Re: [PATCH 0/5] Introduce 'yank' oob qmp command to recover from hanging qemu

[Kevin Wolf <kwolf@redhat.com>]

Am 13.05.2020 um 13:58 hat Paolo Bonzini geschrieben:
> On 13/05/20 13:26, Daniel P. Berrangé wrote:
> > Are you referring to the in-kernel NFS client hangs here ?  AFAIK, it is
> > impossible to do anything to get out of those hangs from userspace, because
> > the thread is stuck in an uninterruptable sleep in kernel space.
> > 
> > If using the in-QEMU NFS client, then there is a network connection that
> > can be yanked just like the NBD client.
> 
> But it's a bad idea to yank it (and also the NBD client) because you're
> not sure which wites have made it to the server (and to the medium) and
> which haven't.

This (undefined content) is generally what guests should expect when a
write request returns an error.

> Effectively, the in-QEMU NFS client and NBD client are always operating
> in "soft" mode, but we should always treat that as a bug (which cannot
> be fixed) and not a feature for read-write images.

It certainly means that you can't continue as if nothing had happened.

However, writing to the same area again (and successfully) restores a
valid state, so with werror=stop you would get into a state where you
can later reconnect and continue where you stopped, but the monitor
would stay responsive during that time.

Or if the disk isn't critical for the guest, you could replace the NBD
block node with a node that always returns I/O errors, so the guest
would see a broken disk (actually, just reporting the errors from
yanking to the guest might be enough to make it give up on the disk and
stop sending new requests). Then you can resume the VM right away.

Kevin