def test_round_robin_execute(self):
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
for client, arg in [(1, 'foo'), (2, 'bar'), (0, 'baz'), (1, 'bax')]:
result = cluster.execute(arg)
self.clients[client].execute.assert_called_with(arg)
self.assertEqual(self.clients[client].execute.return_value, result)
def test_round_robin_execute(self):
"""
seed_clients are chosen in round robin manner
"""
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
for client, arg in [(1, 'foo'), (2, 'bar'), (0, 'baz'), (1, 'bax')]:
result = cluster.execute(arg)
self.clients[client].execute.assert_called_with(arg)
self.assertEqual(self.clients[client].execute.return_value, result)
def test_other_error_propogated(self):
"""
Any error other than subclass of ``ConnectError`` is propgoated
"""
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
rand_err = ValueError('random err')
self.clients[1].execute.return_value = defer.fail(rand_err)
result = cluster.execute(2, 3)
self.assertEqual(self.failureResultOf(result).value, rand_err)
self.assertEqual(self.clients[2].execute.called, False)
self.assertEqual(self.clients[0].execute.called, False)
def test_one_node_down(self):
"""
If a cass node is down, it tries the next node in cluster
"""
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
self.clients[1].execute.return_value = defer.fail(ConnectionRefusedError())
result = cluster.execute(2, 3)
self.assertEqual(self.successResultOf(result), 'exec_ret3')
self.clients[1].execute.assert_called_once_with(2, 3)
self.clients[2].execute.assert_called_once_with(2, 3)
self.assertFalse(self.clients[0].execute.called)
def test_other_error_propogated_on_node_down(self):
"""
If first node gives ``ConnectError`` then second node is tried if it gives
error other than subclass of ``ConnectError`` it is propogated
"""
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
self.clients[1].execute.return_value = defer.fail(NoRouteError())
rand_err = ValueError('random err')
self.clients[2].execute.return_value = defer.fail(rand_err)
result = cluster.execute(2, 3)
self.assertEqual(self.failureResultOf(result).value, rand_err)
self.assertEqual(self.clients[0].execute.called, False)
def test_all_nodes_down(self):
"""
If all cass nodes are down, it gives up eventually by raising the
connection error exception
"""
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
err = ConnectionRefusedError()
for i in range(3):
self.clients[i].execute.side_effect = lambda *_: defer.fail(err)
result = cluster.execute(2, 3)
self.assertEqual(self.failureResultOf(result).value, err)
for i in range(3):
self.clients[i].execute.assert_called_once_with(2, 3)
def test_multiple_clients(self):
"""
When multiple clients execute simultaneously, it does not skip nodes. It basically ensures
that http://bit.ly/1csEIRR situation does not happen
"""
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
def _execute(*args):
cluster.execute(2, 3)
return defer.fail(NoRouteError())
self.clients[1].execute.side_effect = _execute
result = cluster.execute(2, 3)
self.assertEqual(self.successResultOf(result), 'exec_ret3')
self.assertEqual(self.clients[2].execute.call_args_list, [mock.call(2, 3)] * 2)
self.clients[1].execute.assert_called_once_with(2, 3)
def connect_cass_servers(reactor, config):
"""
Connect to Cassandra servers and return the connection
"""
seed_endpoints = [clientFromString(reactor, str(host))
for host in config['seed_hosts']]
return RoundRobinCassandraCluster(
seed_endpoints, config['keyspace'], disconnect_on_cancel=True)
def makeService(config):
"""
Set up the otter-api service.
"""
set_config_data(dict(config))
if not config_value('mock'):
seed_endpoints = [
clientFromString(reactor, str(host))
for host in config_value('cassandra.seed_hosts')
]
cassandra_cluster = LoggingCQLClient(
RoundRobinCassandraCluster(seed_endpoints,
config_value('cassandra.keyspace')),
log.bind(system='otter.silverberg'))
set_store(CassScalingGroupCollection(cassandra_cluster))
bobby_url = config_value('bobby_url')
if bobby_url is not None:
set_bobby(BobbyClient(bobby_url))
cache_ttl = config_value('identity.cache_ttl')
if cache_ttl is None:
# FIXME: Pick an arbitrary cache ttl value based on absolutely no
# science.
cache_ttl = 300
authenticator = CachingAuthenticator(
reactor,
ImpersonatingAuthenticator(config_value('identity.username'),
config_value('identity.password'),
config_value('identity.url'),
config_value('identity.admin_url')),
cache_ttl)
supervisor = Supervisor(authenticator.authenticate_tenant, coiterate)
set_supervisor(supervisor)
s = MultiService()
site = Site(root)
site.displayTracebacks = False
api_service = service(str(config_value('port')), site)
api_service.setServiceParent(s)
if config_value('scheduler') and not config_value('mock'):
scheduler_service = SchedulerService(
int(config_value('scheduler.batchsize')),
int(config_value('scheduler.interval')), cassandra_cluster)
scheduler_service.setServiceParent(s)
return s
def makeService(config):
"""
Set up the otter-api service.
"""
config = dict(config)
set_config_data(config)
parent = MultiService()
region = config_value('region')
seed_endpoints = [
clientFromString(reactor, str(host))
for host in config_value('cassandra.seed_hosts')]
cassandra_cluster = LoggingCQLClient(
TimingOutCQLClient(
reactor,
RoundRobinCassandraCluster(
seed_endpoints,
config_value('cassandra.keyspace'),
disconnect_on_cancel=True),
config_value('cassandra.timeout') or 30),
log.bind(system='otter.silverberg'))
store = CassScalingGroupCollection(
cassandra_cluster, reactor, config_value('limits.absolute.maxGroups'))
admin_store = CassAdmin(cassandra_cluster)
bobby_url = config_value('bobby_url')
if bobby_url is not None:
set_bobby(BobbyClient(bobby_url))
service_configs = get_service_configs(config)
authenticator = generate_authenticator(reactor, config['identity'])
supervisor = SupervisorService(authenticator, region, coiterate,
service_configs)
supervisor.setServiceParent(parent)
set_supervisor(supervisor)
health_checker = HealthChecker(reactor, {
'store': getattr(store, 'health_check', None),
'kazoo': store.kazoo_health_check,
'supervisor': supervisor.health_check
})
# Setup cassandra cluster to disconnect when otter shuts down
if 'cassandra_cluster' in locals():
parent.addService(FunctionalService(stop=partial(
call_after_supervisor, cassandra_cluster.disconnect, supervisor)))
otter = Otter(store, region, health_checker.health_check)
site = Site(otter.app.resource())
site.displayTracebacks = False
api_service = service(str(config_value('port')), site)
api_service.setServiceParent(parent)
# Setup admin service
admin_port = config_value('admin')
if admin_port:
admin = OtterAdmin(admin_store)
admin_site = Site(admin.app.resource())
admin_site.displayTracebacks = False
admin_service = service(str(admin_port), admin_site)
admin_service.setServiceParent(parent)
# setup cloud feed
cf_conf = config.get('cloudfeeds', None)
if cf_conf is not None:
id_conf = deepcopy(config['identity'])
id_conf['strategy'] = 'single_tenant'
add_to_fanout(CloudFeedsObserver(
reactor=reactor,
authenticator=generate_authenticator(reactor, id_conf),
tenant_id=cf_conf['tenant_id'],
region=region,
service_configs=service_configs))
# Setup Kazoo client
if config_value('zookeeper'):
threads = config_value('zookeeper.threads') or 10
disable_logs = config_value('zookeeper.no_logs')
threadpool = ThreadPool(maxthreads=threads)
sync_kz_client = KazooClient(
hosts=config_value('zookeeper.hosts'),
# Keep trying to connect until the end of time with
# max interval of 10 minutes
connection_retry=dict(max_tries=-1, max_delay=600),
logger=None if disable_logs else TxLogger(log.bind(system='kazoo'))
)
kz_client = TxKazooClient(reactor, threadpool, sync_kz_client)
# Don't timeout. Keep trying to connect forever
d = kz_client.start(timeout=None)
def on_client_ready(_):
dispatcher = get_full_dispatcher(reactor, authenticator, log,
get_service_configs(config),
kz_client, store, supervisor,
cassandra_cluster)
# Setup scheduler service after starting
scheduler = setup_scheduler(parent, dispatcher, store, kz_client)
health_checker.checks['scheduler'] = scheduler.health_check
otter.scheduler = scheduler
# Give dispatcher to Otter REST object
otter.dispatcher = dispatcher
# Set the client after starting
# NOTE: There is small amount of time when the start is
# not finished and the kz_client is not set in which case
# policy execution and group delete will fail
store.kz_client = kz_client
# Setup kazoo to stop when shutting down
parent.addService(FunctionalService(
stop=partial(call_after_supervisor,
kz_client.stop, supervisor)))
setup_converger(
parent, kz_client, dispatcher,
config_value('converger.interval') or 10,
config_value('converger.build_timeout') or 3600,
config_value('converger.limited_retry_iterations') or 10,
config_value('converger.step_limits') or {})
d.addCallback(on_client_ready)
d.addErrback(log.err, 'Could not start TxKazooClient')
return parent
def test_disconnect(self):
cluster = RoundRobinCassandraCluster(['one', 'two', 'three'], 'keyspace')
cluster.disconnect()
for client in self.clients:
client.disconnect.assert_called_with()
