[GitHub] [bookkeeper] fpj commented on issue #2482: Bookkeeper client might not close the channel for several minutes after a Bookie crashes

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

fpj commented on issue #2482:
URL: https://github.com/apache/bookkeeper/issues/2482#issuecomment-725364061


   This situation seems to be happening because `channelInactive` is called when the underlying TCP closes, and it is taking long to time out in the case the other end of the connection is gone.
   
   I see here in `PerChannelBookieClient`:
   
   https://github.com/apache/bookkeeper/blob/master/bookkeeper-server/src/main/java/org/apache/bookkeeper/proto/PerChannelBookieClient.java#L567
   
   that we are setting `SO_KEEPALIVE`, but the default configuration is such that the first probe is really long after the last packet sent. One option might be to configure the related options, like `TCP_KEEPIDLE`, `TCP_KEEPINTVL`, and `TCP_KEEPCNT`. `TCP_USER_TIMEOUT` is another option.
   
   I'm a bit puzzled about the 15-minute value interval, though. I see references to docker swarm issues, but I can't map to the scenario we have.


----------------------------------------------------------------
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