This message was deleted.

Each groupBy query could use up to 2 merge buffers (

druid.processing.numMergeBuffers

) to process the query. For higher concurrency of group by queries, you might want to increase that. Make sure that you have enough direct memory in the JVM setting

-XX:MaxDirectMemorySize

to accommodate the extra merge buffers. The full formula for direct memory is here.

These changes shouldn’t affect the availability of the merge buffer.