import arvados.events
import arvados.commands.keepdocker
import arvados.commands.run
+import arvados.collection
+import arvados.util
import cwltool.draft2tool
import cwltool.workflow
import cwltool.main
from cwltool.process import shortname
+from cwltool.errors import WorkflowException
import threading
import cwltool.docker
import fnmatch
import logging
import re
import os
+import sys
from cwltool.process import get_feature
+from arvados.api import OrderedJsonModel
logger = logging.getLogger('arvados.cwl-runner')
logger.setLevel(logging.INFO)
-def arv_docker_get_image(api_client, dockerRequirement, pull_image):
+crunchrunner_pdh = "83db29f08544e1c319572a6bd971088a+140"
+crunchrunner_download = "https://cloud.curoverse.com/collections/download/qr1hi-4zz18-n3m1yxd0vx78jic/1i1u2qtq66k1atziv4ocfgsg5nu5tj11n4r6e0bhvjg03rix4m/crunchrunner"
+certs_download = "https://cloud.curoverse.com/collections/download/qr1hi-4zz18-n3m1yxd0vx78jic/1i1u2qtq66k1atziv4ocfgsg5nu5tj11n4r6e0bhvjg03rix4m/ca-certificates.crt"
+
+tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
+outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
+keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
+
+
+def arv_docker_get_image(api_client, dockerRequirement, pull_image, project_uuid):
if "dockerImageId" not in dockerRequirement and "dockerPull" in dockerRequirement:
dockerRequirement["dockerImageId"] = dockerRequirement["dockerPull"]
if not images:
imageId = cwltool.docker.get_image(dockerRequirement, pull_image)
- args = [image_name]
+ args = ["--project-uuid="+project_uuid, image_name]
if image_tag:
args.append(image_tag)
- logger.info("Uploading Docker image %s", ":".join(args))
+ logger.info("Uploading Docker image %s", ":".join(args[1:]))
arvados.commands.keepdocker.main(args)
return dockerRequirement["dockerImageId"]
(docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
if docker_req and kwargs.get("use_container") is not False:
- runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image)
+ runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
+
+ resources = self.builder.resources
+ if resources is not None:
+ runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
+ runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
+ runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
try:
response = self.arvrunner.api.jobs().create(body={
+ "owner_uuid": self.arvrunner.project_uuid,
"script": "crunchrunner",
- "repository": kwargs["repository"],
+ "repository": "arvados",
"script_version": "master",
- "script_parameters": {"tasks": [script_parameters]},
+ "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
+ "script_parameters": {"tasks": [script_parameters], "crunchrunner": crunchrunner_pdh+"/crunchrunner"},
"runtime_constraints": runtime_constraints
}, find_or_create=kwargs.get("enable_reuse", True)).execute(num_retries=self.arvrunner.num_retries)
try:
outputs = {}
if record["output"]:
+ logc = arvados.collection.Collection(record["log"])
+ log = logc.open(logc.keys()[0])
+ tmpdir = None
+ outdir = None
+ keepdir = None
+ for l in log.readlines():
+ g = tmpdirre.match(l)
+ if g:
+ tmpdir = g.group(1)
+ g = outdirre.match(l)
+ if g:
+ outdir = g.group(1)
+ g = keepre.match(l)
+ if g:
+ keepdir = g.group(1)
+
+ # It turns out if the job fails and restarts it can
+ # come up on a different compute node, so we have to
+ # read the log to the end to be sure instead of taking the
+ # easy way out.
+ #
+ #if tmpdir and outdir and keepdir:
+ # break
+
+ self.builder.outdir = outdir
+ self.builder.pathmapper.keepdir = keepdir
outputs = self.collect_outputs("keep:" + record["output"])
+ except WorkflowException as e:
+ logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
+ processStatus = "permanentFail"
except Exception as e:
- logger.exception("Got exception while collecting job outputs:")
+ logger.exception("Got unknown exception while collecting job outputs:")
processStatus = "permanentFail"
self.output_callback(outputs, processStatus)
class ArvPathMapper(cwltool.pathmapper.PathMapper):
- def __init__(self, arvrunner, referenced_files, basedir, **kwargs):
+ def __init__(self, arvrunner, referenced_files, basedir,
+ collection_pattern, file_pattern, **kwargs):
self._pathmap = arvrunner.get_uploaded()
uploadfiles = []
for src in referenced_files:
if isinstance(src, basestring) and pdh_path.match(src):
- self._pathmap[src] = (src, "$(task.keep)/%s" % src[5:])
+ self._pathmap[src] = (src, collection_pattern % src[5:])
if src not in self._pathmap:
ab = cwltool.pathmapper.abspath(src, basedir)
- st = arvados.commands.run.statfile("", ab, fnPattern="$(task.keep)/%s/%s")
+ st = arvados.commands.run.statfile("", ab, fnPattern=file_pattern)
if kwargs.get("conformance_test"):
self._pathmap[src] = (src, ab)
elif isinstance(st, arvados.commands.run.UploadFile):
arvrunner.api,
dry_run=kwargs.get("dry_run"),
num_retries=3,
- fnPattern="$(task.keep)/%s/%s")
+ fnPattern=file_pattern,
+ project=arvrunner.project_uuid)
for src, ab, st in uploadfiles:
arvrunner.add_uploaded(src, (ab, st.fn))
self._pathmap[src] = (ab, st.fn)
+ self.keepdir = None
+
+ def reversemap(self, target):
+ if target.startswith("keep:"):
+ return (target, target)
+ elif self.keepdir and target.startswith(self.keepdir):
+ return (target, "keep:" + target[len(self.keepdir)+1:])
+ else:
+ return super(ArvPathMapper, self).reversemap(target)
class ArvadosCommandTool(cwltool.draft2tool.CommandLineTool):
def __init__(self, arvrunner, toolpath_object, **kwargs):
- super(ArvadosCommandTool, self).__init__(toolpath_object, outdir="$(task.outdir)", tmpdir="$(task.tmpdir)", **kwargs)
+ super(ArvadosCommandTool, self).__init__(toolpath_object, **kwargs)
self.arvrunner = arvrunner
def makeJobRunner(self):
return ArvadosJob(self.arvrunner)
def makePathMapper(self, reffiles, input_basedir, **kwargs):
- return ArvPathMapper(self.arvrunner, reffiles, input_basedir, **kwargs)
+ return ArvPathMapper(self.arvrunner, reffiles, input_basedir,
+ "$(task.keep)/%s",
+ "$(task.keep)/%s/%s",
+ **kwargs)
class ArvCwlRunner(object):
def on_message(self, event):
if "object_uuid" in event:
- if event["object_uuid"] in self.jobs and event["event_type"] == "update":
- if event["properties"]["new_attributes"]["state"] == "Running" and self.jobs[event["object_uuid"]].running is False:
- uuid = event["object_uuid"]
- with self.lock:
- j = self.jobs[uuid]
- logger.info("Job %s (%s) is Running", j.name, uuid)
- j.running = True
- j.update_pipeline_component(event["properties"]["new_attributes"])
- elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
- uuid = event["object_uuid"]
- try:
- self.cond.acquire()
- j = self.jobs[uuid]
- logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"])
- j.done(event["properties"]["new_attributes"])
- self.cond.notify()
- finally:
- self.cond.release()
+ if event["object_uuid"] in self.jobs and event["event_type"] == "update":
+ if event["properties"]["new_attributes"]["state"] == "Running" and self.jobs[event["object_uuid"]].running is False:
+ uuid = event["object_uuid"]
+ with self.lock:
+ j = self.jobs[uuid]
+ logger.info("Job %s (%s) is Running", j.name, uuid)
+ j.running = True
+ j.update_pipeline_component(event["properties"]["new_attributes"])
+ elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
+ uuid = event["object_uuid"]
+ try:
+ self.cond.acquire()
+ j = self.jobs[uuid]
+ logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"])
+ j.done(event["properties"]["new_attributes"])
+ self.cond.notify()
+ finally:
+ self.cond.release()
def get_uploaded(self):
return self.uploaded.copy()
def add_uploaded(self, src, pair):
self.uploaded[src] = pair
+ def upload_docker(self, tool):
+ pass
+
+ def submit(self, tool, job_order, input_basedir, args, **kwargs):
+ files = set()
+ def visitFiles(self, path):
+ files.add(path)
+
+ adjustFiles(process.scandeps("", tool.tool,
+ set(("run")),
+ set(("$schemas", "path"))),
+ visitFiles)
+ adjustFiles(job_order, visitFiles)
+
+ mapper = ArvPathMapper(self, files, "",
+ "$(task.keep)/%s",
+ "$(task.keep)/%s/%s",
+ **kwargs)
+
+ job_order = adjustFiles(job_order, lambda p: mapper.mapper(p))
+
+ response = self.api.jobs().create(body={
+ "script": "cwl-runner",
+ "script_version": "8654-arv-jobs-cwl-runner",
+ "repository": "arvados",
+ "script_parameters": job_order,
+ "runtime_constraints": {
+ "docker_image": "arvados/jobs"
+ }
+ }, find_or_create=args.enable_reuse).execute(num_retries=self.num_retries)
+ print response["uuid"]
+ return None
+
+
def arvExecutor(self, tool, job_order, input_basedir, args, **kwargs):
+ if args.submit:
+ self.submit(tool, job_order, input_basedir, args, **kwargs)
+ return
+
events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message)
- self.pipeline = self.api.pipeline_instances().create(body={"name": shortname(tool.tool["id"]),
- "components": {},
- "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
+ self.debug = args.debug
+
+ try:
+ self.api.collections().get(uuid=crunchrunner_pdh).execute()
+ except arvados.errors.ApiError as e:
+ import httplib2
+ h = httplib2.Http(ca_certs=arvados.util.ca_certs_path())
+ resp, content = h.request(crunchrunner_download, "GET")
+ resp2, content2 = h.request(certs_download, "GET")
+ with arvados.collection.Collection() as col:
+ with col.open("crunchrunner", "w") as f:
+ f.write(content)
+ with col.open("ca-certificates.crt", "w") as f:
+ f.write(content2)
+
+ col.save_new("crunchrunner binary", ensure_unique_name=True)
self.fs_access = CollectionFsAccess(input_basedir)
kwargs["fs_access"] = self.fs_access
kwargs["enable_reuse"] = args.enable_reuse
- kwargs["repository"] = args.repository
+
+ kwargs["outdir"] = "$(task.outdir)"
+ kwargs["tmpdir"] = "$(task.tmpdir)"
+
+ useruuid = self.api.users().current().execute()["uuid"]
+ self.project_uuid = args.project_uuid if args.project_uuid else useruuid
if kwargs.get("conformance_test"):
return cwltool.main.single_job_executor(tool, job_order, input_basedir, args, **kwargs)
else:
+ self.pipeline = self.api.pipeline_instances().create(
+ body={
+ "owner_uuid": self.project_uuid,
+ "name": shortname(tool.tool["id"]),
+ "components": {},
+ "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
+
+ logger.info("Pipeline instance %s", self.pipeline["uuid"])
+
jobiter = tool.job(job_order,
- input_basedir,
- self.output_callback,
- **kwargs)
+ input_basedir,
+ self.output_callback,
+ docker_outdir="$(task.outdir)",
+ **kwargs)
- for runnable in jobiter:
- if runnable:
- with self.lock:
+ try:
+ self.cond.acquire()
+ # Will continue to hold the lock for the duration of this code
+ # except when in cond.wait(), at which point on_message can update
+ # job state and process output callbacks.
+
+ for runnable in jobiter:
+ if runnable:
runnable.run(**kwargs)
- else:
- if self.jobs:
- try:
- self.cond.acquire()
- self.cond.wait()
- finally:
- self.cond.release()
else:
- logger.error("Workflow cannot make any more progress.")
- break
+ if self.jobs:
+ self.cond.wait(1)
+ else:
+ logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
+ break
- while self.jobs:
- try:
- self.cond.acquire()
- self.cond.wait()
- finally:
- self.cond.release()
+ while self.jobs:
+ self.cond.wait(1)
- events.close()
+ events.close()
- if self.final_output is None:
- raise cwltool.workflow.WorkflowException("Workflow did not return a result.")
+ if self.final_output is None:
+ raise cwltool.workflow.WorkflowException("Workflow did not return a result.")
+
+ # create final output collection
+ except:
+ if sys.exc_info()[0] is KeyboardInterrupt:
+ logger.error("Interrupted, marking pipeline as failed")
+ else:
+ logger.error("Caught unhandled exception, marking pipeline as failed. Error was: %s", sys.exc_info()[0], exc_info=(sys.exc_info()[1] if self.debug else False))
+ self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
+ body={"state": "Failed"}).execute(num_retries=self.num_retries)
+ finally:
+ self.cond.release()
return self.final_output
def main(args, stdout, stderr, api_client=None):
- runner = ArvCwlRunner(api_client=arvados.api('v1'))
args.insert(0, "--leave-outputs")
parser = cwltool.main.arg_parser()
exgroup = parser.add_mutually_exclusive_group()
exgroup.add_argument("--enable-reuse", action="store_true",
- default=False, dest="enable_reuse",
+ default=True, dest="enable_reuse",
help="")
exgroup.add_argument("--disable-reuse", action="store_false",
- default=False, dest="enable_reuse",
+ default=True, dest="enable_reuse",
help="")
-
- parser.add_argument('--repository', type=str, default="peter/crunchrunner", help="Repository containing the 'crunchrunner' program.")
+ parser.add_argument("--project-uuid", type=str, help="Project that will own the workflow jobs")
+ parser.add_argument("--submit", type=str, help="Submit job and print job uuid.")
+
+ try:
+ runner = ArvCwlRunner(api_client=arvados.api('v1', model=OrderedJsonModel()))
+ except Exception as e:
+ logger.error(e)
+ return 1
return cwltool.main.main(args, executor=runner.arvExecutor, makeTool=runner.arvMakeTool, parser=parser)