[GitHub] [activemq-artemis] franz1981 edited a comment on pull request #3566: ARTEMIS-3282 Expose replication response batching tuning

[GitBox <gi...@apache.org>]

franz1981 edited a comment on pull request #3566:
URL: https://github.com/apache/activemq-artemis/pull/3566#issuecomment-833811477


   I was expecting exposing batching size was meant to give a good speedup with kernel bypass drivers but seems not: but I still think the current implementation can still hide a subtle regression with paging/large messages.
   I'm going to perform some test to verify it, but seems related to the fact that, on backup, large msgs write and page writes are still happening on the Netty thread and given that they perform synchronous operations on file channels (some form of FileChannel::write) they can delay for some time (untile Netty decide there are no more incoming data from the replication socket) any already batched response, causing it to be delayed too much.
   
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org