Re: [PATCH v2 3/5] a standone-alone tool to directly share disk image file via vhost-user protocol

[Kevin Wolf]

Am 31.01.2020 um 17:42 hat Coiby Xu geschrieben:
> > Yes, I think at least for the moment it should work fine this way.
> > Eventually, I'd like to integrate it with --export (and associated QMP
> > commands, which are still to be created), too. Maybe at that point we
> > want to make the QOM object not user creatable any more.
> 
> Does it mean TYPE_USER_CREATABLE interface in QOM will become
> deprecated in the future? I'm curious what are the reasons for making
> QOM object no user creatable? Because we may still need to start
> vhost-user block device backend through HMP or QMP instead of stating
> it as a standalone-alone daemon.

Not in general, but if we have something like a block-export-add QMP
command, the QOM interface would be redundant. We could still leave it
there and have both a low-level and a high-level interface, but whether
we would want to is something we still have to decide.

> > As for test cases, do you think it would be hard to just modify the
> > tests to send an explicit 'quit' command to the daemon?
> 
> Accroding to https://patchew.org/QEMU/address@hidden/address@hidden/,
> 
> > +static bool exit_requested = false;
> > +
> > +void qemu_system_killed(int signal, pid_t pid)
> > +{
> > +    exit_requested = true;
> > +}
> 
> if exit_requested = true, qemu-storage-daemon will exit the main loop
> and then quit. So is calling qemu_system_killed by what you means "to
> send an explicit 'quit' command to the daemon"?

qemu_system_killed() is call in the signal handlers for, amongst others,
SIGTERM and SIGINT. This is one way to stop the storage daemon (for
manual use, sending SIGINT with Ctrl-C is probably the easiest way).

What I actually meant is the 'quit' QMP command which will cause
qmp_quit() to be run, which contains the same code. But if sending a
signal is more convenient, that's just as good.

Kevin

> On Fri, Jan 17, 2020 at 6:12 PM Kevin Wolf <address@hidden> wrote:
> >
> > Am 17.01.2020 um 09:12 hat Coiby Xu geschrieben:
> > > Excellent! I will add an option (or object property) for
> > > vhost-user-blk server oject which will tell the daemon process to exit
> > > when the client disconnects, thus "make check-qtest" will not get held
> > > by this daemon process. After that since Kevin's qemu-storage-daemon
> > > support "-object" option
> > > (https://patchew.org/QEMU/address@hidden/address@hidden/)
> > > and vhost-user-server is a user-creatable QOM object, it will work out
> > > of the box.
> >
> > Yes, I think at least for the moment it should work fine this way.
> > Eventually, I'd like to integrate it with --export (and associated QMP
> > commands, which are still to be created), too. Maybe at that point we
> > want to make the QOM object not user creatable any more.
> >
> > Would it make sense to prefix the object type name with "x-" so we can
> > later retire it from the external user interface without a deprecation
> > period?
> >
> > As for test cases, do you think it would be hard to just modify the
> > tests to send an explicit 'quit' command to the daemon?
> >
> > > I'm curious when will be formal version of qemu-storage-daemon
> > > finished so I can take advantage of it? Or should I apply the RFC
> > > PATCHes to my working branch directly and submit them together with
> > > the patches on vhost-user-blk server feature when posting v3?
> >
> > It's the next thing I'm planning to work on after completing the
> > coroutine-base QMP handlers (which I hope to get finished very soon).
> >
> > For the time being I would suggest that you put any patches that depend
> > on qemu-storage-daemon (if you do need it) at the end of your series so
> > that we could apply the first part even if the storage daemon isn't in
> > yet.
> >
> > The latest version of my patches is at:
> >
> >     git://repo.or.cz/qemu/kevin.git storage-daemon
> >
> > But if you just need something for testing your code, I think it would
> > even make sense if you kept your standalone tool around (even though
> > we'll never merge it) and we'll deal with integration in the storage
> > daemon once both parts are ready.
> >
> > Kevin
> >
> 
> 
> -- 
> Best regards,
> Coiby
>