[PATCH] avoid lock oder violation between xattr.tdb and g_lock.tdb

[Volker Lendecke]

On Tue, Jul 12, 2016 at 05:12:48PM +0200, Michael Adam wrote:
> Ouch. So without the below-mentioned rework of dbwrap_watch
> (or another solution), we would need an additional lock oder
> value(4)?

Yep.

> > Just by coincidence, I've started working on an alternative
> > dbwrap_watch implementation that can live without dbwrap_watchers.tdb.
> > Maybe you want to give my dbwrap_watch branch a quick peek. This is
> > work in progress, but I think this might be a way forward.
> 
> Will do, thanks!

The main reason why I'm working on that is speed in the cluster. A
customer of mine has found a measurable performance regression when
using dbwrap_watchers.tdb was introduced for locking.tdb. So this
might be worthwhile to look at for more than one reason now :-)

Volker