CTDB in Samba4

[tridge at samba.org]

Abhi,

Sorry for the slow response. Just got back from a holiday.

 > I'd like to try this out on top of GFS.  What kind of support would be 
 > required from the cluster infrastructure and the cluster filesystems for 
 > CTDB?

Very little filesystem specific support is needed. One of the central
concepts behind CTDB is to avoid using the cluster filesystem for
other than user data. That means all we need is a mountable, shared
filesystem with good data coherence. 

In previous clustering designs for Samba we have relied on extremely
fast (and coherent) clustered byte range locking, but with the new
design we will eventually reach the stage where no locking is done via
the cluster filesystem at all.

The one area where filesystem specific code is likely to be needed is
some way of asking the filesystem what cluster nodes are
present. Ideally we would also have a notifcation mechanism when new
nodes are added. So far we have no hooks for this (we're just using a
flat text file for the list of nodes), but that will change.

 > Can you tell me the timeframe you're looking at to get this
 > integration done?

What we are doing is converting one subsystem at a time to CTDB. I've
only done one so far (the brlock code), and we will need to do at
least 3 more before clustering will be reasonably usable. I'll be
working on these over the next few weeks.

 > Also, are there projects other than CTDB that are currently going on or 
 > in your radar with respect to clustering Samba?

The directory backend code that Andrew Bartlett is working on can be
important for clustering, but the main work is based around CTDB.

Cheers, Tridge