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 .perf import Perf
  17 from . import done
  18
  19 logger = logging.getLogger('arvados.cwl-runner')
  20
  21 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
  22 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
  23 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
  24
  25 class ArvadosJob(object):
  26     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  27
  28     def __init__(self, runner):
  29         self.arvrunner = runner
  30         self.running = False
  31         self.uuid = None
  32
  33     def run(self, dry_run=False, pull_image=True, **kwargs):
  34         script_parameters = {
  35             "command": self.command_line
  36         }
  37         runtime_constraints = {}
  38
  39         if self.generatefiles["listing"]:
  40             vwd = arvados.collection.Collection()
  41             script_parameters["task.vwd"] = {}
  42             generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
  43                                         separateDirs=False)
  44             for f, p in generatemapper.items():
  45                 if p.type == "CreateFile":
  46                     with vwd.open(p.target, "w") as n:
  47                         n.write(p.resolved.encode("utf-8"))
  48             vwd.save_new()
  49             for f, p in generatemapper.items():
  50                 if p.type == "File":
  51                     script_parameters["task.vwd"][p.target] = p.resolved
  52                 if p.type == "CreateFile":
  53                     script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  54
  55         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  56         if self.environment:
  57             script_parameters["task.env"].update(self.environment)
  58
  59         if self.stdin:
  60             script_parameters["task.stdin"] = self.stdin
  61
  62         if self.stdout:
  63             script_parameters["task.stdout"] = self.stdout
  64
  65         if self.stderr:
  66             script_parameters["task.stderr"] = self.stderr
  67
  68         if self.successCodes:
  69             script_parameters["task.successCodes"] = self.successCodes
  70         if self.temporaryFailCodes:
  71             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  72         if self.permanentFailCodes:
  73             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  74
  75         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  76         if docker_req and kwargs.get("use_container") is not False:
  77             runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  78         else:
  79             runtime_constraints["docker_image"] = "arvados/jobs"
  80
  81         resources = self.builder.resources
  82         if resources is not None:
  83             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  84             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  85             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  86
  87         filters = [["repository", "=", "arvados"],
  88                    ["script", "=", "crunchrunner"],
  89                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
  90         if not self.arvrunner.ignore_docker_for_reuse:
  91             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
  92
  93         try:
  94             with Perf(logger, "create %s" % self.name):
  95                 response = self.arvrunner.api.jobs().create(
  96                     body={
  97                         "owner_uuid": self.arvrunner.project_uuid,
  98                         "script": "crunchrunner",
  99                         "repository": "arvados",
 100                         "script_version": "master",
 101                         "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 102                         "script_parameters": {"tasks": [script_parameters]},
 103                         "runtime_constraints": runtime_constraints
 104                     },
 105                     filters=filters,
 106                     find_or_create=kwargs.get("enable_reuse", True)
 107                 ).execute(num_retries=self.arvrunner.num_retries)
 108
 109             self.arvrunner.processes[response["uuid"]] = self
 110
 111             self.update_pipeline_component(response)
 112
 113             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 114
 115             if response["state"] in ("Complete", "Failed", "Cancelled"):
 116                 with Perf(logger, "done %s" % self.name):
 117                     self.done(response)
 118         except Exception as e:
 119             logger.error("Got error %s" % str(e))
 120             self.output_callback({}, "permanentFail")
 121
 122     def update_pipeline_component(self, record):
 123         if self.arvrunner.pipeline:
 124             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 125             with Perf(logger, "update_pipeline_component %s" % self.name):
 126                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 127                                                                                  body={
 128                                                                                     "components": self.arvrunner.pipeline["components"]
 129                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 130         if self.arvrunner.uuid:
 131             try:
 132                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 133                 if job:
 134                     components = job["components"]
 135                     components[self.name] = record["uuid"]
 136                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 137                         body={
 138                             "components": components
 139                         }).execute(num_retries=self.arvrunner.num_retries)
 140             except Exception as e:
 141                 logger.info("Error adding to components: %s", e)
 142
 143     def done(self, record):
 144         try:
 145             self.update_pipeline_component(record)
 146         except:
 147             pass
 148
 149         try:
 150             if record["state"] == "Complete":
 151                 processStatus = "success"
 152             else:
 153                 processStatus = "permanentFail"
 154
 155             outputs = {}
 156             try:
 157                 if record["output"]:
 158                     with Perf(logger, "inspect log %s" % self.name):
 159                         logc = arvados.collection.Collection(record["log"])
 160                         log = logc.open(logc.keys()[0])
 161                         tmpdir = None
 162                         outdir = None
 163                         keepdir = None
 164                         for l in log:
 165                             # Determine the tmpdir, outdir and keepdir paths from
 166                             # the job run.  Unfortunately, we can't take the first
 167                             # values we find (which are expected to be near the
 168                             # top) and stop scanning because if the node fails and
 169                             # the job restarts on a different node these values
 170                             # will different runs, and we need to know about the
 171                             # final run that actually produced output.
 172
 173                             g = tmpdirre.match(l)
 174                             if g:
 175                                 tmpdir = g.group(1)
 176                             g = outdirre.match(l)
 177                             if g:
 178                                 outdir = g.group(1)
 179                             g = keepre.match(l)
 180                             if g:
 181                                 keepdir = g.group(1)
 182
 183                     with Perf(logger, "output collection %s" % self.name):
 184                         outputs = done.done(self, record, tmpdir, outdir, keepdir)
 185             except WorkflowException as e:
 186                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 187                 processStatus = "permanentFail"
 188             except Exception as e:
 189                 logger.exception("Got unknown exception while collecting job outputs:")
 190                 processStatus = "permanentFail"
 191
 192             self.output_callback(outputs, processStatus)
 193         finally:
 194             del self.arvrunner.processes[record["uuid"]]
 195
 196
 197 class RunnerJob(Runner):
 198     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 199
 200     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 201         """Create an Arvados job specification for this workflow.
 202
 203         The returned dict can be used to create a job (i.e., passed as
 204         the +body+ argument to jobs().create()), or as a component in
 205         a pipeline template or pipeline instance.
 206         """
 207
 208         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 209
 210         self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"]).target[5:]
 211         return {
 212             "script": "cwl-runner",
 213             "script_version": "master",
 214             "repository": "arvados",
 215             "script_parameters": self.job_order,
 216             "runtime_constraints": {
 217                 "docker_image": "arvados/jobs"
 218             }
 219         }
 220
 221     def run(self, *args, **kwargs):
 222         job_spec = self.arvados_job_spec(*args, **kwargs)
 223         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 224
 225         response = self.arvrunner.api.jobs().create(
 226             body=job_spec,
 227             find_or_create=self.enable_reuse
 228         ).execute(num_retries=self.arvrunner.num_retries)
 229
 230         self.uuid = response["uuid"]
 231         self.arvrunner.processes[self.uuid] = self
 232
 233         logger.info("Submitted job %s", response["uuid"])
 234
 235         if kwargs.get("submit"):
 236             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 237                 body={
 238                     "owner_uuid": self.arvrunner.project_uuid,
 239                     "name": shortname(self.tool.tool["id"]),
 240                     "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } },
 241                     "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries)
 242
 243         if response["state"] in ("Complete", "Failed", "Cancelled"):
 244             self.done(response)
 245
 246
 247 class RunnerTemplate(object):
 248     """An Arvados pipeline template that invokes a CWL workflow."""
 249
 250     type_to_dataclass = {
 251         'boolean': 'boolean',
 252         'File': 'File',
 253         'Directory': 'Collection',
 254         'float': 'number',
 255         'int': 'number',
 256         'string': 'text',
 257     }
 258
 259     def __init__(self, runner, tool, job_order, enable_reuse):
 260         self.runner = runner
 261         self.tool = tool
 262         self.job = RunnerJob(
 263             runner=runner,
 264             tool=tool,
 265             job_order=job_order,
 266             enable_reuse=enable_reuse)
 267
 268     def pipeline_component_spec(self):
 269         """Return a component that Workbench and a-r-p-i will understand.
 270
 271         Specifically, translate CWL input specs to Arvados pipeline
 272         format, like {"dataclass":"File","value":"xyz"}.
 273         """
 274         spec = self.job.arvados_job_spec()
 275
 276         # Most of the component spec is exactly the same as the job
 277         # spec (script, script_version, etc.).
 278         # spec['script_parameters'] isn't right, though. A component
 279         # spec's script_parameters hash is a translation of
 280         # self.tool.tool['inputs'] with defaults/overrides taken from
 281         # the job order. So we move the job parameters out of the way
 282         # and build a new spec['script_parameters'].
 283         job_params = spec['script_parameters']
 284         spec['script_parameters'] = {}
 285
 286         for param in self.tool.tool['inputs']:
 287             param = copy.deepcopy(param)
 288
 289             # Data type and "required" flag...
 290             types = param['type']
 291             if not isinstance(types, list):
 292                 types = [types]
 293             param['required'] = 'null' not in types
 294             non_null_types = set(types) - set(['null'])
 295             if len(non_null_types) == 1:
 296                 the_type = [c for c in non_null_types][0]
 297                 dataclass = self.type_to_dataclass.get(the_type)
 298                 if dataclass:
 299                     param['dataclass'] = dataclass
 300             # Note: If we didn't figure out a single appropriate
 301             # dataclass, we just left that attribute out.  We leave
 302             # the "type" attribute there in any case, which might help
 303             # downstream.
 304
 305             # Title and description...
 306             title = param.pop('label', '')
 307             descr = param.pop('doc', '').rstrip('\n')
 308             if title:
 309                 param['title'] = title
 310             if descr:
 311                 param['description'] = descr
 312
 313             # Fill in the value from the current job order, if any.
 314             param_id = shortname(param.pop('id'))
 315             value = job_params.get(param_id)
 316             if value is None:
 317                 pass
 318             elif not isinstance(value, dict):
 319                 param['value'] = value
 320             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 321                 param['value'] = value['location'][5:]
 322
 323             spec['script_parameters'][param_id] = param
 324         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 325         return spec
 326
 327     def save(self):
 328         job_spec = self.pipeline_component_spec()
 329         response = self.runner.api.pipeline_templates().create(body={
 330             "components": {
 331                 self.job.name: job_spec,
 332             },
 333             "name": self.job.name,
 334             "owner_uuid": self.runner.project_uuid,
 335         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 336         self.uuid = response["uuid"]
 337         logger.info("Created template %s", self.uuid)