7 from cwltool.process import get_feature, shortname
8 from cwltool.errors import WorkflowException
9 from cwltool.draft2tool import revmap_file, CommandLineTool
10 from cwltool.load_tool import fetch_document
11 from cwltool.builder import Builder
12 from cwltool.pathmapper import adjustDirObjs
14 import ruamel.yaml as yaml
16 import arvados.collection
18 from .arvdocker import arv_docker_get_image
19 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_listing
20 from .pathmapper import InitialWorkDirPathMapper
21 from .perf import Perf
23 from ._version import __version__
25 logger = logging.getLogger('arvados.cwl-runner')
26 metrics = logging.getLogger('arvados.cwl-runner.metrics')
28 crunchrunner_re = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.(tmpdir|outdir|keep)\)=(.*)")
30 crunchrunner_git_commit = 'a3f2cb186e437bfce0031b024b2157b73ed2717d'
32 class ArvadosJob(object):
33 """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
35 def __init__(self, runner):
36 self.arvrunner = runner
40 def run(self, dry_run=False, pull_image=True, **kwargs):
42 "command": self.command_line
44 runtime_constraints = {}
46 with Perf(metrics, "generatefiles %s" % self.name):
47 if self.generatefiles["listing"]:
48 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
49 keep_client=self.arvrunner.keep_client,
50 num_retries=self.arvrunner.num_retries)
51 script_parameters["task.vwd"] = {}
52 generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "",
55 with Perf(metrics, "createfiles %s" % self.name):
56 for f, p in generatemapper.items():
57 if p.type == "CreateFile":
58 with vwd.open(p.target, "w") as n:
59 n.write(p.resolved.encode("utf-8"))
61 with Perf(metrics, "generatefiles.save_new %s" % self.name):
64 for f, p in generatemapper.items():
66 script_parameters["task.vwd"][p.target] = p.resolved
67 if p.type == "CreateFile":
68 script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target)
70 script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
72 script_parameters["task.env"].update(self.environment)
75 script_parameters["task.stdin"] = self.stdin
78 script_parameters["task.stdout"] = self.stdout
81 script_parameters["task.stderr"] = self.stderr
84 script_parameters["task.successCodes"] = self.successCodes
85 if self.temporaryFailCodes:
86 script_parameters["task.temporaryFailCodes"] = self.temporaryFailCodes
87 if self.permanentFailCodes:
88 script_parameters["task.permanentFailCodes"] = self.permanentFailCodes
90 with Perf(metrics, "arv_docker_get_image %s" % self.name):
91 (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
92 if docker_req and kwargs.get("use_container") is not False:
93 if docker_req.get("dockerOutputDirectory"):
94 raise SourceLine(docker_req, "dockerOutputDirectory", UnsupportedRequirement).makeError(
95 "Option 'dockerOutputDirectory' of DockerRequirement not supported.")
96 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
98 runtime_constraints["docker_image"] = "arvados/jobs"
100 resources = self.builder.resources
101 if resources is not None:
102 runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
103 runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
104 runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
106 runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
108 if "keep_cache" in runtime_req:
109 runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"]
110 if "outputDirType" in runtime_req:
111 if runtime_req["outputDirType"] == "local_output_dir":
112 script_parameters["task.keepTmpOutput"] = False
113 elif runtime_req["outputDirType"] == "keep_output_dir":
114 script_parameters["task.keepTmpOutput"] = True
116 filters = [["repository", "=", "arvados"],
117 ["script", "=", "crunchrunner"],
118 ["script_version", "in git", crunchrunner_git_commit]]
119 if not self.arvrunner.ignore_docker_for_reuse:
120 filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
123 with Perf(metrics, "create %s" % self.name):
124 response = self.arvrunner.api.jobs().create(
126 "owner_uuid": self.arvrunner.project_uuid,
127 "script": "crunchrunner",
128 "repository": "arvados",
129 "script_version": "master",
130 "minimum_script_version": crunchrunner_git_commit,
131 "script_parameters": {"tasks": [script_parameters]},
132 "runtime_constraints": runtime_constraints
135 find_or_create=kwargs.get("enable_reuse", True)
136 ).execute(num_retries=self.arvrunner.num_retries)
138 self.arvrunner.processes[response["uuid"]] = self
140 self.update_pipeline_component(response)
142 logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"])
144 if response["state"] in ("Complete", "Failed", "Cancelled"):
145 with Perf(metrics, "done %s" % self.name):
147 except Exception as e:
148 logger.exception("%s error" % (self.arvrunner.label(self)))
149 self.output_callback({}, "permanentFail")
151 def update_pipeline_component(self, record):
152 if self.arvrunner.pipeline:
153 self.arvrunner.pipeline["components"][self.name] = {"job": record}
154 with Perf(metrics, "update_pipeline_component %s" % self.name):
155 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
157 "components": self.arvrunner.pipeline["components"]
158 }).execute(num_retries=self.arvrunner.num_retries)
159 if self.arvrunner.uuid:
161 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
163 components = job["components"]
164 components[self.name] = record["uuid"]
165 self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
167 "components": components
168 }).execute(num_retries=self.arvrunner.num_retries)
169 except Exception as e:
170 logger.info("Error adding to components: %s", e)
172 def done(self, record):
174 self.update_pipeline_component(record)
179 if record["state"] == "Complete":
180 processStatus = "success"
182 processStatus = "permanentFail"
187 with Perf(metrics, "inspect log %s" % self.name):
188 logc = arvados.collection.CollectionReader(record["log"],
189 api_client=self.arvrunner.api,
190 keep_client=self.arvrunner.keep_client,
191 num_retries=self.arvrunner.num_retries)
192 log = logc.open(logc.keys()[0])
198 # Determine the tmpdir, outdir and keepdir paths from
199 # the job run. Unfortunately, we can't take the first
200 # values we find (which are expected to be near the
201 # top) and stop scanning because if the node fails and
202 # the job restarts on a different node these values
203 # will different runs, and we need to know about the
204 # final run that actually produced output.
205 g = crunchrunner_re.match(l)
207 dirs[g.group(1)] = g.group(2)
209 if processStatus == "permanentFail":
210 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
212 with Perf(metrics, "output collection %s" % self.name):
213 outputs = done.done(self, record, dirs["tmpdir"],
214 dirs["outdir"], dirs["keep"])
215 except WorkflowException as e:
216 logger.error("%s unable to collect output from %s:\n%s",
217 self.arvrunner.label(self), record["output"], e, exc_info=(e if self.arvrunner.debug else False))
218 processStatus = "permanentFail"
219 except Exception as e:
220 logger.exception("Got unknown exception while collecting output for job %s:", self.name)
221 processStatus = "permanentFail"
223 # Note: Currently, on error output_callback is expecting an empty dict,
224 # anything else will fail.
225 if not isinstance(outputs, dict):
226 logger.error("Unexpected output type %s '%s'", type(outputs), outputs)
228 processStatus = "permanentFail"
230 self.output_callback(outputs, processStatus)
231 if record["uuid"] in self.arvrunner.processes:
232 del self.arvrunner.processes[record["uuid"]]
234 class RunnerJob(Runner):
235 """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner."""
237 def upload_workflow_collection(self, packed):
238 collection = arvados.collection.Collection(api_client=self.arvrunner.api,
239 keep_client=self.arvrunner.keep_client,
240 num_retries=self.arvrunner.num_retries)
241 with collection.open("workflow.cwl", "w") as f:
242 f.write(yaml.round_trip_dump(packed))
244 exists = self.arvrunner.api.collections().list(filters=[["owner_uuid", "=", self.arvrunner.project_uuid],
245 ["portable_data_hash", "=", collection.portable_data_hash()],
246 ["name", "like", self.name+"%"]]).execute(num_retries=self.arvrunner.num_retries)
249 logger.info("Using collection %s", exists["items"][0]["uuid"])
251 collection.save_new(name=self.name,
252 owner_uuid=self.arvrunner.project_uuid,
253 ensure_unique_name=True,
254 num_retries=self.arvrunner.num_retries)
255 logger.info("Uploaded to %s", collection.manifest_locator())
257 return collection.portable_data_hash()
259 def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
260 """Create an Arvados job specification for this workflow.
262 The returned dict can be used to create a job (i.e., passed as
263 the +body+ argument to jobs().create()), or as a component in
264 a pipeline template or pipeline instance.
267 if self.tool.tool["id"].startswith("keep:"):
268 self.job_order["cwl:tool"] = self.tool.tool["id"][5:]
270 packed = packed_workflow(self.arvrunner, self.tool)
271 wf_pdh = self.upload_workflow_collection(packed)
272 self.job_order["cwl:tool"] = "%s/workflow.cwl" % wf_pdh
274 adjustDirObjs(self.job_order, trim_listing)
277 self.job_order["arv:output_name"] = self.output_name
280 self.job_order["arv:output_tags"] = self.output_tags
282 self.job_order["arv:enable_reuse"] = self.enable_reuse
285 self.job_order["arv:on_error"] = self.on_error
288 "script": "cwl-runner",
289 "script_version": "master",
290 "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9",
291 "repository": "arvados",
292 "script_parameters": self.job_order,
293 "runtime_constraints": {
294 "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
295 "min_ram_mb_per_node": self.submit_runner_ram
299 def run(self, *args, **kwargs):
300 job_spec = self.arvados_job_spec(*args, **kwargs)
302 job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
304 job = self.arvrunner.api.jobs().create(
306 find_or_create=self.enable_reuse
307 ).execute(num_retries=self.arvrunner.num_retries)
309 for k,v in job_spec["script_parameters"].items():
310 if v is False or v is None or isinstance(v, dict):
311 job_spec["script_parameters"][k] = {"value": v}
313 del job_spec["owner_uuid"]
314 job_spec["job"] = job
315 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create(
317 "owner_uuid": self.arvrunner.project_uuid,
319 "components": {"cwl-runner": job_spec },
320 "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries)
321 logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"])
323 if kwargs.get("wait") is False:
324 self.uuid = self.arvrunner.pipeline["uuid"]
327 self.uuid = job["uuid"]
328 self.arvrunner.processes[self.uuid] = self
330 if job["state"] in ("Complete", "Failed", "Cancelled"):
334 class RunnerTemplate(object):
335 """An Arvados pipeline template that invokes a CWL workflow."""
337 type_to_dataclass = {
338 'boolean': 'boolean',
340 'Directory': 'Collection',
346 def __init__(self, runner, tool, job_order, enable_reuse, uuid,
347 submit_runner_ram=0, name=None):
350 self.job = RunnerJob(
354 enable_reuse=enable_reuse,
357 submit_runner_ram=submit_runner_ram,
361 def pipeline_component_spec(self):
362 """Return a component that Workbench and a-r-p-i will understand.
364 Specifically, translate CWL input specs to Arvados pipeline
365 format, like {"dataclass":"File","value":"xyz"}.
368 spec = self.job.arvados_job_spec()
370 # Most of the component spec is exactly the same as the job
371 # spec (script, script_version, etc.).
372 # spec['script_parameters'] isn't right, though. A component
373 # spec's script_parameters hash is a translation of
374 # self.tool.tool['inputs'] with defaults/overrides taken from
375 # the job order. So we move the job parameters out of the way
376 # and build a new spec['script_parameters'].
377 job_params = spec['script_parameters']
378 spec['script_parameters'] = {}
380 for param in self.tool.tool['inputs']:
381 param = copy.deepcopy(param)
383 # Data type and "required" flag...
384 types = param['type']
385 if not isinstance(types, list):
387 param['required'] = 'null' not in types
388 non_null_types = set(types) - set(['null'])
389 if len(non_null_types) == 1:
390 the_type = [c for c in non_null_types][0]
391 dataclass = self.type_to_dataclass.get(the_type)
393 param['dataclass'] = dataclass
394 # Note: If we didn't figure out a single appropriate
395 # dataclass, we just left that attribute out. We leave
396 # the "type" attribute there in any case, which might help
399 # Title and description...
400 title = param.pop('label', '')
401 descr = param.pop('doc', '').rstrip('\n')
403 param['title'] = title
405 param['description'] = descr
407 # Fill in the value from the current job order, if any.
408 param_id = shortname(param.pop('id'))
409 value = job_params.get(param_id)
412 elif not isinstance(value, dict):
413 param['value'] = value
414 elif param.get('dataclass') in ('File', 'Collection') and value.get('location'):
415 param['value'] = value['location'][5:]
417 spec['script_parameters'][param_id] = param
418 spec['script_parameters']['cwl:tool'] = job_params['cwl:tool']
424 self.job.name: self.pipeline_component_spec(),
426 "name": self.job.name,
428 if self.runner.project_uuid:
429 body["owner_uuid"] = self.runner.project_uuid
431 self.runner.api.pipeline_templates().update(
432 uuid=self.uuid, body=body).execute(
433 num_retries=self.runner.num_retries)
434 logger.info("Updated template %s", self.uuid)
436 self.uuid = self.runner.api.pipeline_templates().create(
437 body=body, ensure_unique_name=True).execute(
438 num_retries=self.runner.num_retries)['uuid']
439 logger.info("Created template %s", self.uuid)