[PATCH][WIP] Improve replication restore tombstone reanimation

[Andrew Bartlett]

On Tue, 2016-03-29 at 19:51 +1300, Andrew Bartlett wrote:
> On Sun, 2016-03-27 at 15:49 +1300, Andrew Bartlett wrote:
> > G'Day,
> > 
> > I've been working to make our replication code pass a much nastier
> > set
> > of tests than we have attempted in the past, and in the process
> > have
> > found and fixed a large number of issues related to renames,
> > conflict
> > records and in particular the handling for children of changes to
> > parent objects.
> > 
> > The tests pass against Win2012R2, with the exception of the tests
> > that
> > invoke --local.
> > 
> > The patches are in tombstone-reanimation in
> > git://git.catalyst.net.nz/samba.git
> > 
> > I've also been bold enough to restore the tests for tombstone
> > reanimation.  I was never able to reproduce the failure locally or
> > in
> > the Catalyst Cloud, so I'll need to schedule a time when we can
> > again
> > try this on sn-devel for a large run.
> > 
> > Given the issues fixed by this patch, it is reasonable to try again
> > and
> > get tombstone reanimation into git master.  I can't however say
> > that
> > any specific patch is targeted at it, as I don't know what the root
> > cause of the failure was.
> > 
> > Also included are a fixed autobuild 'last 50 lines' patch, and a
> > dbcheck for a missing deleted objects container (this needs a
> > specific
> > test).
> > 
> > Please comment and review,
> 
> This reliably triggers a replication failure due to incorrect
> attribute
> ID values created by custom schema in 'make test'.  I'm looking
> forward
> to investigating this (much for fun than investigating unreliable
> failures, even if I have to wait for a full make test run...).
> 
> We saw a similar error in some of the flapping dbcheck runs in
> master. 
>  What we need to do is assert on the contents of replPropertyMetaData
> in the ldap and drs schema tests.

With many more fixes, particularly to schema replication, the tombstone
-reanimation branch is back.  As we always suspected, the tombstone
reanimation patches were fine, but the rest of the replication code was
pretty busted.

Replication still has oddities, but this branch just passed 5/6
autobuilds on the Catalyst cloud, with the one failure due to duplicate
random numbers in the schema test (I'll improve those).

In particular, we found the cause of the dbcheck failures where the DN
was a different value to the name attribute.  I understand the failure,
and plan to write a specific test, now that we have the overall test
passing reliably.  This means we can remove dbcheck from flapping
again, which is really important.

Please comment, as we will be looking to merge these soon once we tidy
up some of the top commits, and add a test for the first patch
(dbcheck). 

Andrew Bartlett

-- 
Andrew Bartlett                       http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba