Denis Cardon
2015-Feb-25 09:33 UTC
[Samba] performance issue with large group and dbcheck / dc join
Hi everyone, yesterday I had to debug a failing join on a large domain for a group of primary schools with 1000+ desktop and 10k accounts. The domain run mostly ok except for some replications issues, however joining a new DC just takes eons and so does dbcheck. After some investigation into dbcheck (4.1.17) and some "human powered" valgrind like evalution using pudb, I came to the conclusion that it was the computation of the memberof attribute that was to blame. Disabling that attribute check in the dbcheck code did solve the issue. Actually there was a "pupils" group where all new pupils were automaticaly added (and which had in fact no real usage). So that was a group with about 10k members. Deleting the group resolved the dbcheck performance issue and the join at the same time. The dbcheck went for a few hours of run to just 6 minutes. I think that thread was experiencing a similar issue: https://lists.samba.org/archive/samba/2014-July/182819.html Doing such large group is probably not a good idea, be it on a samba4 AD or on a MS AD. But for people with linux scripting background, it is much easier to just add a line in their adduser script that screw up the whole thing. Should I open a bug for that? Cheers, Denis -- Denis Cardon Tranquil IT Systems Les Espaces Jules Verne, b?timent A 12 avenue Jules Verne 44230 Saint S?bastien sur Loire tel : +33 (0) 2.40.97.57.55 http://www.tranquil-it-systems.fr
Jeremy Allison
2015-Feb-25 22:06 UTC
[Samba] performance issue with large group and dbcheck / dc join
On Wed, Feb 25, 2015 at 10:33:57AM +0100, Denis Cardon wrote:> Hi everyone, > > yesterday I had to debug a failing join on a large domain for a > group of primary schools with 1000+ desktop and 10k accounts. The > domain run mostly ok except for some replications issues, however > joining a new DC just takes eons and so does dbcheck. > > After some investigation into dbcheck (4.1.17) and some "human > powered" valgrind like evalution using pudb, I came to the > conclusion that it was the computation of the memberof attribute > that was to blame. Disabling that attribute check in the dbcheck > code did solve the issue. > > Actually there was a "pupils" group where all new pupils were > automaticaly added (and which had in fact no real usage). So that > was a group with about 10k members. Deleting the group resolved the > dbcheck performance issue and the join at the same time. The dbcheck > went for a few hours of run to just 6 minutes. I think that thread > was experiencing a similar issue: > https://lists.samba.org/archive/samba/2014-July/182819.html > > Doing such large group is probably not a good idea, be it on a > samba4 AD or on a MS AD. But for people with linux scripting > background, it is much easier to just add a line in their adduser > script that screw up the whole thing. > > Should I open a bug for that?Yes, if it's causing a domain join fail then we need a bug to track it. Thanks !