samba_3 and the build farm

Jeremy Allison jra at samba.org
Thu Sep 7 12:19:36 GMT 2006


On Thu, Sep 07, 2006 at 03:23:12PM +1000, tridge at samba.org wrote:
> Jerry and Volker,
> 
> I had to disable the test phase of samba_3_0 in the build farm as it
> is leaving stray smbd processes around. From the SVN tag in the
> binaries the problem seems to have started at around rev 17797 (may
> not be very precise).
> 
> Unfortunately I don't have gdb on the machine I noticed this on yet,
> but strace and ps shows that its leaving behind two smbd processes per
> run, a parent and a child. The parent is sitting in select() like
> this:
> 
>    select(23, [20 21 22], NULL, NULL, NULL      
> 
> the child is sitting in pause().
> 
> I don't understand why the timelimit code isn't kicking in. It should
> be killing these stray processes, but it isn't. Perhaps something to
> do with the architecture? It's a PPC64 (Volker, its tridge.anu.edu.au,
> which I think you have an account on).
> 
> I also don't know how many different hosts are affected. Hopefully we
> can narrow this down to just a few hosts or fix the bug soon and get
> samba_3_0 testing again.

This might be my recent change to lib/events.c - just realised...

I'll look at this as soon as I actually wake up (this is a
middle of the night "oh dear, this might have caused it" email :-).

I think the change is correct - I need to check all the
called event functions to make sure they correctly delete
themselves (if they don't it might cause such a spin).

Jeremy.


More information about the samba-technical mailing list