On Tue, 2021-08-10 at 01:28 -0700, Nadav Amit wrote: > > Happy it could help. > > Unfortunately, there seems to be yet another issue (unless my code > somehow caused it). It seems that when SQPOLL is used, there are > cases > in which we get stuck in io_uring_cancel_sqpoll() when > tctx_inflight() > never goes down to zero. > > Debugging... (while also trying to make some progress with my code) You are on something. io_uring starts to be very solid but it isn't 100% flawless yet. I am a heavy user of SQPOLL which now run flawlessly for me with 5.13.9 (Was running flawlessly since 5.12 minus few patches I did submit recently) with my simple use-case (my SQPOLL thread isn't spawning any threads like in your use-case). The best is yet to come. I'm salivating by seeing all the performance optimizations that Jens and Pavel are putting in place lately...
