Scott, We are not yet able to fix this issue.
Our findings suggest that we have one housekeeping application which cleanups old znodes ( we are yet not using ttl znodes) every hour. This application reads thousands of znodes using get children command which is causing throttling for other client applications and results in client disconnects. Interestingly application works fine with NIO server cxn factory in the same ZK version. So we believe the issue is at Netty level. I am waiting for someone from ZK dev team who can help us to understand this netty behaviour. There might be some netty limitations. Thanks Srikant Kalani On Thu, 30 Jul 2020 at 9:08 PM, Scott Guminy <[email protected]> wrote: > Srikant, > > Did you ever resolve this issue? I might be seeing something similar. I'm > also on 3.5.5 with quorum SSL enabled. > > On Mon, Jul 13, 2020 at 10:42 PM shrikant kalani <[email protected] > > > wrote: > > > When I turned on Debug logs I can see the requests are getting throttled > > and even PING request from client is throttled. Later event buffer was > full > > and it discarded the Autoread.enable message. > > > > Is there a way to avoid discarding of message or increase the throttling > > limit ? > > > > The cluster memory is well under controlled. > > > > Thanks > > Srikant Kalani > > > > On Mon, 13 Jul 2020 at 11:21 PM, shrikant kalani < > [email protected] > > > > > wrote: > > > > > Adding one more email list > > > > > > On Mon, 13 Jul 2020 at 10:49 PM, shrikant kalani < > > [email protected]> > > > wrote: > > > > > >> Hi > > >> > > >> We are seeing a very uncommon behaviour. We implemented SSL for quorum > > >> communication in version 3.5.5. After the change we are seeing ZK > client > > >> applications are frequently getting crashed with session expired > > message. > > >> While there are no signs of GC in the application, it is hard to > figure > > out > > >> why ZK server is disconnecting those application. > > >> > > >> Any thoughts ? > > >> > > >> Thanks > > >> Srikant Kalani > > >> > > > > > >
