Re: [Bug 208827] [fio io_uring] io_uring write data crc32c verify failed

[Jens Axboe <axboe@kernel.dk>]

On 8/12/20 9:24 AM, Jeff Moyer wrote:
> Jens Axboe <axboe@kernel.dk> writes:
> 
>> Yes, it would ideally wait, or at least trigger on the last one. I'll
>> see if I can improve that. For any of my testing, the amount of
>> triggered short reads is minimal. For the verify case that we just
>> ran, we're talking 8-12 ios out of 820 thousand, or 0.001% of them.
>> So nothing that makes a performance difference in practical terms,
>> though it would be nice to not hand back short reads if we can avoid
>> it. Not for performance reasons, but for usage reasons.
> 
> I think you could make the case that handing back a short read is a
> bug (unless RWF_NOWAIT was specified, of course).  At the very least,
> it violates the principle of least surprise, and the fact that it
> happens infrequently actually makes it a worse problem when it comes
> to debugging.

It's definitely on my list to ensure that we handle the somewhat
expected case by just retrying it internally, because I do agree that it
can be surprising. FWIW, this isn't a change with 5.9-rc, io_uring has
always potentially short buffered reads when when the io-wq offload was
done.

While it may happen infrequently with this specific test case, others
can trigger it more often.

-- 
Jens Axboe