On Wed, 25 Oct 2023 17:27:57 GMT, Daniel Jeliński <[email protected]> wrote:
> This patch fixes a race between the selector thread and the IdleTimeoutTask > which closes connections that were idle for longer than IDLE_INTERVAL. With > this patch, only the thread that successfully removes the connection from > idleConnections is permitted to change it. If any data arrives after the > IdleTimeoutTask removes the connection from idleConnections, the data is > ignored. > > Additionally, this patch reduces the time interval between when the > IdleTimeoutTask removes the connection from idleConnections and when the > connection is closed, back to pre-JDK-8286918 level. > > No new test; with platform sockets it's next to impossible to reliably get > the timing right. Existing tier1-3 tests pass. Thank you Daniel for addressing this. This looks OK to me, I just have one related question which I've added inline. Finally, given this relates to intermittent/racy issue, it would be good to verify that this change continues to work with a test-repeat of maybe 50. ------------- PR Comment: https://git.openjdk.org/jdk/pull/16366#issuecomment-1784543527
