I was doing performance test on Queens. When I counted the number of connections to rabbitmq, found that zun-api's count was too large.
1 nova-consoleaut
3 nova-scheduler
4 nova-compute
4 zun-compute
5 neutron-dhcp-ag
7 neutron-l3-agen
11 neutron-linuxbr
42 neutron-metadat
80 nova-api
171 nova-conductor
329 neutron-server
3580 zun-api
And it increases as I invoke the zunclient, but never gets down.
Here is my fix:
diff --git a/zun/zun/api/hooks.py b/zun/zun/api/hooks.py
index 1f2d43b..ef25881 100644
--- a/zun/zun/api/hooks.py
+++ b/zun/zun/api/hooks.py
@@ -80,6 +80,8 @@ class RPCHook(hooks.PecanHook):
context = state.request.context
state.request.compute_api = compute_api.API(context)
+ def after(self, state):
+ state.request.compute_api.rpcapi.cleanup()
class NoExceptionTracebackHook(hooks.PecanHook):
"""Workaround rpc.common: deserialize_remote_exception.
diff --git a/zun/zun/common/rpc_service.py b/zun/zun/common/rpc_service.py
index 157ffb4..2f3656b 100644
--- a/zun/zun/common/rpc_service.py
+++ b/zun/zun/common/rpc_service.py
@@ -112,3 +112,6 @@ class API(object):
def echo(self, message):
self._cast('echo', message=message)
+
+ def cleanup(self):
+ self._client.transport.cleanup()
Fix proposed to branch: master
Review: https:/