Comment 18 for bug 1621541

I tested nova performance in such cases:

1) with this patch https://review.openstack.org/#/c/391869/, all controllers up:

http://paste.openstack.org/show/587643/

haproxy log - http://paste.openstack.org/show/587651/

2) with this patch https://review.openstack.org/#/c/391869/, 1 controller down:

http://paste.openstack.org/show/587644/

there are big times of request serving in keystone, it can be seen in haproxy log - http://paste.openstack.org/show/587649/

3) with this patch https://review.openstack.org/#/c/391869/, and keystone pointed to local memcached, 1 controller down:

http://paste.openstack.org/show/587647/

4) without patch https://review.openstack.org/#/c/391869/, keystone is pointed to local memcached, 1 controller down:

http://paste.openstack.org/show/587646/

So as you can see performance is improved only when:

nova.conf:
[keystone_authtoken]
memcached_servers=*local_ip*:port

AND

keystone.conf:
[cache]
memcache_servers = *local_ip*:port

so @bbobrov how do you think is it possible to set memcache_servers = *local_ip*:port in keystone.conf? or it can lead to some issues?