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