On Tue, Aug 6, 2024 at 11:08 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> On Sun, Aug 04, 2024 at 04:02:51PM +0800, Yafang Shao wrote:
> > One solution we're currently exploring is leveraging the preadv2(2)
> > syscall. By using the RWF_NOWAIT flag, preadv2(2) can avoid the XFS inode
> > lock hung task. This can be illustrated as follows:
> >
> > retry:
> > if (preadv2(fd, iovec, cnt, offset, RWF_NOWAIT) < 0) {
> > sleep(n)
> > goto retry;
>
> But that's not how you're supposed to use RWF_NOWAIT! You're supposed
> to try it _once_ in the thread that can't block, then hand the I/O off
> to another thread which _can_ block. Because that single thread is the
> one which does all the blocking I/O, there's no lock contention.
>
> So this is a kernel workaround for bad application design, and should
> be rejected.
They are different applications, but not different threads within a
single application.
A simple example:
$ ps -eLo pid,comm
1 systemd
2 tail -f /var/log/messages.
In this case, tail is reading /var/log/messages while systemd is
writing to it. Are you suggesting we should forbid `tail -f` on Linux
servers? If you had taken the time to understand what filebeat is
doing, you wouldn't have reached this arbitrary conclusion.
--
Regards
Yafang
