Open Stack

On 03/12/2013 09:09 AM, Baktha Ashokan wrote:
> Hi There,
>   I'am trying to setup an highly redundant openstack setup. where i used
> replicated mysql DB, clustered Rabbitmq and software loadbalanced using
> HAproxy and keepalived suggested by mirantis. but while using VIP seems
> we need to change ip's and endpoint's in conf files of glance, keystone
> and nova. could you please tell me how may i achive that?
> 
> the guide i followed is
> http://www.mirantis.com/blog/software-high-availability-load-balancing-openstack-cloud-api-servic/
> 
> i followed that guide as it is but it doesnt seems to be working does
> anyone worked in it before?

We didn't follow that particular guide, but we definitely have a
highly-available OpenStack setup. For each deployment zone, we use MySQL
Galera cluster, spread over four nodes for write scaling. On those same
four nodes we also have RabbitMQ set up. On another four nodes, we run
the OpenStack "controller stack": nova API endpoints, nova scheduler,
cinder API and scheduler, glance API and registry, nova-cert, and dashboard.

Each endpoint has an associated VIP in our load balancer setup that
round-robins requests across the four underlying controller or
database/message queue nodes.

We use an active/passive setup for our F5 BigIP load balancers and an
active/active setup for all network switches, from top-of-racks out to
the spine switches.

Works a charm. We have seen an issue where we get connection reset by
peer messages occasionally from the AMQP code in Oslo, but we're not
sure if it's something to do with our setup or something to do with the
upstream code, which has recently been enhanced by many folks, including
Ray P from Dell, and we plan on moving to that codebase sooner rather
than later.

Best,
-jay