Deploying OpenStack in production at any scale, upgrade support is one of the requirements to have a successful deployment. Without upgrade management, adeployment will have bugs and security issues from day 1. Also in longer term, it will miss the latest features that OpenStack offers.
People often think an upgrade is based on upgrading OpenStack packages but unfortunatly that's not enough.
You also have to deal with:
- system upgrades: new kernel, core components needed by OpenStack (apache, HAproxy, etc).
- availability: your architecture has to remain available during the upgrade
- orchestration: reduce or avoid service downtime by orchestrating an upgrade in the smartest way
- scalability: the number of nodes to upgrades should not affect the time to upgrade (ie: packaging upgrade method is too slow at scale and consumes a lot more resources)
- reproduceability: be able to test the upgrade on another plateform before doing it in production
- Testing: before and after the upgrade it's crucialto validate that no regressions are occuring, and be able to track them in the worst case.
Attend this presentation if you want to know:
- why and how we use image-based deployments
- how an orchestrated upgrade works
- how to prepare the orchestration
- and learn from our experience which is building against past mistakes that we would love you to avoid
Two years of experience on this subject, we have learnt good and bad things that are leading us to think about a new way to prepare the upgrades. After a formal presentation of our current state, we will open the discussion to a project that we envisioning: automatic computation of the orchestration