Releasing Samba 4.0 RC1 on 4 Sep 2012?

[Andrew Bartlett]

On Thu, 2012-08-23 at 21:11 +0200, Stefan (metze) Metzmacher wrote:
> Hi Andrew,
> 
> > I agree this is a key feature, as durable handles are a major part of
> > what is really special in SMB3.  
> >
> > Is there anything I can do to assist with this?
> >
> > More broadly, one of the biggest challenges for but also the biggest
> > achievement of the Samba 4.0 release process is bringing our two major
> > development strands together into a single release.  We have much to be
> > proud of!
> 
> Indeed!
> 
> > That our synchronization on a first-suggestion date for RC1 is only off
> > by a few days or a week is pretty impressive, and I'm sure we can work
> > it out.   
> 
> yes :-)
> 
> > Given the feedback earlier, I think it's best that we do the RC before
> > SDC.  That way, we can focus on testing what features we have, not
> > adding the next major feature to this particular release. 
> > 
> >> If possible we should add some magic that rejects to work as AD DC in a
> >> domain
> >> which also has Windows DCs. As that can't work as SYSVOL replication
> >> isn't supported
> >> and we could also corrupt the Windows database if we send replication
> >> notifications
> >> while a windows dc is currently replicating from us. (We could have an
> >> option to
> >> allow this, but that should be off similar to the "dsdb:schema update
> >> allowed"
> >> option.
> > 
> > This is the first I've heard of this.  Can you explain more about the
> > corruption you fear?
> 
> While doing migrations from samba3 to windows via samba4, it was
> needed to disable (at least) the drepl service in samba4.
> 
> What happened is this you add about 5000 computers, 5000 users and 5000
> groups
> and about 10000 groupmemberships to the samba4 ldb
> (after a vampire run, but before starting 'samba')
> 
> Then you start 'samba', as the replication of this amount of objects/links
> takes some time, it's very likely that some local task in 'samba' triggers
> a modification to the domain ldb file. This is picked up by our drepl task
> which sends a notification to the windows dc, which is currently within
> a replication cycle with our drsuapi task.
> 
> This somehow confuses the windows (at least 2008R2) dc, It cancels the
> running
> replication cycle and starts a new one (but with the wrong meta data!)
> All objects/links which were pending in the running replication cycle
> are lost
> and they won't ever be replicated to the windows dc, as he believes that
> they're already replicated.
> 
> I was able to reproduce this and disabling the 'drepl' task fixed it.
> I used samba-tool ldapcmp to compare the servers.

Thanks for the details.  Sorting this out sounds like a perfect task for
the IO Lab with Microsoft.  Do you happen to still have a test script to
make the broken domain?

Did you try and fix the windows domain using drepl commands
(/fullsync?)?

> >>> This is a lot of work for 2 weeks, but I want to avoid doing it in the
> >>> week before SDC as that will be busy with preparation for travel.
> >>
> >> Maybe we should aim for September 7th? Which would mean 3 additional
> >> days for me.
> >> But still I can't promise that it's ready by then.
> > 
> > I'm happy with Sep 7 or Sep 10 (at the latest) if Karolin is happy with
> > me doing the RC1.  (That close to SDC, I would prefer to just handle it
> > with existing processes and scripts personally than also co-ordinate it
> > with someone else). 
> 
> Let's see where we are next week.

Thanks,

Andrew Bartlett

-- 
Andrew Bartlett                                http://samba.org/~abartlet/
Authentication Developer, Samba Team           http://samba.org