Re: [PATCH v6] Work around vhost-user-blk-test hang

[Michael S. Tsirkin]

On Tue, Oct 19, 2021 at 02:57:46PM +0100, Stefan Hajnoczi wrote:
> On Thu, Oct 14, 2021 at 04:32:23AM +0000, Raphael Norwitz wrote:
> > The vhost-user-blk-test qtest has been hanging intermittently for a
> > while. The root cause is not yet fully understood, but the hang is
> > impacting enough users that it is important to merge a workaround for
> > it.
> > 
> > The race which causes the hang occurs early on in vhost-user setup,
> > where a vhost-user message is never received by the backend. Forcing
> > QEMU to wait until the storage-daemon has had some time to initialize
> > prevents the hang. Thus the existing storage-daemon pidfile option can
> > be used to implement a workaround cleanly and effectively, since it
> > creates a file only once the storage-daemon initialization is complete.
> > 
> > This change implements a workaround for the vhost-user-blk-test hang by
> > making QEMU wait until the storage-daemon has written out a pidfile
> > before attempting to connect and send messages over the vhost-user
> > socket.
> > 
> > Some relevent mailing list discussions:
> > 
> > [1] 
> > https://lore.kernel.org/qemu-devel/CAFEAcA8kYpz9LiPNxnWJAPSjc=nv532bEdyfynaBeMeohqBp3A@mail.gmail.com/
> > [2] 
> > https://lore.kernel.org/qemu-devel/YWaky%2FKVbS%2FKZjlV@stefanha-x1.localdomain/
> 
> I sent a fix for the hang so the workaround is no longer needed:
> "[PATCH] vhost-user-blk-test: pass vhost-user socket fds to QSD"
> 
> Stefan

Woot! Thanks a bunch Stefan! Great job debugging this!

-- 
MST