Hello everyone, Apache Druid restart takes 4-5 hours and during this time we wait for historical nodes. This time is too long but I don't know if it is normal? Also there are 4 historical nodes, is it advantageous for the general architecture to add more historical and add the data to more nodes? I think it is better to have fewer servers if there is no resource shortage since it will shuffle while querying the data. I wanted to ask you too. Can you help me?

hi, I guess it would depend a lot at your setup, cluster size and maybe even query type (for the shuffle you've mentioned). I can say we can have more than 5 hours time to do rolling restart of some of our clusters, but it won't really tell you much I guess.

Check out

druid.segmentCache.numLoadingThreads

... it is set to #vCPU/6 by default, but you can bump this up a bit to speed up the Historical segment loading process. I have seen it bumped up very high for cluster restarts, then brought back down after the restart was completed. You should be able to change this without losing the segment load progress, by changing the local properties file for the Historical and restarting (i.e. kill and let Overlord restart) the Historical service on the server. The existing segment cache should be recognized is already present and not try to reload.