Re: Replication failure after error fixed

[Quanah Gibson-Mount <quanah@stanford.edu>]

--On Friday, April 20, 2007 11:55 AM +1000 Dave Horsfall <daveh@ci.com.au> 
wrote:

> OpenLDAP 2.3.32 (our policy is to run STABLE unless there's a bugfix we
> need).
>
> Most of our sites replicate direct to each other (SyncRepl; you need to
> know that data for a country is mastered in that country), except for one
> situation:
>
> A <-> B <-> C
>
> A and C are masters for their data, and B is a pure slave.  For political
> reasons (i.e. it won't get fixed) A and C cannot replicate direct.
>
> Because a schema change was not made on B, some updated data on A did not
> get through.  All well and good, we fix the schema on B, and wait for the
> update (we use refreshAndPersist).
>
> Except it never happened.  Blowing away the slave on B caused it to
> update (of course), except it still never reached C, until it in turn
> was repopulated.
>
> Am I looking at a replication bug?  It seems to me that once the schema
> was fixed, the replication should have happened.  Or am I not
> understanding how SyncRepl works?

This sounds strikingly similar to a bug I've encountered in the past with 
delta-syncrepl where the CSN was incorrectly updated after a failed MOD 
(due to differences because the replicas had an overlay on that the master 
didn't).  I've had it on my to-do to really get the logs for this, but have 
been busy on other things.  I'll see if I can set some time aside to 
re-produce this and get the necessary information so it can be fixed.

--Quanah

--
Quanah Gibson-Mount
Senior Systems Software Developer
ITS/Shared Application Services
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html