6 import arvados.commands.keepdocker
7 import arvados.commands.run
8 import arvados.collection
10 import cwltool.draft2tool
11 import cwltool.workflow
13 from cwltool.process import shortname
22 from cwltool.process import get_feature
23 from arvados.api import OrderedJsonModel
25 logger = logging.getLogger('arvados.cwl-runner')
26 logger.setLevel(logging.INFO)
28 crunchrunner_pdh = "83db29f08544e1c319572a6bd971088a+140"
29 crunchrunner_download = "https://cloud.curoverse.com/collections/download/qr1hi-4zz18-n3m1yxd0vx78jic/1i1u2qtq66k1atziv4ocfgsg5nu5tj11n4r6e0bhvjg03rix4m/crunchrunner"
30 certs_download = "https://cloud.curoverse.com/collections/download/qr1hi-4zz18-n3m1yxd0vx78jic/1i1u2qtq66k1atziv4ocfgsg5nu5tj11n4r6e0bhvjg03rix4m/ca-certificates.crt"
32 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
33 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
34 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
37 def arv_docker_get_image(api_client, dockerRequirement, pull_image):
38 if "dockerImageId" not in dockerRequirement and "dockerPull" in dockerRequirement:
39 dockerRequirement["dockerImageId"] = dockerRequirement["dockerPull"]
41 sp = dockerRequirement["dockerImageId"].split(":")
43 image_tag = sp[1] if len(sp) > 1 else None
45 images = arvados.commands.keepdocker.list_images_in_arv(api_client, 3,
46 image_name=image_name,
50 imageId = cwltool.docker.get_image(dockerRequirement, pull_image)
53 args.append(image_tag)
54 logger.info("Uploading Docker image %s", ":".join(args))
55 arvados.commands.keepdocker.main(args)
57 return dockerRequirement["dockerImageId"]
60 class CollectionFsAccess(cwltool.process.StdFsAccess):
61 def __init__(self, basedir):
63 self.basedir = basedir
65 def get_collection(self, path):
67 if p[0].startswith("keep:") and arvados.util.keep_locator_pattern.match(p[0][5:]):
69 if pdh not in self.collections:
70 self.collections[pdh] = arvados.collection.CollectionReader(pdh)
71 return (self.collections[pdh], "/".join(p[1:]))
75 def _match(self, collection, patternsegments, parent):
76 if not patternsegments:
79 if not isinstance(collection, arvados.collection.RichCollectionBase):
83 # iterate over the files and subcollections in 'collection'
84 for filename in collection:
85 if patternsegments[0] == '.':
86 # Pattern contains something like "./foo" so just shift
88 ret.extend(self._match(collection, patternsegments[1:], parent))
89 elif fnmatch.fnmatch(filename, patternsegments[0]):
90 cur = os.path.join(parent, filename)
91 if len(patternsegments) == 1:
94 ret.extend(self._match(collection[filename], patternsegments[1:], cur))
97 def glob(self, pattern):
98 collection, rest = self.get_collection(pattern)
99 patternsegments = rest.split("/")
100 return self._match(collection, patternsegments, "keep:" + collection.manifest_locator())
102 def open(self, fn, mode):
103 collection, rest = self.get_collection(fn)
105 return collection.open(rest, mode)
107 return open(self._abs(fn), mode)
109 def exists(self, fn):
110 collection, rest = self.get_collection(fn)
112 return collection.exists(rest)
114 return os.path.exists(self._abs(fn))
116 class ArvadosJob(object):
117 def __init__(self, runner):
118 self.arvrunner = runner
121 def run(self, dry_run=False, pull_image=True, **kwargs):
122 script_parameters = {
123 "command": self.command_line
125 runtime_constraints = {}
127 if self.generatefiles:
128 vwd = arvados.collection.Collection()
129 script_parameters["task.vwd"] = {}
130 for t in self.generatefiles:
131 if isinstance(self.generatefiles[t], dict):
132 src, rest = self.arvrunner.fs_access.get_collection(self.generatefiles[t]["path"].replace("$(task.keep)/", "keep:"))
133 vwd.copy(rest, t, source_collection=src)
135 with vwd.open(t, "w") as f:
136 f.write(self.generatefiles[t])
138 for t in self.generatefiles:
139 script_parameters["task.vwd"][t] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), t)
141 script_parameters["task.env"] = {"TMPDIR": "$(task.tmpdir)"}
143 script_parameters["task.env"].update(self.environment)
146 script_parameters["task.stdin"] = self.pathmapper.mapper(self.stdin)[1]
149 script_parameters["task.stdout"] = self.stdout
151 (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
152 if docker_req and kwargs.get("use_container") is not False:
153 runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image)
155 resources = self.builder.resources
156 if resources is not None:
157 keys = resources.keys()
158 if "coresMin" in keys:
160 runtime_constraints["min_cores_per_node"] = int(resources["coresMin"])
162 runtime_constraints["min_cores_per_node"] = None
165 runtime_constraints["min_ram_mb_per_node"] = int(resources["ramMin"])
167 runtime_constraints["min_ram_mb_per_node"] = None
168 if "tmpdirMin" in keys:
170 runtime_constraints["min_scratch_mb_per_node"] = int(resources["tmpdirMin"])
172 runtime_constraints["min_scratch_mb_per_node"] = None
175 response = self.arvrunner.api.jobs().create(body={
176 "script": "crunchrunner",
177 "repository": "arvados",
178 "script_version": "8488-cwl-crunchrunner-collection",
179 "script_parameters": {"tasks": [script_parameters], "crunchrunner": crunchrunner_pdh+"/crunchrunner"},
180 "runtime_constraints": runtime_constraints
181 }, find_or_create=kwargs.get("enable_reuse", True)).execute(num_retries=self.arvrunner.num_retries)
183 self.arvrunner.jobs[response["uuid"]] = self
185 self.arvrunner.pipeline["components"][self.name] = {"job": response}
186 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
188 "components": self.arvrunner.pipeline["components"]
189 }).execute(num_retries=self.arvrunner.num_retries)
191 logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
193 if response["state"] in ("Complete", "Failed", "Cancelled"):
195 except Exception as e:
196 logger.error("Got error %s" % str(e))
197 self.output_callback({}, "permanentFail")
199 def update_pipeline_component(self, record):
200 self.arvrunner.pipeline["components"][self.name] = {"job": record}
201 self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
203 "components": self.arvrunner.pipeline["components"]
204 }).execute(num_retries=self.arvrunner.num_retries)
206 def done(self, record):
208 self.update_pipeline_component(record)
213 if record["state"] == "Complete":
214 processStatus = "success"
216 processStatus = "permanentFail"
221 logc = arvados.collection.Collection(record["log"])
222 log = logc.open(logc.keys()[0])
226 for l in log.readlines():
227 g = tmpdirre.match(l)
230 g = outdirre.match(l)
236 if tmpdir and outdir and keepdir:
239 self.builder.outdir = outdir
240 self.builder.pathmapper.keepdir = keepdir
241 outputs = self.collect_outputs("keep:" + record["output"])
242 except Exception as e:
243 logger.exception("Got exception while collecting job outputs:")
244 processStatus = "permanentFail"
246 self.output_callback(outputs, processStatus)
248 del self.arvrunner.jobs[record["uuid"]]
251 class ArvPathMapper(cwltool.pathmapper.PathMapper):
252 def __init__(self, arvrunner, referenced_files, basedir, **kwargs):
253 self._pathmap = arvrunner.get_uploaded()
256 pdh_path = re.compile(r'^keep:[0-9a-f]{32}\+\d+/.+')
258 for src in referenced_files:
259 if isinstance(src, basestring) and pdh_path.match(src):
260 self._pathmap[src] = (src, "$(task.keep)/%s" % src[5:])
261 if src not in self._pathmap:
262 ab = cwltool.pathmapper.abspath(src, basedir)
263 st = arvados.commands.run.statfile("", ab, fnPattern="$(task.keep)/%s/%s")
264 if kwargs.get("conformance_test"):
265 self._pathmap[src] = (src, ab)
266 elif isinstance(st, arvados.commands.run.UploadFile):
267 uploadfiles.append((src, ab, st))
268 elif isinstance(st, arvados.commands.run.ArvFile):
269 self._pathmap[src] = (ab, st.fn)
271 raise cwltool.workflow.WorkflowException("Input file path '%s' is invalid" % st)
274 arvados.commands.run.uploadfiles([u[2] for u in uploadfiles],
276 dry_run=kwargs.get("dry_run"),
278 fnPattern="$(task.keep)/%s/%s")
280 for src, ab, st in uploadfiles:
281 arvrunner.add_uploaded(src, (ab, st.fn))
282 self._pathmap[src] = (ab, st.fn)
286 def reversemap(self, target):
287 if target.startswith("keep:"):
289 elif self.keepdir and target.startswith(self.keepdir):
290 return "keep:" + target[len(self.keepdir)+1:]
292 return super(ArvPathMapper, self).reversemap(target)
295 class ArvadosCommandTool(cwltool.draft2tool.CommandLineTool):
296 def __init__(self, arvrunner, toolpath_object, **kwargs):
297 super(ArvadosCommandTool, self).__init__(toolpath_object, **kwargs)
298 self.arvrunner = arvrunner
300 def makeJobRunner(self):
301 return ArvadosJob(self.arvrunner)
303 def makePathMapper(self, reffiles, input_basedir, **kwargs):
304 return ArvPathMapper(self.arvrunner, reffiles, input_basedir, **kwargs)
307 class ArvCwlRunner(object):
308 def __init__(self, api_client):
309 self.api = api_client
311 self.lock = threading.Lock()
312 self.cond = threading.Condition(self.lock)
313 self.final_output = None
317 def arvMakeTool(self, toolpath_object, **kwargs):
318 if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
319 return ArvadosCommandTool(self, toolpath_object, **kwargs)
321 return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
323 def output_callback(self, out, processStatus):
324 if processStatus == "success":
325 logger.info("Overall job status is %s", processStatus)
326 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
327 body={"state": "Complete"}).execute(num_retries=self.num_retries)
330 logger.warn("Overall job status is %s", processStatus)
331 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
332 body={"state": "Failed"}).execute(num_retries=self.num_retries)
333 self.final_output = out
336 def on_message(self, event):
337 if "object_uuid" in event:
338 if event["object_uuid"] in self.jobs and event["event_type"] == "update":
339 if event["properties"]["new_attributes"]["state"] == "Running" and self.jobs[event["object_uuid"]].running is False:
340 uuid = event["object_uuid"]
343 logger.info("Job %s (%s) is Running", j.name, uuid)
345 j.update_pipeline_component(event["properties"]["new_attributes"])
346 elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
347 uuid = event["object_uuid"]
351 logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"])
352 j.done(event["properties"]["new_attributes"])
357 def get_uploaded(self):
358 return self.uploaded.copy()
360 def add_uploaded(self, src, pair):
361 self.uploaded[src] = pair
363 def arvExecutor(self, tool, job_order, input_basedir, args, **kwargs):
364 events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message)
367 self.api.collections().get(uuid=crunchrunner_pdh).execute()
368 except arvados.errors.ApiError as e:
370 h = httplib2.Http(ca_certs=arvados.util.ca_certs_path())
371 resp, content = h.request(crunchrunner_download, "GET")
372 resp2, content2 = h.request(certs_download, "GET")
373 with arvados.collection.Collection() as col:
374 with col.open("crunchrunner", "w") as f:
376 with col.open("ca-certificates.crt", "w") as f:
379 col.save_new("crunchrunner binary", ensure_unique_name=True)
381 self.fs_access = CollectionFsAccess(input_basedir)
383 kwargs["fs_access"] = self.fs_access
384 kwargs["enable_reuse"] = args.enable_reuse
386 kwargs["outdir"] = "$(task.outdir)"
387 kwargs["tmpdir"] = "$(task.tmpdir)"
389 if kwargs.get("conformance_test"):
390 return cwltool.main.single_job_executor(tool, job_order, input_basedir, args, **kwargs)
392 self.pipeline = self.api.pipeline_instances().create(body={"name": shortname(tool.tool["id"]),
394 "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
396 jobiter = tool.job(job_order,
398 self.output_callback,
399 docker_outdir="$(task.outdir)",
403 for runnable in jobiter:
406 runnable.run(**kwargs)
417 logger.error("Workflow cannot make any more progress.")
431 if self.final_output is None:
432 raise cwltool.workflow.WorkflowException("Workflow did not return a result.")
435 if sys.exc_info()[0] is not KeyboardInterrupt:
436 logger.exception("Caught unhandled exception, marking pipeline as failed")
437 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
438 body={"state": "Failed"}).execute(num_retries=self.num_retries)
440 return self.final_output
443 def main(args, stdout, stderr, api_client=None):
444 args.insert(0, "--leave-outputs")
445 parser = cwltool.main.arg_parser()
446 exgroup = parser.add_mutually_exclusive_group()
447 exgroup.add_argument("--enable-reuse", action="store_true",
448 default=False, dest="enable_reuse",
450 exgroup.add_argument("--disable-reuse", action="store_false",
451 default=False, dest="enable_reuse",
455 runner = ArvCwlRunner(api_client=arvados.api('v1', model=OrderedJsonModel()))
456 except Exception as e:
460 return cwltool.main.main(args, executor=runner.arvExecutor, makeTool=runner.arvMakeTool, parser=parser)