import threading
from neco.util import guid
+from neco.util.parallel import ParallelRun
from neco.util.timefuncs import strfnow, strfdiff, strfvalid
-from neco.execution.resource import ResourceFactory
+from neco.execution.resource import ResourceFactory, ResourceAction, \
+ ResourceState
from neco.execution.scheduler import HeapScheduler, Task, TaskStatus
-from neco.util.parallel import ParallelRun
+from neco.execution.trace import TraceAttr
+
+# TODO: use multiprocessing instead of threading
+# TODO: Improve speed. Too slow... !!
+
+class ECState(object):
+ RUNNING = 1
+ FAILED = 2
+ TERMINATED = 3
class ExperimentController(object):
- def __init__(self, root_dir = "/tmp", loglevel = 'error'):
+ def __init__(self, exp_id = None, root_dir = "/tmp"):
super(ExperimentController, self).__init__()
# root directory to store files
self._root_dir = root_dir
+ # experiment identifier given by the user
+ self._exp_id = exp_id or "nepi-exp-%s" % os.urandom(8).encode('hex')
+
# generator of globally unique ids
self._guid_generator = guid.GuidGenerator()
self._tasks = dict()
# Event processing thread
- self._stop = False
self._cond = threading.Condition()
self._thread = threading.Thread(target = self._process)
+ self._thread.setDaemon(True)
self._thread.start()
+ # EC state
+ self._state = ECState.RUNNING
+
# Logging
- self._logger = logging.getLogger("neco.execution.ec")
- self._logger.setLevel(getattr(logging, loglevel.upper()))
+ self._logger = logging.getLogger("ExperimentController")
+
+ @property
+ def logger(self):
+ return self._logger
+
+ @property
+ def ecstate(self):
+ return self._state
+ @property
+ def exp_id(self):
+ exp_id = self._exp_id
+ if not exp_id.startswith("nepi-"):
+ exp_id = "nepi-" + exp_id
+ return exp_id
+
+ @property
+ def finished(self):
+ return self.ecstate in [ECState.FAILED, ECState.TERMINATED]
+
+ def wait_finished(self, guids):
+ while not all([self.state(guid) == ResourceState.FINISHED \
+ for guid in guids]) and not self.finished:
+ time.sleep(1)
+
def get_task(self, tid):
return self._tasks.get(tid)
def resources(self):
return self._resources.keys()
- def register_resource(self, rtype, guid = None, creds = None):
+ def register_resource(self, rtype, guid = None):
# Get next available guid
guid = self._guid_generator.next(guid)
# Instantiate RM
- rm = ResourceFactory.create(rtype, self, guid, creds)
+ rm = ResourceFactory.create(rtype, self, guid)
# Store RM
self._resources[guid] = rm
rm1.connect(guid2)
rm2.connect(guid1)
- def discover_resource(self, guid, filters):
- rm = self.get_resource(guid)
- return rm.discover(filters)
+ def register_condition(self, group1, action, group2, state,
+ time = None):
+ """ Registers an action START or STOP for all RM on group1 to occur
+ time 'time' after all elements in group2 reached state 'state'.
- def provision_resource(self, guid, filters):
- rm = self.get_resource(guid)
- return rm.provision(filters)
+ :param group1: List of guids of RMs subjected to action
+ :type group1: list
- def register_start(self, group1, time, after_status, group2):
- if isinstance(group1, int):
- group1 = list[group1]
- if isinstance(group2, int):
- group2 = list[group2]
+ :param action: Action to register (either START or STOP)
+ :type action: ResourceAction
- for guid1 in group1:
- for guid2 in group2:
- rm = self.get_resource(guid)
- rm.start_after(time, after_status, guid2)
+ :param group2: List of guids of RMs to we waited for
+ :type group2: list
- def register_stop(self, group1, time, after_status, group2):
- if isinstance(group1, int):
- group1 = list[group1]
- if isinstance(group2, int):
- group2 = list[group2]
+ :param state: State to wait for on RMs (STARTED, STOPPED, etc)
+ :type state: ResourceState
- for guid1 in group1:
- for guid2 in group2:
- rm = self.get_resource(guid)
- rm.stop_after(time, after_status, guid2)
+ :param time: Time to wait after group2 has reached status
+ :type time: string
- def register_set(self, name, value, group1, time, after_status, group2):
+ """
if isinstance(group1, int):
- group1 = list[group1]
+ group1 = [group1]
if isinstance(group2, int):
- group2 = list[group2]
+ group2 = [group2]
for guid1 in group1:
- for guid2 in group2:
- rm = self.get_resource(guid)
- rm.set_after(name, value, time, after_status, guid2)
+ rm = self.get_resource(guid1)
+ rm.register_condition(action, group2, state, time)
+
+ def register_trace(self, guid, name):
+ """ Enable trace
+
+ :param name: Name of the trace
+ :type name: str
+ """
+ rm = self.get_resource(guid)
+ rm.register_trace(name)
+
+ def trace(self, guid, name, attr = TraceAttr.ALL, block = 512, offset = 0):
+ """ Get information on collected trace
+
+ :param name: Name of the trace
+ :type name: str
+
+ :param attr: Can be one of:
+ - TraceAttr.ALL (complete trace content),
+ - TraceAttr.STREAM (block in bytes to read starting at offset),
+ - TraceAttr.PATH (full path to the trace file),
+ - TraceAttr.SIZE (size of trace file).
+ :type attr: str
+
+ :param block: Number of bytes to retrieve from trace, when attr is TraceAttr.STREAM
+ :type name: int
+
+ :param offset: Number of 'blocks' to skip, when attr is TraceAttr.STREAM
+ :type name: int
+
+ :rtype: str
+ """
+ rm = self.get_resource(guid)
+ return rm.trace(name, attr, block, offset)
+
+ def discover(self, guid, filters):
+ rm = self.get_resource(guid)
+ return rm.discover(filters)
+
+ def provision(self, guid, filters):
+ rm = self.get_resource(guid)
+ return rm.provision(filters)
def get(self, guid, name):
rm = self.get_resource(guid)
rm = self.get_resource(guid)
return rm.set(name, value)
- def status(self, guid):
+ def state(self, guid):
rm = self.get_resource(guid)
- return rm.status()
+ return rm.state
def stop(self, guid):
rm = self.get_resource(guid)
return rm.stop()
- def deploy(self, group = None, start_when_all_ready = True):
+ def start(self, guid):
+ rm = self.get_resource(guid)
+ return rm.start()
+
+ def set_with_conditions(self, name, value, group1, group2, state,
+ time = None):
+ """ Set value 'value' on attribute with name 'name' on all RMs of
+ group1 when 'time' has elapsed since all elements in group2
+ have reached state 'state'.
+
+ :param name: Name of attribute to set in RM
+ :type name: string
+
+ :param value: Value of attribute to set in RM
+ :type name: string
+
+ :param group1: List of guids of RMs subjected to action
+ :type group1: list
+
+ :param action: Action to register (either START or STOP)
+ :type action: ResourceAction
+
+ :param group2: List of guids of RMs to we waited for
+ :type group2: list
+
+ :param state: State to wait for on RMs (STARTED, STOPPED, etc)
+ :type state: ResourceState
+
+ :param time: Time to wait after group2 has reached status
+ :type time: string
+
+ """
+ if isinstance(group1, int):
+ group1 = [group1]
+ if isinstance(group2, int):
+ group2 = [group2]
+
+ for guid1 in group1:
+ rm = self.get_resource(guid)
+ rm.set_with_conditions(name, value, group2, state, time)
+
+ def stop_with_conditions(self, guid):
+ rm = self.get_resource(guid)
+ return rm.stop_with_conditions()
+
+ def start_with_conditions(self, guid):
+ rm = self.get_resource(guid)
+ return rm.start_with_condition()
+
+ def deploy(self, group = None, wait_all_ready = True):
+ """ Deploy all resource manager in group
+
+ :param group: List of guids of RMs to deploy
+ :type group: list
+
+ :param wait_all_ready: Wait until all RMs are ready in
+ order to start the RMs
+ :type guid: int
+
+ """
+ self.logger.debug(" ------- DEPLOY START ------ ")
+
+ def steps(rm):
+ rm.deploy()
+ rm.start_with_conditions()
+
+ # Only if the RM has STOP consitions we
+ # schedule a stop. Otherwise the RM will stop immediately
+ if rm.conditions.get(ResourceAction.STOP):
+ rm.stop_with_conditions()
+
if not group:
group = self.resources
for guid in group:
rm = self.get_resource(guid)
- kwargs = {'target': rm.deploy}
- if start_when_all_ready:
+ if wait_all_ready:
towait = list(group)
towait.remove(guid)
- kwargs['args'] = towait
+ self.register_condition(guid, ResourceAction.START,
+ towait, ResourceState.READY)
- thread = threading.Thread(kwargs)
+ thread = threading.Thread(target = steps, args = (rm,))
threads.append(thread)
+ thread.setDaemon(True)
thread.start()
- for thread in threads:
- thread.join()
+ while list(threads) and not self.finished:
+ thread = threads[0]
+ # Time out after 5 seconds to check EC not terminated
+ thread.join(5)
+ if not thread.is_alive():
+ threads.remove(thread)
def release(self, group = None):
if not group:
rm = self.get_resource(guid)
thread = threading.Thread(target=rm.release)
threads.append(thread)
+ thread.setDaemon(True)
thread.start()
- for thread in threads:
- thread.join()
+ while list(threads) and not self.finished:
+ thread = threads[0]
+ # Time out after 5 seconds to check EC not terminated
+ thread.join(5)
+ if not thread.is_alive():
+ threads.remove(thread)
+
+ self._state = ECState.TERMINATED
def shutdown(self):
self.release()
- self._stop = True
self._cond.acquire()
self._cond.notify()
self._cond.release()
+
if self._thread.is_alive():
self._thread.join()
def schedule(self, date, callback, track = False):
- """
+ """ Schedule a callback to be executed at time date.
+
date string containing execution time for the task.
It can be expressed as an absolute time, using
timestamp format, or as a relative time matching
runner.start()
try:
- while not self._stop:
+ while not self.finished:
self._cond.acquire()
task = self._scheduler.next()
self._cond.release()
else:
# Process tasks in parallel
runner.put(self._execute, task)
- except:
+
+ except:
import traceback
err = traceback.format_exc()
self._logger.error("Error while processing tasks in the EC: %s" % err)
+ self._state = ECState.FAILED
+ return
+
+ # Mark EC state as terminated
+ if self.ecstate == ECState.RUNNING:
+ self._state = ECState.TERMINATED
+
def _execute(self, task):
# Invoke callback
task.status = TaskStatus.DONE
except:
import traceback
err = traceback.format_exc()
- self._logger.error("Error while executing event: %s" % err)
-
task.result = err
task.status = TaskStatus.ERROR
+
+ self._logger.error("Error occurred while executing task: %s" % err)
+
+ # Mark the EC as failed
+ self._state = ECState.FAILED
+
+ # Wake up the EC in case it was sleeping
+ self._cond.acquire()
+ self._cond.notify()
+ self._cond.release()
+
+ # Propage error to the ParallelRunner
+ raise