CTDB in master?

Michael Adam obnox at samba.org
Tue Oct 15 06:11:45 MDT 2013


sorry for the delay in following up, I was offline last week...

On 2013-10-07 at 10:16 +1300, Andrew Bartlett wrote:
> G'Day,
> While we were at the Microsoft IO Lab we discussed what actually needed
> to be done to get CTDB in the master tree.
> Why am I suddenly interested?  Well I really like the idea of automated
> testing, and I would like to learn CTDB a little.  I've offered to help
> because I've got a good grasp of autobuild, and have experience adding
> new dependent environments to our selftest system.  The goal as I
> understand it is to have a new environment, dependent on 3 or 4 CTDB
> nodes, that presents a SERVER_URLLIST (or so) in the format that
> smbtorture wants for the list of multiple servers.
> I understand metze has a branch with the ctdb history imported, and
> obnox already has much of the autobuild work done.  We could import the
> build with autoconf for now if that helps things move faster, but
> naturally we would want to move to waf to keep just one system. 
> With Samba 4.1 releasing soon, and our desire to have this done for
> Samba 4.2, we really should take the opportunity of having many of the
> parts in place, and I'm happy to help out for my part.
> Metze/Obnox: 
> Is this the current tree that we would import?
> https://gitweb.samba.org/?p=obnox/samba/samba-obnox.git;a=shortlog;h=refs/heads/master-ctdb

Right, this is my branch of work on integrating ctdb with our
autobuild and selftest that was on top of Metze's snapshot
import of ctdb master. Metze's ctdb-import would have to be
redone at the point in time we really switch.

Concering my work:

- I have not switched the ctdb to waf or anything.
  It is built as before with autoconf based make.

- The patch that adds a separate ctdb target
  can be considered complete. It does not run the
  ctdb internal "make test", but could be extended to
  do so, but I think this would be for a later time.

- The patch "TODO:REMOVE: autobuild: temporarily remove
  random-sleep for samba-ctdb target" is not to be pushed
  upstream: I only added it to be able to test the
  "samba-ctdb" target without delay.

- The patch "TODO: autobuild: extend samba-ctdb target to build
  ctdb also" could be considered complete for a start:
  It creates a target with ctdb compiled against the tdb
  under lib/tdb (not the one under ctdb/lib/tdb), installs
  this ctdb, and then compiles samba against this ctdb.

> It seems almost finished:

I think this patchset would already be reasonable for a start
and is indeed rather finished.

> What can I do to help?  Is there any
> particular part of our selftest system that is blocking you here?

The next patch is really unfinished, and this is where
of course help would be appreciated: It adds a clusteredmember
environment to our selftest, for a start a 1-node cluster.
This patch has not been run yet. It is the start (done at the
IO-Lab) of my effort of continuing my work from 2 years ago,
when I first started this:


This is the branch where this sort of ran alrady.
(I only squashed and reordered my commits from the original branch


There you can see how I tried to integrate this into our "make
test" to run various smbtorture tests against the cluster to
excercise the samba->ctdb codepath.

The main point blocking me at the moment is ressources.
Andrew, your help would be appreciated: if you could look
into the unfinished patch and check whether this looks
like a vaguely reasonable approach to you, we could continue
into this direction.

This could also be completed after integrating the ctdb code,
just as subsystem reconsilation can be done afterwards.

Cheers - Michael

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 215 bytes
Desc: Digital signature
URL: <http://lists.samba.org/pipermail/samba-technical/attachments/20131015/d462951c/attachment.pgp>

More information about the samba-technical mailing list