openstack team mailing list archive

[Samuel Winchenbach <swinchen@xxxxxxxxx>]

Hi All,

I am attempting to set up a high availability openstack cluster.
 Currently, using pacemaker, I create a Virtual IP for all the highly
available service, launch haproxy to proxy all the requests and clone
keystone to all the nodes.   The idea being that the requests come into
haproxy and are load balanced across all the nodes.


To do this I have keystone listen on 26000 for admin, and 26001 for public.
 haproxy listens on 35357 and 5000 respectively (these ports are bound to
the VIP).  The problem with setup is that my log is filling (MB/min) with
this warning:

2013-06-13 09:20:18     INFO [access] 127.0.0.1 - - [13/Jun/2013:13:20:18
+0000] "GET http://10.80.255.1:35357/v2.0/users HTTP/1.0" 200 915
2013-06-13 09:20:18  WARNING [keystone.contrib.stats.core] Unable to
resolve API as either public or admin: 10.80.255.1:35357
...
...

where 10.80.255.1 is my VIP for highly available services.   I traced down
that module and added a few lines of code for debugging and it turns out
that if checks to see if the incoming connection matches a port in the
config file.  In my case it does not.

I can not just bind keystone to the internal ip and leave the port as their
defaults because the way pacemaker checks to see if services are alive is
by sending requests to service it is monitoring, and I do not want to send
requests to the VIP because any instance of keystone could respond.
Basically I would I have to write a pacemaker rule for each node and it
would become messy quite quickly.

Does anyone see something I could do differently, or a fix for my current
situation?

Thanks,
Sam