Dead space in sessionid.tdb

[Richard Sharpe]

On Sat, Aug 23, 2014 at 1:15 AM, Volker Lendecke
<Volker.Lendecke at sernet.de> wrote:
>
> On Fri, Aug 22, 2014 at 08:29:46AM -0700, Hemanth Thummala wrote:
> > Hi Volker,
> >
> > We are running 3.6.12 + (critical and security fixes).
> > Have seen few threads and patches related to cleaning up of messages.tdb
> > and locking.tdb files. Here is one we found for cleaning up dead records in
> > messages.tdb file
> >
> > https://lists.samba.org/archive/samba-technical/2014-April/098856.html
> >
> > But this is specific to messages.tdb file cleanup. Should we have similar
> > cleanup routine for sessionid.tdb as well?
> >
> > Also would like to understand the reason for having huge dead record at one
> > point. Any pointers here will be really helpful.
>
> Well, this is for the recovery area inside a transaction.
> tdb_summary counts that as dead. The fact that it is so
> large might just be a bug that Rusty fixed a few years ago
> with 5767224b7f4703c3195aa69eef4352d80980f95e and
> 3a2a755e3380a8f81374009d463cd06161352507. While searching
> through the tdb commit log for these two I came across so
> many fixes and performance improvements in tdb that I'd
> really recommend just dumping the current tdb implementation
> into your tree and use it. Freelist contention reduced,
> freelist fragmentation improved, all sorts of improvements
> all over the place. And it should be completely compatible.
> If you want to really improve things, you might also look
> at the mutex support, which dramatically changes things
> under high load.

They are using FreeBSD. Has the mutex stuff been tested well under FreeBSD?

-- 
Regards,
Richard Sharpe
(何以解憂？唯有杜康。--曹操)