Re: [PATCH v3 3/3] multifd: Send using asynchronous write on nocomp to send RAM pages.

[Peter Xu <peterx@redhat.com>]

On Fri, Sep 24, 2021 at 06:43:49PM +0100, Daniel P. Berrangé wrote:
> > @@ -546,6 +546,7 @@ void multifd_save_cleanup(void)
> >          MultiFDSendParams *p = &multifd_send_state->params[i];
> >          Error *local_err = NULL;
> >  
> > +        qio_channel_async_flush(p->c, NULL);
> >          socket_send_channel_destroy(p->c);
> >          p->c = NULL;
> >          qemu_mutex_destroy(&p->mutex);
> 
> This isn't reliable beucase qio_channel_async_flush will return early
> even if not everything is flushed.

Right, though I think the problem is in patch 2 where we should make sure
flush() will only return if all data sent.

And at the meantime we may also want to move it to before we send the devices
states for both pre/post copy.  multifd_save_cleanup() is called only until
migration completed and we're cleaning stuffs, I'm afraid it's too late, so
potentially the device states can arrive and dest vm running without the latest
src VM memories.

-- 
Peter Xu