process shared robust mutexes for tdb

[Volker Lendecke]

On Mon, Mar 25, 2013 at 08:36:39AM -0400, Ira Cooper wrote:
> On Fri, Mar 22, 2013 at 12:10 PM, Volker Lendecke <Volker.Lendecke at sernet.de
> > wrote:
> 
> > On Sat, Jan 12, 2013 at 10:45:57AM +0100, Volker Lendecke wrote:
> > > Hi!
> > >
> > > On Sun, Jan 06, 2013 at 07:18:00PM -0500, Ira Cooper wrote:
> > > > Can you run the same test with 1 bucket?  Not that we'd actually want
> > to
> > > > deploy that way, but, the performance of fcntl TDB shouldn't drop that
> > > > much... and it'd give us an estimate of how much we need the 2048
> > locks vs.
> > > > just using 1.
> > >
> > > Under the usual branch
> > >
> > > https://git.samba.org/?p=vl/samba.git/.git;a=shortlog;h=refs/heads/tdb
> > >
> > > Now find a patchset that survived autobuild for me. It
> > > contains some additional stuff, like a tdb feature flags
> > > field. This could for example take a flag for a dual-linked
> > > freelist and other future improvements.
> > >
> > > This is something that I would like to see broader testing
> > > and review of. Getting rid of the fcntl locks in tdb was
> > > something I have wanted to do for a very long time, and
> > > while this is not complete it is a start in that direction.
> >
> > The above branch now contains code that has survived testing
> > in a cluster with thousands of clients. The cluster survived
> > a recovery just fine, something which was entirely
> > impossible due to the fcntl thundering herd before. The fact
> > that unmodified ctdb is happy with it shows that the API is
> > covered pretty completely.
> >
> > I squashed all the patches into one. In the end, I had
> > almost 100 unsorted patches, no way to sort them into a sane
> > order.
> >
> > Comments?
> >
> 
> Looks good from a "review" point of view.  I have some work to do on the
> "OS" side to support this patch, then I'd like to test it in our lab,
> before giving the "all clear."
> 
> Are you looking to have these patches pushed "as is", or looking for more
> feedback before a final round?

As metze wrote, the tdb piece to me is pretty complete, the
libreplace patches need work. The tdb code has seen some
heavy cluster testing within 3.6 and ctdb from the 1.2.40
branch, but only on Linux. Testing on Solaris would be very
much appreciated. It's not very much code, but it is so
central that I am very cautious about it.

Volker

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen
http://www.sernet.de, mailto:kontakt at sernet.de