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(api_client=self.arvrunner.api,
  44                                                     keep_client=self.arvrunner.keep_client,
  45                                                     num_retries=self.arvrunner.num_retries)
  46                 script_parameters["task.vwd"] = {}
  47                 generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
  48                                                           separateDirs=False)
  49
  50                 with Perf(metrics, "createfiles %s" % self.name):
  51                     for f, p in generatemapper.items():
  52                         if p.type == "CreateFile":
  53                             with vwd.open(p.target, "w") as n:
  54                                 n.write(p.resolved.encode("utf-8"))
  55
  56                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
  57                     vwd.save_new()
  58
  59                 for f, p in generatemapper.items():
  60                     if p.type == "File":
  61                         script_parameters["task.vwd"][p.target] = p.resolved
  62                     if p.type == "CreateFile":
  63                         script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
  64
  65         script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  66         if self.environment:
  67             script_parameters["task.env"].update(self.environment)
  68
  69         if self.stdin:
  70             script_parameters["task.stdin"] = self.stdin
  71
  72         if self.stdout:
  73             script_parameters["task.stdout"] = self.stdout
  74
  75         if self.stderr:
  76             script_parameters["task.stderr"] = self.stderr
  77
  78         if self.successCodes:
  79             script_parameters["task.successCodes"] = self.successCodes
  80         if self.temporaryFailCodes:
  81             script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
  82         if self.permanentFailCodes:
  83             script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
  84
  85         with Perf(metrics, "arv_docker_get_image %s" % self.name):
  86             (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  87             if docker_req and kwargs.get("use_container") is not False:
  88                 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  89             else:
  90                 runtime_constraints["docker_image"] = "arvados/jobs"
  91
  92         resources = self.builder.resources
  93         if resources is not None:
  94             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  95             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  96             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  97
  98         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
  99         if runtime_req:
 100             if "keep_cache" in runtime_req:
 101                 runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
 102             if "outputDirType" in runtime_req:
 103                 if runtime_req["outputDirType"] == "local_output_dir":
 104                     script_parameters["task.keepTmpOutput"] = False
 105                 elif runtime_req["outputDirType"] == "keep_output_dir":
 106                     script_parameters["task.keepTmpOutput"] = True
 107
 108         filters = [["repository", "=", "arvados"],
 109                    ["script", "=", "crunchrunner"],
 110                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
 111         if not self.arvrunner.ignore_docker_for_reuse:
 112             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
 113
 114         try:
 115             with Perf(metrics, "create %s" % self.name):
 116                 response = self.arvrunner.api.jobs().create(
 117                     body={
 118                         "owner_uuid": self.arvrunner.project_uuid,
 119                         "script": "crunchrunner",
 120                         "repository": "arvados",
 121                         "script_version": "master",
 122                         "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
 123                         "script_parameters": {"tasks": [script_parameters]},
 124                         "runtime_constraints": runtime_constraints
 125                     },
 126                     filters=filters,
 127                     find_or_create=kwargs.get("enable_reuse", True)
 128                 ).execute(num_retries=self.arvrunner.num_retries)
 129
 130             self.arvrunner.processes[response["uuid"]] = self
 131
 132             self.update_pipeline_component(response)
 133
 134             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
 135
 136             if response["state"] in ("Complete", "Failed", "Cancelled"):
 137                 with Perf(metrics, "done %s" % self.name):
 138                     self.done(response)
 139         except Exception as e:
 140             logger.error("Got error %s" % str(e))
 141             self.output_callback({}, "permanentFail")
 142
 143     def update_pipeline_component(self, record):
 144         if self.arvrunner.pipeline:
 145             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 146             with Perf(metrics, "update_pipeline_component %s" % self.name):
 147                 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 148                                                                                  body={
 149                                                                                     "components": self.arvrunner.pipeline["components"]
 150                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 151         if self.arvrunner.uuid:
 152             try:
 153                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 154                 if job:
 155                     components = job["components"]
 156                     components[self.name] = record["uuid"]
 157                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 158                         body={
 159                             "components": components
 160                         }).execute(num_retries=self.arvrunner.num_retries)
 161             except Exception as e:
 162                 logger.info("Error adding to components: %s", e)
 163
 164     def done(self, record):
 165         try:
 166             self.update_pipeline_component(record)
 167         except:
 168             pass
 169
 170         try:
 171             if record["state"] == "Complete":
 172                 processStatus = "success"
 173             else:
 174                 processStatus = "permanentFail"
 175
 176             outputs = {}
 177             try:
 178                 if record["output"]:
 179                     with Perf(metrics, "inspect log %s" % self.name):
 180                         logc = arvados.collection.CollectionReader(record["log"],
 181                                                                    api_client=self.arvrunner.api,
 182                                                                    keep_client=self.arvrunner.keep_client,
 183                                                                    num_retries=self.arvrunner.num_retries)
 184                         log = logc.open(logc.keys()[0])
 185                         tmpdir = None
 186                         outdir = None
 187                         keepdir = None
 188                         for l in log:
 189                             # Determine the tmpdir, outdir and keepdir paths from
 190                             # the job run.  Unfortunately, we can't take the first
 191                             # values we find (which are expected to be near the
 192                             # top) and stop scanning because if the node fails and
 193                             # the job restarts on a different node these values
 194                             # will different runs, and we need to know about the
 195                             # final run that actually produced output.
 196
 197                             g = tmpdirre.match(l)
 198                             if g:
 199                                 tmpdir = g.group(1)
 200                             g = outdirre.match(l)
 201                             if g:
 202                                 outdir = g.group(1)
 203                             g = keepre.match(l)
 204                             if g:
 205                                 keepdir = g.group(1)
 206
 207                     with Perf(metrics, "output collection %s" % self.name):
 208                         outputs = done.done(self, record, tmpdir, outdir, keepdir)
 209             except WorkflowException as e:
 210                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 211                 processStatus = "permanentFail"
 212                 outputs = None
 213             except Exception as e:
 214                 logger.exception("Got unknown exception while collecting job outputs:")
 215                 processStatus = "permanentFail"
 216                 outputs = None
 217
 218             self.output_callback(outputs, processStatus)
 219         finally:
 220             del self.arvrunner.processes[record["uuid"]]
 221
 222
 223 class RunnerJob(Runner):
 224     """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
 225
 226     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 227         """Create an Arvados job specification for this workflow.
 228
 229         The returned dict can be used to create a job (i.e., passed as
 230         the +body+ argument to jobs().create()), or as a component in
 231         a pipeline template or pipeline instance.
 232         """
 233
 234         workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 235
 236         self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"]).target[5:]
 237         if self.output_name:
 238             self.job_order["arv:output_name"] = self.output_name
 239         return {
 240             "script": "cwl-runner",
 241             "script_version": "master",
 242             "repository": "arvados",
 243             "script_parameters": self.job_order,
 244             "runtime_constraints": {
 245                 "docker_image": "arvados/jobs"
 246             }
 247         }
 248
 249     def run(self, *args, **kwargs):
 250         job_spec = self.arvados_job_spec(*args, **kwargs)
 251         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 252
 253         response = self.arvrunner.api.jobs().create(
 254             body=job_spec,
 255             find_or_create=self.enable_reuse
 256         ).execute(num_retries=self.arvrunner.num_retries)
 257
 258         self.uuid = response["uuid"]
 259         self.arvrunner.processes[self.uuid] = self
 260
 261         logger.info("Submitted job %s", response["uuid"])
 262
 263         if kwargs.get("submit"):
 264             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
 265                 body={
 266                     "owner_uuid": self.arvrunner.project_uuid,
 267                     "name": shortname(self.tool.tool["id"]),
 268                     "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } },
 269                     "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries)
 270
 271         if response["state"] in ("Complete", "Failed", "Cancelled"):
 272             self.done(response)
 273
 274
 275 class RunnerTemplate(object):
 276     """An Arvados pipeline template that invokes a CWL workflow."""
 277
 278     type_to_dataclass = {
 279         'boolean': 'boolean',
 280         'File': 'File',
 281         'Directory': 'Collection',
 282         'float': 'number',
 283         'int': 'number',
 284         'string': 'text',
 285     }
 286
 287     def __init__(self, runner, tool, job_order, enable_reuse):
 288         self.runner = runner
 289         self.tool = tool
 290         self.job = RunnerJob(
 291             runner=runner,
 292             tool=tool,
 293             job_order=job_order,
 294             enable_reuse=enable_reuse,
 295             output_name=None)
 296
 297     def pipeline_component_spec(self):
 298         """Return a component that Workbench and a-r-p-i will understand.
 299
 300         Specifically, translate CWL input specs to Arvados pipeline
 301         format, like {"dataclass":"File","value":"xyz"}.
 302         """
 303         spec = self.job.arvados_job_spec()
 304
 305         # Most of the component spec is exactly the same as the job
 306         # spec (script, script_version, etc.).
 307         # spec['script_parameters'] isn't right, though. A component
 308         # spec's script_parameters hash is a translation of
 309         # self.tool.tool['inputs'] with defaults/overrides taken from
 310         # the job order. So we move the job parameters out of the way
 311         # and build a new spec['script_parameters'].
 312         job_params = spec['script_parameters']
 313         spec['script_parameters'] = {}
 314
 315         for param in self.tool.tool['inputs']:
 316             param = copy.deepcopy(param)
 317
 318             # Data type and "required" flag...
 319             types = param['type']
 320             if not isinstance(types, list):
 321                 types = [types]
 322             param['required'] = 'null' not in types
 323             non_null_types = set(types) - set(['null'])
 324             if len(non_null_types) == 1:
 325                 the_type = [c for c in non_null_types][0]
 326                 dataclass = self.type_to_dataclass.get(the_type)
 327                 if dataclass:
 328                     param['dataclass'] = dataclass
 329             # Note: If we didn't figure out a single appropriate
 330             # dataclass, we just left that attribute out.  We leave
 331             # the "type" attribute there in any case, which might help
 332             # downstream.
 333
 334             # Title and description...
 335             title = param.pop('label', '')
 336             descr = param.pop('doc', '').rstrip('\n')
 337             if title:
 338                 param['title'] = title
 339             if descr:
 340                 param['description'] = descr
 341
 342             # Fill in the value from the current job order, if any.
 343             param_id = shortname(param.pop('id'))
 344             value = job_params.get(param_id)
 345             if value is None:
 346                 pass
 347             elif not isinstance(value, dict):
 348                 param['value'] = value
 349             elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
 350                 param['value'] = value['location'][5:]
 351
 352             spec['script_parameters'][param_id] = param
 353         spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
 354         return spec
 355
 356     def save(self):
 357         job_spec = self.pipeline_component_spec()
 358         response = self.runner.api.pipeline_templates().create(body={
 359             "components": {
 360                 self.job.name: job_spec,
 361             },
 362             "name": self.job.name,
 363             "owner_uuid": self.runner.project_uuid,
 364         }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries)
 365         self.uuid = response["uuid"]
 366         logger.info("Created template %s", self.uuid)