You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@kafka.apache.org by Robert Quinlivan <rq...@signal.co> on 2017/03/01 17:33:50 UTC

Messages rejected by broker due to size

Hello,

I'm looking for a bit of clarification on the outcome of a
RecordTooLargeException in the broker that was not evident from the
documentation.

First, what happens to a large record that is rejected? Is it ignored by
the broker or is there a way to recover a rejected message?

Second, is the message size condition checked before or after
compression/batching is applied? What exactly is "message.max.bytes"
looking at?

Third, what happens if a broker accepts the record from the producer but
the "replica.fetch.max.bytes" setting is lower than the "message.max.bytes"
setting? This seems like an edge case to me. The leader would accept the
record but replicas would not be able to receive it, so it would be lost.
Or does the replica take the max of those two settings in order to avoid
this condition?

Thanks in advance!
-- 
Robert Quinlivan
Software Engineer, Signal