def run(self):
while True:
# Get current time to retrieve state
env, now, now_string = self.environment, datetime_now(format=None), datetime_now()
try:
env.auto = True # Really better like that ;-)
index, event = env.events.get(now, default_value={})
print(u'[{0}] Update charts at index {1}.'.format(self.name, index))
for service, stats in env.statistics.iteritems():
label = SERVICE_TO_LABEL.get(service, service)
units_api, tasks_api = SERVICE_TO_UNITS_API[service], SERVICE_TO_TASKS_API[service]
planned = event.get(service, None)
api_client = env.api_client
api_client.auth = env.daemons_auth
if env.enable_units_status:
if env.enable_units_api:
units = getattr(api_client, units_api).list()
else:
units = env.get_units(service)
else:
units = {k: {u'agent-state': py_juju.STARTED} for k in range(planned)}
tasks = getattr(api_client, tasks_api).list(head=True) if env.enable_tasks_status else None
stats.update(now_string, planned, units, tasks)
stats.generate_units_pie_chart_by_status(env.charts_path)
stats.generate_units_line_chart(env.charts_path)
stats.generate_tasks_line_chart(env.charts_path)
stats.write()
except (ConnectionError, Timeout) as e:
# FIXME do something here ...
print(u'[{0}] WARNING! Communication error, details: {1}.'.format(self.name, e))
self.sleep()
def run(self):
while True:
# Get current time to retrieve state
env, now, now_string = self.environment, datetime_now(format=None), datetime_now()
try:
env.auto = True # Really better like that ;-)
index, event = env.events.get(now, default_value={})
print(u'[{0}] Handle scaling at index {1}.'.format(self.name, index))
for service, stats in env.statistics.iteritems():
label = SERVICE_TO_LABEL.get(service, service)
units_api = SERVICE_TO_UNITS_API[service]
planned = event.get(service, None)
if env.enable_units_api:
api_client = env.api_client
api_client.auth = env.daemons_auth
units = getattr(api_client, units_api).list()
else:
units = env.get_units(service)
if len(units) != planned:
print(u'[{0}] Ensure {1} instances of service {2}'.format(self.name, planned, label))
env.ensure_num_units(service, service, num_units=planned)
env.cleanup_machines() # Safer way to terminate machines !
else:
print(u'[{0}] Nothing to do !'.format(self.name))
# Recover faulty units
# FIXME only once and then destroy and warn admin by mail ...
for number, unit_dict in units.iteritems():
if unit_dict.get(u'agent-state') in ERROR_STATES:
unit = u'{0}/{1}'.format(service, number)
juju_do(u'resolved', environment=env.name, options=[u'--retry', unit], fail=False)
except (ConnectionError, Timeout) as e:
# FIXME do something here ...
print(u'[{0}] WARNING! Communication error, details: {1}.'.format(self.name, e))
self.sleep()
def read_or_default(environment, service, **kwargs):
label = SERVICE_TO_LABEL.get(service, service)
filename = os.path.join(STATISTICS_PATH, u'{0}_{1}.pkl'.format(environment, label))
statistics = ServiceStatistics.read(filename, store_filename=True, create_if_error=True,
environment=environment, service=service, **kwargs)
print(u'Read {0} with {1} measures.'.format(os.path.basename(filename), len(statistics.units_planned)))
if statistics.unknown_states:
print(u'[WARNING] Unknown states: {0}.'.format(statistics.unknown_states))
return statistics
def service_label(self):
return SERVICE_TO_LABEL.get(self.service, self.service)