved
  53                 if p.type == "CreateFile":
  54                     script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  55
  56         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  57         if self.environment:
  58             script_parameters["task.env"].update(self.environment)
  59
  60         if self.stdin:
  61             script_parameters["task.stdin"] = self.stdin
  62
  63         if self.stdout:
  64             script_parameters["task.stdout"] = self.stdout
  65
  66         if self.stderr:
  67             script_parameters["task.stderr"] = self.stderr
  68
  69         if self.successCodes:
  70             script_parameters["task.successCodes"] = self.successCodes
  71         if self.temporaryFailCodes:
  72             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  73         if self.permanentFailCodes:
  74             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  75
  76         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  77         if docker_req and kwargs.get("use_container") is not False:
  78             runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  79         else:
  80             runtime_constraints["docker_image"] = "arvados/jobs"
  81
  82         resources = self.builder.resources
  83         if resources is not None:
  84             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  85             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  86             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  87
  88         filters = [["repository", "=", "arvados"],
  89                    ["script", "=", "crunchrunner"],
  90                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
  91         if not self.arvrunner.ignore_docker_for_reuse:
  92             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
  93
  94         try:
  95             with Perf(logger, "create %s" % self.name):
  96                 response = self.arvrunner.api.jobs().create(
  97                     body={
  98                         "owner_uuid": self.arvrunner.project_uuid,
  99                         "script": "crunchrunner",
 100                         "repository": "arvados",
 101                         "script_version": "master",
 102                         "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 103                         "script_parameters": {"tasks": [script_parameters]},
 104                         "runtime_constraints": runtime_constraints
 105                     },
 106                     filters=filters,
 107                     find_or_create=kwargs.get("enable_reuse", True)
 108                 ).execute(num_retries=self.arvrunner.num_retries)
 109
 110             self.arvrunner.processes[response["uuid"]] = self
 111
 112             self.update_pipeline_component(response)
 113
 114             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 115
 116             if response["state"] in ("Complete", "Failed", "Cancelled"):
 117                 with Perf(logger, "done %s" % self.name):
 118                     self.done(response)
 119         except Exception as e:
 120             logger.error("Got error %s" % str(e))
 121             self.output_callback({}, "permanentFail")
 122
 123     def update_pipeline_component(self, record):
 124         if self.arvrunner.pipeline:
 125             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 126             with Perf(logger, "update_pipeline_component %s" % self.name):
 127                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 128                                                                                  body={
 129                                                                                     "components": self.arvrunner.pipeline["components"]
 130                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 131         if self.arvrunner.uuid:
 132             try:
 133                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 134                 if job:
 135                     components = job["components"]
 136                     components[self.name] = record["uuid"]
 137                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 138                         body={
 139                             "components": components
 140                         }).execute(num_retries=self.arvrunner.num_retries)
 141             except Exception as e:
 142                 logger.info("Error adding to components: %s", e)
 143
 144     def done(self, record):
 145         try:
 146             self.update_pipeline_component(record)
 147         except:
 148             pass
 149
 150         try:
 151             if record["state"] == "Complete":
 152                 processStatus = "success"
 153             else:
 154                 processStatus = "permanentFail"
 155
 156             outputs = {}
 157             try:
 158                 if record["output"]:
 159                     with Perf(logger, "inspect log %s" % self.name):
 160                         logc = arvados.collection.Collection(record["log"])
 161                         log = logc.open(logc.keys()[0])
 162                         tmpdir = None
 163                         outdir = None
 164                         keepdir = None
 165                         for l in log:
 166                             # Determine the tmpdir, outdir and keepdir paths from
 167                             # the job run.  Unfortunately, we can't take the first
 168                             # values we find (which are expected to be near the
 169                             # top) and stop scanning because if the node fails and
 170                             # the job restarts on a different node these values
 171                             # will different runs, and we need to know about the
 172                             # final run that actually produced output.
 173
 174                             g = tmpdirre.match(l)
 175                             if g:
 176                                 tmpdir = g.group(1)
 177                             g = outdirre.match(l)
 178                             if g:
 179                                 outdir = g.group(1)
 180                             g = keepre.match(l)
 181                             if g:
 182                                 keepdir = g.group(1)
 183
 184                     with Perf(logger, "output collection %s" % self.name):
 185                         outputs = done.done(self, record, tmpdir, outdir, keepdir)
 186             except WorkflowException as e:
 187                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 188                 processStatus = "permanentFail"
 189                 outputs = None
 190             except Exception as e:
 191                 logger.exception("Got unknown exception while collecting job outputs:")
 192                 processStatus = "permanentFail"
 193                 outputs = None
 194
 195             self.output_callback(outputs, processStatus)
 196         finally:
 197             del self.arvrunner.processes[record["uuid"]]
 198
 199
 200 class RunnerJob(Runner):
 201     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 202
 203     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 204         """Create an Arvados job specification for this workflow.
 205
 206         The returned dict can be used to create a job (i.e., passed as
 207         the +body+ argument to jobs().create()), or as a component in
 208         a pipeline template or pipeline instance.
 209         """
 210
 211         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 212
 213         self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"]).target[5:]
 214         return {
 215             "script": "cwl-runner",
 216             "script_version": "master",
 217             "repository": "arvados",
 218             "script_parameters": self.job_order,
 219             "runtime_constraints": {
 220                 "docker_image": "arvados/jobs"
 221             }
 222         }
 223
 224     def run(self, *args, **kwargs):
 225         job_spec = self.arvados_job_spec(*args, **kwargs)
 226         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 227
 228         response = self.arvrunner.api.jobs().create(
 229             body=job_spec,
 230             find_or_create=self.enable_reuse
 231         ).execute(num_retries=self.arvrunner.num_retries)
 232
 233         self.uuid = response["uuid"]
 234         self.arvrunner.processes[self.uuid] = self
 235
 236         logger.info("Submitted job %s", response["uuid"])
 237
 238         if kwargs.get("submit"):
 239             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 240                 body={
 241                     "owner_uuid": self.arvrunner.project_uuid,
 242                     "name": shortname(self.tool.tool["id"]),
 243                     "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } },
 244                     "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries)
 245
 246         if response["state"] in ("Complete", "Failed", "Cancelled"):
 247             self.done(response)
 248
 249
 250 class RunnerTemplate(object):
 251     """An Arvados pipeline template that invokes a CWL workflow."""
 252
 253     type_to_dataclass = {
 254         'boolean': 'boolean',
 255         'File': 'File',
 256         'Directory': 'Collection',
 257         'float': 'number',
 258         'int': 'number',
 259         'string': 'text',
 260     }
 261
 262     def __init__(self, runner, tool, job_order, enable_reuse):
 263         self.runner = runner
 264         self.tool = tool
 265         self.job = RunnerJob(
 266             runner=runner,
 267             tool=tool,
 268             job_order=job_order,
 269             enable_reuse=enable_reuse)
 270
 271     def pipeline_component_spec(self):
 272         """Return a component that Workbench and a-r-p-i will understand.
 273
 274         Specifically, translate CWL input specs to Arvados pipeline
 275         format, like {"dataclass":"File","value":"xyz"}.
 276         """
 277         spec = self.job.arvados_job_spec()
 278
 279         # Most of the component spec is exactly the same as the job
 280         # spec (script, script_version, etc.).
 281         # spec['script_parameters'] isn't right, though. A component
 282         # spec's script_parameters hash is a translation of
 283         # self.tool.tool['inputs'] with defaults/overrides taken from
 284         # the job order. So we move the job parameters out of the way
 285         # and build a new spec['script_parameters'].
 286         job_params = spec['script_parameters']
 287         spec['script_parameters'] = {}
 288
 289         for param in self.tool.tool['inputs']:
 290             param = copy.deepcopy(param)
 291
 292             # Data type and "required" flag...
 293             types = param['type']
 294             if not isinstance(types, list):
 295                 types = [types]
 296             param['required'] = 'null' not in types
 297             non_null_types = set(types) - set(['null'])
 298             if len(non_null_types) == 1:
 299                 the_type = [c for c in non_null_types][0]
 300                 dataclass = self.type_to_dataclass.get(the_type)
 301                 if dataclass:
 302                     param['dataclass'] = dataclass
 303             # Note: If we didn't figure out a single appropriate
 304             # dataclass, we just left that attribute out.  We leave
 305             # the "type" attribute there in any case, which might help
 306             # downstream.
 307
 308             # Title and description...
 309             title = param.pop('label', '')
 310             descr = param.pop('doc', '').rstrip('\n')
 311             if title:
 312                 param['title'] = title
 313             if descr:
 314                 param['description'] = descr
 315
 316             # Fill in the value from the current job order, if any.
 317             param_id = shortname(param.pop('id'))
 318             value = job_params.get(param_id)
 319             if value is None:
 320                 pass
 321             elif not isinstance(value, dict):
 322                 param['value'] = value
 323             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 324                 param['value'] = value['location'][5:]
 325
 326             spec['script_parameters'][param_id] = param
 327         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 328         return spec
 329
 330     def save(self):
 331         job_spec = self.pipeline_component_spec()
 332         response = self.runner.api.pipeline_templates().create(body={
 333             "components": {
 334                 self.job.name: job_spec,
 335             },
 336             "name": self.job.name,
 337             "owner_uuid": self.runner.project_uuid,
 338         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 339         self.uuid = response["uuid"]
 340         logger.info("Created template %s", self.uuid)