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