def _get_capacity(self):
assert self.state.IsInitialized()
assert self.master
# If there are no resources in the state we're done.
if len(self.state.resource) == 0:
return
# Creates the RPC request object.
request = GetCapacityRequest()
request.client_id = self.state.client_id
# Goes through all the resources in the state and adds a subrequest
# for that resource to the request.
for res in self.state.resource:
req = request.resource.add()
req.resource_id = res.resource_id
req.priority = res.priority
req.wants = res.wants
if res.HasField('has'):
req.has.CopyFrom(res.has)
# Calls the GetCapacity RPC in the master.
response = self.master.GetCapacity_RPC(request)
# If that failed we did not get new capacities. Blimey! Most probably
# the server we talked to is no longer the master, so schedule a discovery
# asap. When that discovery succeeds it will call us again.
if not response:
logger.error('%s GetCapacity request failed!' % self.get_client_id())
Counter.get('client.GetCapacity_RPC.failure').inc()
return False
else:
# Goes through the response and copies the capacity information back into
# the client state.
for r in response.response:
assert r.gets.capacity >= 0
resource = self._find_resource(r.resource_id)
resource.has.CopyFrom(r.gets)
# Schedules an action at the expiry time to clear out the lease.
scheduler.add_absolute(
resource.has.expiry_time,
lambda: self._maybe_lease_expired(r.resource_id))
if r.HasField('safe_capacity'):
resource.safe_capacity = r.safe_capacity
else:
resource.ClearField('safe_capacity')
return True
def find_resource(self, resource_id):
# This can only happen in the master!
assert self.server.is_master()
# Lineair scan through all the resources in the state to find this
# resource.
for r in self.wrapped_state.resource:
if r.resource_id == resource_id:
return r
# The resource was not found. Find the resource template which
# describes this resource.
t = global_config.find_resource_template(resource_id)
# No template found.
if not t:
logger.error(
'Cannot create server state entry for resource %s (no template)' %
resource_id)
return None
# We got the template. Now create a new blank resource entry in the server
# state.
logger.info(
'%s creating new resource %s' %
(self.get_server_id(), resource_id))
r = self.wrapped_state.resource.add()
r.template.CopyFrom(t)
r.resource_id = resource_id
# Calculates the time this resource went out of learning mode.
# Note: This time may be in the past.
r.learning_mode_expiry_time = self.wrapped_state.election_victory_time + \
AlgorithmImpl.create(t, self.wrapped_state.server_level).get_max_lease_duration()
if r.learning_mode_expiry_time > clock.get_time():
logger.info(
'%s putting resource %s in learning mode until T=%d' %
(self.get_server_id(), resource_id,
r.learning_mode_expiry_time))
# Schedules an action to log a message when this resource leaves
# learning mode.
scheduler.add_absolute(
r.learning_mode_expiry_time + 1, _LeaveLearningMode(),
(self.get_server_id(), r.resource_id))
# Note: At this point this server has not capacity lease for this resource.
# It is up to the caller to deal with this.
assert r.IsInitialized()
return r
def find_resource(self, resource_id):
# This can only happen in the master!
assert self.server.is_master()
# Lineair scan through all the resources in the state to find this
# resource.
for r in self.wrapped_state.resource:
if r.resource_id == resource_id:
return r
# The resource was not found. Find the resource template which
# describes this resource.
t = global_config.find_resource_template(resource_id)
# No template found.
if not t:
logger.error(
'Cannot create server state entry for resource %s (no template)'
% resource_id)
return None
# We got the template. Now create a new blank resource entry in the server
# state.
logger.info('%s creating new resource %s' %
(self.get_server_id(), resource_id))
r = self.wrapped_state.resource.add()
r.template.CopyFrom(t)
r.resource_id = resource_id
# Calculates the time this resource went out of learning mode.
# Note: This time may be in the past.
r.learning_mode_expiry_time = self.wrapped_state.election_victory_time + \
AlgorithmImpl.create(t, self.wrapped_state.server_level).get_max_lease_duration()
if r.learning_mode_expiry_time > clock.get_time():
logger.info('%s putting resource %s in learning mode until T=%d' %
(self.get_server_id(), resource_id,
r.learning_mode_expiry_time))
# Schedules an action to log a message when this resource leaves
# learning mode.
scheduler.add_absolute(r.learning_mode_expiry_time + 1,
_LeaveLearningMode(),
(self.get_server_id(), r.resource_id))
# Note: At this point this server has not capacity lease for this resource.
# It is up to the caller to deal with this.
assert r.IsInitialized()
return r
def process_capacity_response(self, response):
for resp in response.resource:
assert resp.gets.capacity >= 0
resource = self.find_resource(resp.resource_id)
n = sum_leases(resource)
if resp.gets.capacity < n:
logger.warning(
'%s shortfall for %s: getting %lf, but has %lf outstanding leases' %
(self.get_server_id(), resource.resource_id,
resp.gets.capacity, n))
Counter.get('server_capacity_shortfall').inc()
Gauge.get('server.%s.shortfall' %
self.get_server_id()).set(resp.gets.capacity - n)
resource.has.CopyFrom(resp.gets)
# Schedules an action at the expirty time to clear out the lease.
scheduler.add_absolute(
resource.has.expiry_time,
lambda: self._maybe_lease_expired(resource.resource_id))
def process_capacity_response(self, response):
for resp in response.resource:
assert resp.gets.capacity >= 0
resource = self.find_resource(resp.resource_id)
n = sum_leases(resource)
if resp.gets.capacity < n:
logger.warning(
'%s shortfall for %s: getting %lf, but has %lf outstanding leases'
% (self.get_server_id(), resource.resource_id,
resp.gets.capacity, n))
Counter.get('server_capacity_shortfall').inc()
Gauge.get('server.%s.shortfall' %
self.get_server_id()).set(resp.gets.capacity - n)
resource.has.CopyFrom(resp.gets)
# Schedules an action at the expirty time to clear out the lease.
scheduler.add_absolute(
resource.has.expiry_time,
lambda: self._maybe_lease_expired(resource.resource_id))
def scenario_six(reporter):
scenario_five(reporter)
reporter.set_filename('scenario_six')
scheduler.add_absolute(150, lambda: spike())
def scenario_seven(reporter):
scenario_five(reporter)
reporter.set_filename('scenario_seven')
scheduler.add_absolute(60, lambda: random_mishap())
def scenario_six(reporter):
scenario_five(reporter)
reporter.set_filename('scenario_six')
scheduler.add_absolute(150, lambda: spike())
def scenario_seven(reporter):
scenario_five(reporter)
reporter.set_filename('scenario_seven')
scheduler.add_absolute(60, lambda: random_mishap())
