Inconsistent query result with multiple broker

Hi all,

I deployed druid using kubernetes with multiple broker docker containers. When I run a query, the query may get load balanced into different broker. And surprisingly I get different result set. Any idea what went wrong and what to look for ?

If you’re using one of the approximate algorithms (topn/hyperunique/etc) it’s possible the differences are just due to the approximation arriving at different results at different times. The differences should be small though. If you see big differences from broker to broker then it’s possible one of your brokers is out of sync with the cluster in some way – maybe check it to see if it’s healthy (not too many GCs, has good connectivity to ZK).

Hi Gian,

Thanks for the reply. I think I might hit a basic issue. I did a simple record count query, and each query gave me a different count, and its the difference is quite big, ie. 200k, 150k, 180k, even after I stop ingestion.

Its like all the brokers are out of sync.

How do I check if all the brokers are in sync ?

I don’t see any alarming messages from the broker log.