[Samba] samba getting stuck, highwatermark replication issue?

[Andrew Bartlett]

On Thu, 2017-10-12 at 09:17 +0200, mj via samba wrote:
> Hi all, James,
> 
> After following James' suggestions fixing the several dbcheck errors, 
> and having observed things for a few days, I'd like to update this 
> issue, and hope for some new input again. :-)
> 
> Summary: three DCs, all three running Version 
> 4.5.10-SerNet-Debian-16.wheezy, samba-tool dbcheck --cross-ncs reports 
> no errors, except for two (supposedly innocent) dangling forward links 
> that I'm ignoring for now. Time is synced. Very basic smb.conf, posted 
> earlier, can post again if needed.
> 
> samba-tool ldapcmp dcX dcY --filter=whenChanged shows that they are in 
> sync, and also samba-tool drs showrepl shows that replication seems to 
> be stable.
> 
> The "getting stuck" from the subject line has not occured for a few 
> days, perhaps the dbcheck fixes have solved that, or perhaps we've just 
> been lucky.
> 
> All in all this appears pretty healthy, but there is a remaing problem:
> 
> At ANY given time, ONE RANDOM single DC shows high cpu usage on one 
> samba process. And on that DC (can be any of the three DCs) the logs 
> fill up with this:

I would upgrade to Samba 4.7.  The work on locking in LDB and the
mention of replication issues was serious.  Likewise we fixed a number
of other issues in over-replication of linked attributes (group
memberships) for 4.6.

We are carefully following the reports here, but we do expect
replication should be much more stable with Samba 4.7.

Thanks,

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