This is neither a bug report nor feature request. I figured I use Github issues to solicit peoples experience and knowledge about Netty's native epoll transports. We've been running Finagle with native epoll (on Linux) for quite a while internally and finally decided to switch our OSS version to it.

To our surprise, switching over to a native epoll caused out showdown tests to fail. Turns out, our previous graceful timeout (order of a couple of seconds) didn't do it for native epoll socket/channel (shutting down an HTTP server running with epoll=true). We had to bump our timeouts significantly to ensure the test passes: think of going from 2s to 30s.

Before we start digging into why closing epoll channels (server channels if that matters) takes so much longer I decided to ask around if somebody has seen this before. Any pointers would be totally appreciated!

I will try to post with thread with the updates should we know more.