sdk/cwl/arvados_cwl/arvjob.py

   1 import logging
   2 import re
   3 import copy
   4
   5 from cwltool.process import get_feature, shortname
   6 from cwltool.errors import WorkflowException
   7 from cwltool.draft2tool import revmap_file, CommandLineTool
   8 from cwltool.load_tool import fetch_document
   9 from cwltool.builder import Builder
  10
  11 import arvados.collection
  12
  13 from .arvdocker import arv_docker_get_image
  14 from .runner import Runner
  15 from .pathmapper import InitialWorkDirPathMapper
  16 from . import done
  17
  18 logger = logging.getLogger('arvados.cwl-runner')
  19
  20 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
  21 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
  22 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
  23
  24 class ArvadosJob(object):
  25     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  26
  27     def __init__(self, runner):
  28         self.arvrunner = runner
  29         self.running = False
  30         self.uuid = None
  31
  32     def run(self, dry_run=False, pull_image=True, **kwargs):
  33         script_parameters = {
  34             "command": self.command_line
  35         }
  36         runtime_constraints = {}
  37
  38         if self.generatefiles["listing"]:
  39             vwd = arvados.collection.Collection()
  40             script_parameters["task.vwd"] = {}
  41             generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
  42                                         separateDirs=False)
  43             for f, p in generatemapper.items():
  44                 if p.type == "CreateFile":
  45                     with vwd.open(p.target, "w") as n:
  46                         n.write(p.resolved.encode("utf-8"))
  47             vwd.save_new()
  48             for f, p in generatemapper.items():
  49                 if p.type == "File":
  50                     script_parameters["task.vwd"][p.target] = p.resolved
  51                 if p.type == "CreateFile":
  52                     script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  53
  54         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  55         if self.environment:
  56             script_parameters["task.env"].update(self.environment)
  57
  58         if self.stdin:
  59             script_parameters["task.stdin"] = self.stdin
  60
  61         if self.stdout:
  62             script_parameters["task.stdout"] = self.stdout
  63
  64         if self.stderr:
  65             script_parameters["task.stderr"] = self.stderr
  66
  67         if self.successCodes:
  68             script_parameters["task.successCodes"] = self.successCodes
  69         if self.temporaryFailCodes:
  70             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  71         if self.permanentFailCodes:
  72             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  73
  74         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  75         if docker_req and kwargs.get("use_container") is not False:
  76             runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  77         else:
  78             runtime_constraints["docker_image"] = "arvados/jobs"
  79
  80         resources = self.builder.resources
  81         if resources is not None:
  82             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  83             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  84             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  85
  86         filters = [["repository", "=", "arvados"],
  87                    ["script", "=", "crunchrunner"],
  88                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
  89         if not self.arvrunner.ignore_docker_for_reuse:
  90             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
  91
  92         try:
  93             response = self.arvrunner.api.jobs().create(
  94                 body={
  95                     "owner_uuid": self.arvrunner.project_uuid,
  96                     "script": "crunchrunner",
  97                     "repository": "arvados",
  98                     "script_version": "master",
  99                     "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 100                     "script_parameters": {"tasks": [script_parameters]},
 101                     "runtime_constraints": runtime_constraints
 102                 },
 103                 filters=filters,
 104                 find_or_create=kwargs.get("enable_reuse", True)
 105             ).execute(num_retries=self.arvrunner.num_retries)
 106
 107             self.arvrunner.processes[response["uuid"]] = self
 108
 109             self.update_pipeline_component(response)
 110
 111             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 112
 113             if response["state"] in ("Complete", "Failed", "Cancelled"):
 114                 self.done(response)
 115         except Exception as e:
 116             logger.error("Got error %s" % str(e))
 117             self.output_callback({}, "permanentFail")
 118
 119     def update_pipeline_component(self, record):
 120         if self.arvrunner.pipeline:
 121             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 122             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 123                                                                                  body={
 124                                                                                     "components": self.arvrunner.pipeline["components"]
 125                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 126         if self.arvrunner.uuid:
 127             try:
 128                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 129                 if job:
 130                     components = job["components"]
 131                     components[self.name] = record["uuid"]
 132                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 133                         body={
 134                             "components": components
 135                         }).execute(num_retries=self.arvrunner.num_retries)
 136             except Exception as e:
 137                 logger.info("Error adding to components: %s", e)
 138
 139     def done(self, record):
 140         try:
 141             self.update_pipeline_component(record)
 142         except:
 143             pass
 144
 145         try:
 146             if record["state"] == "Complete":
 147                 processStatus = "success"
 148             else:
 149                 processStatus = "permanentFail"
 150
 151             outputs = {}
 152             try:
 153                 if record["output"]:
 154                     logc = arvados.collection.Collection(record["log"])
 155                     log = logc.open(logc.keys()[0])
 156                     tmpdir = None
 157                     outdir = None
 158                     keepdir = None
 159                     for l in log:
 160                         # Determine the tmpdir, outdir and keepdir paths from
 161                         # the job run.  Unfortunately, we can't take the first
 162                         # values we find (which are expected to be near the
 163                         # top) and stop scanning because if the node fails and
 164                         # the job restarts on a different node these values
 165                         # will different runs, and we need to know about the
 166                         # final run that actually produced output.
 167
 168                         g = tmpdirre.match(l)
 169                         if g:
 170                             tmpdir = g.group(1)
 171                         g = outdirre.match(l)
 172                         if g:
 173                             outdir = g.group(1)
 174                         g = keepre.match(l)
 175                         if g:
 176                             keepdir = g.group(1)
 177
 178                     outputs = done.done(self, record, tmpdir, outdir, keepdir)
 179             except WorkflowException as e:
 180                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 181                 processStatus = "permanentFail"
 182             except Exception as e:
 183                 logger.exception("Got unknown exception while collecting job outputs:")
 184                 processStatus = "permanentFail"
 185
 186             self.output_callback(outputs, processStatus)
 187         finally:
 188             del self.arvrunner.processes[record["uuid"]]
 189
 190
 191 class RunnerJob(Runner):
 192     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 193
 194     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 195         """Create an Arvados job specification for this workflow.
 196
 197         The returned dict can be used to create a job (i.e., passed as
 198         the +body+ argument to jobs().create()), or as a component in
 199         a pipeline template or pipeline instance.
 200         """
 201
 202         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 203
 204         self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"])[1]
 205         return {
 206             "script": "cwl-runner",
 207             "script_version": "master",
 208             "repository": "arvados",
 209             "script_parameters": self.job_order,
 210             "runtime_constraints": {
 211                 "docker_image": "arvados/jobs"
 212             }
 213         }
 214
 215     def run(self, *args, **kwargs):
 216         job_spec = self.arvados_job_spec(*args, **kwargs)
 217         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 218
 219         response = self.arvrunner.api.jobs().create(
 220             body=job_spec,
 221             find_or_create=self.enable_reuse
 222         ).execute(num_retries=self.arvrunner.num_retries)
 223
 224         self.uuid = response["uuid"]
 225         self.arvrunner.processes[self.uuid] = self
 226
 227         logger.info("Submitted job %s", response["uuid"])
 228
 229         if kwargs.get("submit"):
 230             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 231                 body={
 232                     "owner_uuid": self.arvrunner.project_uuid,
 233                     "name": shortname(self.tool.tool["id"]),
 234                     "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } },
 235                     "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries)
 236
 237         if response["state"] in ("Complete", "Failed", "Cancelled"):
 238             self.done(response)
 239
 240
 241 class RunnerTemplate(object):
 242     """An Arvados pipeline template that invokes a CWL workflow."""
 243
 244     type_to_dataclass = {
 245         'boolean': 'boolean',
 246         'File': 'File',
 247         'float': 'number',
 248         'int': 'number',
 249         'string': 'text',
 250     }
 251
 252     def __init__(self, runner, tool, job_order, enable_reuse):
 253         self.runner = runner
 254         self.tool = tool
 255         self.job = RunnerJob(
 256             runner=runner,
 257             tool=tool,
 258             job_order=job_order,
 259             enable_reuse=enable_reuse)
 260
 261     def pipeline_component_spec(self):
 262         """Return a component that Workbench and a-r-p-i will understand.
 263
 264         Specifically, translate CWL input specs to Arvados pipeline
 265         format, like {"dataclass":"File","value":"xyz"}.
 266         """
 267         spec = self.job.arvados_job_spec()
 268
 269         # Most of the component spec is exactly the same as the job
 270         # spec (script, script_version, etc.).
 271         # spec['script_parameters'] isn't right, though. A component
 272         # spec's script_parameters hash is a translation of
 273         # self.tool.tool['inputs'] with defaults/overrides taken from
 274         # the job order. So we move the job parameters out of the way
 275         # and build a new spec['script_parameters'].
 276         job_params = spec['script_parameters']
 277         spec['script_parameters'] = {}
 278
 279         for param in self.tool.tool['inputs']:
 280             param = copy.deepcopy(param)
 281
 282             # Data type and "required" flag...
 283             types = param['type']
 284             if not isinstance(types, list):
 285                 types = [types]
 286             param['required'] = 'null' not in types
 287             non_null_types = set(types) - set(['null'])
 288             if len(non_null_types) == 1:
 289                 the_type = [c for c in non_null_types][0]
 290                 dataclass = self.type_to_dataclass.get(the_type)
 291                 if dataclass:
 292                     param['dataclass'] = dataclass
 293             # Note: If we didn't figure out a single appropriate
 294             # dataclass, we just left that attribute out.  We leave
 295             # the "type" attribute there in any case, which might help
 296             # downstream.
 297
 298             # Title and description...
 299             title = param.pop('label', '')
 300             descr = param.pop('doc', '').rstrip('\n')
 301             if title:
 302                 param['title'] = title
 303             if descr:
 304                 param['description'] = descr
 305
 306             # Fill in the value from the current job order, if any.
 307             param_id = shortname(param.pop('id'))
 308             value = job_params.get(param_id)
 309             if value is None:
 310                 pass
 311             elif not isinstance(value, dict):
 312                 param['value'] = value
 313             elif param.get('dataclass') == 'File' and value.get('location'):
 314                 param['value'] = value['location']
 315
 316             spec['script_parameters'][param_id] = param
 317         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 318         return spec
 319
 320     def save(self):
 321         job_spec = self.pipeline_component_spec()
 322         response = self.runner.api.pipeline_templates().create(body={
 323             "components": {
 324                 self.job.name: job_spec,
 325             },
 326             "name": self.job.name,
 327             "owner_uuid": self.runner.project_uuid,
 328         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 329         self.uuid = response["uuid"]
 330         logger.info("Created template %s", self.uuid)