memcached_servers timeout causes poor API response time

[Craig Anderson]

In nova.conf, when configured for HA by setting the memcached_servers parameter to several memcached servers in the nova API cluster, e.g.:

memcached_servers=192.168.50.11:11211,192.168.50.12:11211,192.168.50.13:11211

If there are memcached servers on this list that are down, the time it takes to complete Nova API requests increases from < 1 second to 3-6 seconds.

It seems to me that Nova should protect itself from such performance degradation in an HA scenario.

[Sahid Orentino (sahid-ferdjaoui)]

This look to be related with the client python-memcached.

https://bugs.launchpad.net/python-memcached/+bug/794550

[Sahid Orentino (sahid-ferdjaoui)]

Actually after some investigation it looks like we use the version 1.48 and this version handles a param 'socket_timeout' in the client constructor.

We can add a option to configure it.

[Sahid Orentino (sahid-ferdjaoui)]

But this is not related to nova. It's related to oslo-incubator

[Bogdan Dobrelya (bogdando)]

Action items: add new option which configures socket_timeout for python-memcached