from functools import partial
import pkg_resources # part of setuptools
import Queue
+import time
+import signal
+import thread
from cwltool.errors import WorkflowException
import cwltool.main
import arvados.config
from arvados.keep import KeepClient
from arvados.errors import ApiError
+import arvados.commands._util as arv_cmd
from .arvcontainer import ArvadosContainer, RunnerContainer
from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate
-from. runner import Runner, upload_docker, upload_job_order, upload_workflow_deps, upload_dependencies
+from. runner import Runner, upload_docker, upload_job_order, upload_workflow_deps
from .arvtool import ArvadosCommandTool
from .arvworkflow import ArvadosWorkflow, upload_workflow
from .fsaccess import CollectionFsAccess, CollectionFetcher, collectionResolver, CollectionCache
from .perf import Perf
from .pathmapper import NoFollowPathMapper
+from .task_queue import TaskQueue
from ._version import __version__
from cwltool.pack import pack
"""
- def __init__(self, api_client, work_api=None, keep_client=None, output_name=None, output_tags=None, num_retries=4):
+ def __init__(self, api_client, work_api=None, keep_client=None,
+ output_name=None, output_tags=None, num_retries=4,
+ thread_count=4):
self.api = api_client
self.processes = {}
- self.in_flight = 0
self.workflow_eval_lock = threading.Condition(threading.RLock())
self.final_output = None
self.final_status = None
- self.uploaded = {}
self.num_retries = num_retries
self.uuid = None
self.stop_polling = threading.Event()
self.intermediate_output_ttl = 0
self.intermediate_output_collections = []
self.trash_intermediate = False
- self.runnable_queue = Queue.Queue()
+ self.thread_count = thread_count
+ self.poll_interval = 12
if keep_client is not None:
self.keep_client = keep_client
self.collection_cache = CollectionCache(self.api, self.keep_client, self.num_retries)
+ self.fetcher_constructor = partial(CollectionFetcher,
+ api_client=self.api,
+ fs_access=CollectionFsAccess("", collection_cache=self.collection_cache),
+ num_retries=self.num_retries)
+
self.work_api = None
expected_api = ["jobs", "containers"]
for api in expected_api:
def arv_make_tool(self, toolpath_object, **kwargs):
kwargs["work_api"] = self.work_api
- kwargs["fetcher_constructor"] = partial(CollectionFetcher,
- api_client=self.api,
- fs_access=CollectionFsAccess("", collection_cache=self.collection_cache),
- num_retries=self.num_retries)
+ kwargs["fetcher_constructor"] = self.fetcher_constructor
kwargs["resolver"] = partial(collectionResolver, self.api, num_retries=self.num_retries)
if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
return ArvadosCommandTool(self, toolpath_object, **kwargs)
body={"state": "Failed"}).execute(num_retries=self.num_retries)
self.final_status = processStatus
self.final_output = out
+ self.workflow_eval_lock.notifyAll()
- def runnable_queue_thread(self):
- while True:
- runnable, kwargs = self.runnable_queue.get()
- runnable.run(**kwargs)
def start_run(self, runnable, kwargs):
- with self.workflow_eval_lock:
- self.in_flight += 1
- self.runnable_queue.put((runnable, kwargs))
+ self.task_queue.add(partial(runnable.run, **kwargs))
def process_submitted(self, container):
with self.workflow_eval_lock:
self.processes[container.uuid] = container
- self.in_flight -= 1
- def process_done(self, uuid):
+ def process_done(self, uuid, record):
with self.workflow_eval_lock:
- if uuid in self.processes:
- del self.processes[uuid]
+ j = self.processes[uuid]
+ logger.info("%s %s is %s", self.label(j), uuid, record["state"])
+ self.task_queue.add(partial(j.done, record))
+ del self.processes[uuid]
def wrapped_callback(self, cb, obj, st):
with self.workflow_eval_lock:
cb(obj, st)
+ self.workflow_eval_lock.notifyAll()
def get_wrapped_callback(self, cb):
return partial(self.wrapped_callback, cb)
def on_message(self, event):
- if "object_uuid" in event:
- if event["object_uuid"] in self.processes and event["event_type"] == "update":
- if event["properties"]["new_attributes"]["state"] == "Running" and self.processes[event["object_uuid"]].running is False:
- uuid = event["object_uuid"]
- with self.workflow_eval_lock:
- j = self.processes[uuid]
- logger.info("%s %s is Running", self.label(j), uuid)
+ if event.get("object_uuid") in self.processes and event["event_type"] == "update":
+ uuid = event["object_uuid"]
+ if event["properties"]["new_attributes"]["state"] == "Running":
+ with self.workflow_eval_lock:
+ j = self.processes[uuid]
+ if j.running is False:
j.running = True
j.update_pipeline_component(event["properties"]["new_attributes"])
- elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled", "Final"):
- uuid = event["object_uuid"]
- with self.workflow_eval_lock:
- j = self.processes[uuid]
- logger.info("%s %s is %s", self.label(j), uuid, event["properties"]["new_attributes"]["state"])
- with Perf(metrics, "done %s" % j.name):
- j.done(event["properties"]["new_attributes"])
- self.workflow_eval_lock.notify()
+ logger.info("%s %s is Running", self.label(j), uuid)
+ elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled", "Final"):
+ self.process_done(uuid, event["properties"]["new_attributes"])
def label(self, obj):
return "[%s %s]" % (self.work_api[0:-1], obj.name)
"""
try:
+ remain_wait = self.poll_interval
while True:
- self.stop_polling.wait(15)
+ if remain_wait > 0:
+ self.stop_polling.wait(remain_wait)
if self.stop_polling.is_set():
break
with self.workflow_eval_lock:
keys = list(self.processes.keys())
if not keys:
+ remain_wait = self.poll_interval
continue
+ begin_poll = time.time()
if self.work_api == "containers":
table = self.poll_api.container_requests()
elif self.work_api == "jobs":
proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries)
except Exception as e:
logger.warn("Error checking states on API server: %s", e)
+ remain_wait = self.poll_interval
continue
for p in proc_states["items"]:
"new_attributes": p
}
})
+ finish_poll = time.time()
+ remain_wait = self.poll_interval - (finish_poll - begin_poll)
except:
- logger.error("Fatal error in state polling thread.", exc_info=(sys.exc_info()[1] if self.debug else False))
- with workflow_eval_lock:
+ logger.exception("Fatal error in state polling thread.")
+ with self.workflow_eval_lock:
self.processes.clear()
- self.workflow_eval_lock.notify()
+ self.workflow_eval_lock.notifyAll()
finally:
self.stop_polling.set()
- def get_uploaded(self):
- return self.uploaded.copy()
-
- def add_uploaded(self, src, pair):
- self.uploaded[src] = pair
-
def add_intermediate_output(self, uuid):
if uuid:
self.intermediate_output_collections.append(uuid)
self.api.collections().delete(uuid=i).execute(num_retries=self.num_retries)
except:
logger.warn("Failed to delete intermediate output: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
- if sys.exc_info()[0] is KeyboardInterrupt:
+ if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit:
break
def check_features(self, obj):
if self.intermediate_output_ttl < 0:
raise Exception("Invalid value %d for --intermediate-output-ttl, cannot be less than zero" % self.intermediate_output_ttl)
+ if kwargs.get("submit_request_uuid") and self.work_api != "containers":
+ raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api))
+
if not kwargs.get("name"):
kwargs["name"] = self.name = tool.tool.get("label") or tool.metadata.get("label") or os.path.basename(tool.tool["id"])
logger.info("Pipeline instance %s", self.pipeline["uuid"])
if runnerjob and not kwargs.get("wait"):
- runnerjob.run(wait=kwargs.get("wait"))
+ submitargs = kwargs.copy()
+ submitargs['submit'] = False
+ runnerjob.run(**submitargs)
return (runnerjob.uuid, "success")
self.poll_api = arvados.api('v1')
self.polling_thread = threading.Thread(target=self.poll_states)
self.polling_thread.start()
- threading.Thread(target=self.runnable_queue_thread).start()
+ self.task_queue = TaskQueue(self.workflow_eval_lock, self.thread_count)
if runnerjob:
jobiter = iter((runnerjob,))
if self.stop_polling.is_set():
break
+ if self.task_queue.error is not None:
+ raise self.task_queue.error
+
if runnable:
with Perf(metrics, "run"):
self.start_run(runnable, kwargs)
else:
- if (self.in_flight + len(self.processes)) > 0:
- self.workflow_eval_lock.wait(1)
+ if (self.task_queue.in_flight + len(self.processes)) > 0:
+ self.workflow_eval_lock.wait(3)
else:
- logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
+ logger.error("Workflow is deadlocked, no runnable processes and not waiting on any pending processes.")
break
loopperf.__enter__()
loopperf.__exit__()
- while self.processes:
- self.workflow_eval_lock.wait(1)
+ while (self.task_queue.in_flight + len(self.processes)) > 0:
+ if self.task_queue.error is not None:
+ raise self.task_queue.error
+ self.workflow_eval_lock.wait(3)
except UnsupportedRequirement:
raise
except:
- if sys.exc_info()[0] is KeyboardInterrupt:
- logger.error("Interrupted, marking pipeline as failed")
+ if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit:
+ logger.error("Interrupted, workflow will be cancelled")
else:
logger.error("Execution failed: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
if self.pipeline:
body={"priority": "0"}).execute(num_retries=self.num_retries)
finally:
self.workflow_eval_lock.release()
+ self.task_queue.drain()
self.stop_polling.set()
self.polling_thread.join()
+ self.task_queue.join()
if self.final_status == "UnsupportedRequirement":
raise UnsupportedRequirement("Check log for details.")
help="Docker image for workflow runner job, default arvados/jobs:%s" % __version__,
default=None)
+ parser.add_argument("--submit-request-uuid", type=str,
+ default=None,
+ help="Update and commit supplied container request instead of creating a new one (containers API only).")
+
parser.add_argument("--name", type=str,
help="Name to use for workflow execution instance.",
default=None)
action="store_true", default=False,
help=argparse.SUPPRESS)
+ parser.add_argument("--thread-count", type=int,
+ default=4, help="Number of threads to use for job submit and output collection.")
+
exgroup = parser.add_mutually_exclusive_group()
exgroup.add_argument("--trash-intermediate", action="store_true",
default=False, dest="trash_intermediate",
"http://arvados.org/cwl#ReuseRequirement"
])
-def main(args, stdout, stderr, api_client=None, keep_client=None):
+def exit_signal_handler(sigcode, frame):
+ logger.error("Caught signal {}, exiting.".format(sigcode))
+ sys.exit(-sigcode)
+
+def main(args, stdout, stderr, api_client=None, keep_client=None,
+ install_sig_handlers=True):
parser = arg_parser()
job_order_object = None
arvargs = parser.parse_args(args)
+ if install_sig_handlers:
+ arv_cmd.install_signal_handlers()
+
if arvargs.update_workflow:
if arvargs.update_workflow.find('-7fd4e-') == 5:
want_api = 'containers'
keep_client = arvados.keep.KeepClient(api_client=api_client, num_retries=4)
runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client,
num_retries=4, output_name=arvargs.output_name,
- output_tags=arvargs.output_tags)
+ output_tags=arvargs.output_tags,
+ thread_count=arvargs.thread_count)
except Exception as e:
logger.error(e)
return 1
projects / arvados.git / blobdiff