also after starting the server, you need to wait for health checks to make sure everything is loaded before moving on to next server

still not sure why queries failing. If we have only 3 servers and 3 replicas, failure of 1 should not break query as there is enough data. Can it be an issue with the client?

what client are you using

via presto

And pinot

are you restarting pinot brokers as well or just pinot server

Everything, with rollingupgrade k8s strategy, 1 node at a time.