Anyone who s ever tried to build an OpenStack cloud knows how difficult it is to bring together all the components and tune them for optimized performance, especially for Highly Available installations. That s why so many deployment automation tools have emerged, including Fuel, Crowbar, Packstack, etc. Although it might seem straightforward to make all the OpenStack services HA after you complete a simple multi-host installation phase, be careful, as there will be dragons: AMQP message brokers not handling failover correctly, the database being put under a risk of corruption or need of manual reassembling, and the whole deployment finally being interfered by the way OpenStack messaging and database code is working.
In this talk, we ll explain how we re overcoming these problems in Mirantis OpenStack by using techniques such as:
- Patching the vanilla OpenStack code
- Optimizing deployment parts
- Modifying Puppet providers and other tooling, to make the whole OpenStack cluster work as true Highly Available software
After attending this presentation, you ll know how to create a working HA cluster with either Mirantis OpenStack or vanilla OpenStack, and perhaps learn other techniques to help make your own installation more robust.