❓ We occasionally do

server

reboots for installing latest security patches on instances. We've noticed elevated failure rates on pinot queries during these times. We use replication factor of 3 to ensure HA. My understanding was that

broker

would gracefully handle

dead

server and scatter queries to a replica group that contains only

alive

servers. Is that not the case? If not, is there a recommended path to taking a server at-a-time offline to perform reboot while maintaining full availability of the cluster? Does `disable`/`enable`-ing the server state (using

​/instances​/{instanceName}​/state

) during reboot cycle help here?

Broker should not route the query to the server as long as it went through graceful shutdown. Sounds like the way you are restarting it, it is not shutting down gracefully

awesome, thanks for the input! we are not currently doing a graceful shutdown, will give this a try. 🙇