Virtualization management requires processing continuous streams of data, mostly for monitoring purposes. It is therefore challenging to make sure the different services are able to act properly, rather than become overloaded with incoming data. "Reactive" concepts can help overcome such challenges. In this presentation I'll show how oVirt embraced these concepts and evolved to Reactive architecture, and show the impact it had on overall system performance and stability.
During the process of refactoring how oVirt processes data we have been learning what it is required to change model from proactive to reactive. I want to share our successes and failures during the process because we are/were heavily constrained by backward compatibility. People thinking about performing similar changes would benefit by learning from mistakes that we made.