qemu-block
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v4 20/32] nbd/client-connection: implement connection retry


From: Eric Blake
Subject: Re: [PATCH v4 20/32] nbd/client-connection: implement connection retry
Date: Mon, 22 Nov 2021 15:51:45 -0600
User-agent: NeoMutt/20211029-26-f6989f

On Mon, Nov 22, 2021 at 08:17:34PM +0300, Vladimir Sementsov-Ogievskiy wrote:
> 22.11.2021 19:30, Eric Blake wrote:
> > Reviving this thread, as a good as place as any for my question:
> > 

> >   But I can't
> > find anywhere in the code base that ensures that on a reconnect, the
> > new server supplies at least as many extensions as the original
> > server, nor anywhere that we would be able to gracefully handle an
> > in-flight block status command that can no longer be successfully
> > continued because the reconnect landed on a downgraded server.
> > 

> 
> Yes that's a problem. We previously noted it here 
> https://lists.gnu.org/archive/html/qemu-block/2021-06/msg00458.html

Aha!  And oops that we've let it slide this long.  But now that we
know the problem was also present in 6.1, and not a new regression in
6.2, it is less urgent to get a fix in.  If we get one written in
time, it's still game for inclusion for -rc2 or maybe -rc3, but as we
get further along in the process, it should not hold up the final
release (it would not be -rc4 material, for example).

> 
> Honestly, I didn't start any fix for that :(.. I agree, it would be good to 
> fix it somehow in 6.2. I'll try to make something simple this week. Or did 
> you already started doing some fix?

I haven't started writing anything on the topic.  I've got some
patches that I hope to post soon targetting 7.0 that allow an
extension for NBD_CMD_BLOCK_STATUS to do a full round-trip 64-bit
operation, which is why I noticed it (if the 64-bit extension is not
present on reconnect, we have a problem - but it's no worse than the
existing problems).  But I'm currently so focused on getting the new
feature interoperating between qemu, libnbd, and nbdkit, plus the US
Thanksgiving break this week, that I'm happy to let you take first
shot at client-side validation that a server reconnect does not lose
capabilities.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org




reply via email to

[Prev in Thread] Current Thread [Next in Thread]