Changing Druid Coordinator Cost Algorithm

Hi all,

Checking in to see if it is generally safe to test out the different Coordinator Cost algorithms by rolling restarting coordinators to change the algorithm. We are currently using the default, but are considering changing it up. I have tested out making the change in a sandbox environment and it seemed to change in place without any unexpected side effects. However, I wanted to confirm this here before making any changes in larger clusters. What I am mainly trying to validate is that the existing cluster state will not be dropped and re-loaded using the new coordination algorithm, but rather going forward, the coordinator will use said new algorithm for coordination choices going forward.


Hey Lucas,

Your understanding is right – the coordinator will not drop and reload if all you do is change the algorithm. It will also still respect the various throttles, especially maxSegmentsToMove and maxSegmentsInNodeLoadingQueue.