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