Re: [Qemu-devel] [PATCH v7 00/23] Migration: postcopy failure recovery

[Peter Xu <peterx@redhat.com>]

On Thu, Mar 15, 2018 at 04:19:37PM +0800, Fam Zheng wrote:
> On Mon, 03/12 12:45, Peter Xu wrote:
> > On Fri, Mar 09, 2018 at 10:59:03AM -0600, Eric Blake wrote:
> > > On 03/09/2018 10:54 AM, Eric Blake wrote:
> > > > On 03/09/2018 03:15 AM, Peter Xu wrote:
> > > > > Based-on: <20180309090006.10018-1-peterx@redhat.com>
> > > > > 
> > > > > This series is based on the series:
> > > > > 
> > > > >    [PATCH v8 00/23] QMP: out-of-band (OOB) execution support
> > > > 
> > > > Let's spell that in a way patchew understands:
> > > > 
> > > > Based-on: <20180309090006.10018-1-peterx@redhat.com>
> > > > 
> > > 
> > > Never mind - I missed that you snuck that in on the very first line (which
> > > I'm used to overlooking because it's often a 'From:' author attribution,
> > > instead of a patchew directive).  I saw the patchew failure messages, but
> > > didn't read them closely to see why it is failing, and merely assumed it was
> > > because of the missing dependency...
> > 
> > Yeh, but there still seems to be something wrong elsewhere since the
> > series can pass compilation and docker tests on my host...
> > 
> > Fam, what would happen if I firstly post OOB series, then postcopy
> > series (which will have "based-on" on OOB series) right away?  Could
> > Patchew just ignore the "based-on" if I post the 2nd series too fast
> > since Patchew may have not yet went through the first one?
> 
> Could be. Git apply of the first one races with the appearance of the second one
> in patchew's database. Or if the emails arrive in patchew's INBOX out-of-order,
> the based-on tag will not work as expected.

Thanks for confirming.  Based-on is not frequently used, so I think
what I need to do next time is I manually query and only send the 2nd
series after patchew.org reported okay on the first one.

It'll be good too if patchew can stop running tests if there is
unknown based-on tag (either fully unknown, or work-in-progress), and
report that "unknown tag found" message directly instead of test
failures (since if there is a based-on, it's very possible that it'll
fail).

Further - I'm not sure whether serializing the work will be hard or
not. Say, if series B has a based-on of A that is known but during
processing, then postpone processing B until A finished.  I won't ask
for that though especially if it's not trivial.

Thanks,

-- 
Peter Xu