Closed bryce-anderson closed 11 months ago
Hi! Try checking the sequences on https://github.com/netty/netty-incubator-transport-io_uring/pull/193 explained in https://github.com/netty/netty-incubator-transport-io_uring/issues/192 ( exactly at https://github.com/netty/netty-incubator-transport-io_uring/issues/192#issuecomment-1435012607)
I have basically logged all the sqe events and cqe events + dump the threads (as you have done already) to solve it
Hi @franz1981! I think I have it sorted out via https://github.com/netty/netty-incubator-transport-io_uring/pull/227. The logging showed that for stalled loops we never enter the .submitAndWait()
call so we never flushed the initial interest in the eventFd
.
The latest commit shows how the sequence can happen deterministically but it requires modifications to the IOUringEventLoop
that clearly wouldn't be acceptable to merge.
I'm trying to debug a problem where I can't close down an IOUring event loop which ends up stuck in the
ioUringWaitCqe
state. This seems to only happen if the quiet period and graceful shutdown window are 0, or a very small duration.