[PATCH] Re: irpc/tevent losing replies

Andrew Bartlett abartlet at samba.org
Sat Jan 26 01:31:08 MST 2013


On Fri, 2013-01-25 at 21:34 -0800, Matthieu Patou wrote:
> On 01/25/2013 04:46 PM, Andrew Bartlett wrote:
> > On Thu, 2013-01-24 at 23:16 -0800, Matthieu Patou wrote:
> >> On 01/24/2013 06:22 PM, Andrew Bartlett wrote:
> >>> On Wed, 2013-01-23 at 23:17 -0800, Matthieu Patou wrote:
> >>>> Hello
> >>>>
> >>>> While working on DRS issues with andrew's branch I started to notice
> >>>> message like "IRPC callback failed for DsReplicaSync - NT_STATUS_IO_TIMEOUT"
> >>>>
> >>>> The drepl_replica_sync returns quickly but somehow it seems that either
> >>>> the reply is not sent or it is lost, when I force the dc to be run in
> >>>> single mode (so that all the tasks are in the same process) the error
> >>>> message goes away.
> >>> Per our IRC conversation today, attached is the patch I think we were
> >>> both discussing.
> >>>
> >>> Please review, or consider this my review if your patch is essentially
> >>> identical.
> >> Yeah that's more or less the content of my patch (comments changed), so
> >> I'll review it.
> > In trying to chase down your autobuild failure with patch 4 here, I
> > wrote up these additional patches to ensure our python code uses the
> > correct types.
> >
> > The ldb fix in my other mail needs to be applied first.
> >
> > The combined set now passes autobuild.
> 
> So for patch 2 it's ok, for patch3 I'm not too confident with it I would 
> love to have feedback from Metze or Jelmer, patch 4 is almost ok but I 
> would merge the python part to patch 3 and have just patch 4 for the task.
> 
> Also I think the commit message is not correct because when we are in 
> standard mode I don't think that the single_new_task is used that's 
> where it was tricky: task were created with the prefork model but some 
> of them were using the single_new_connection method for handling new 
> connections.

single_new_task will be used, if a service, forced into single mode
(like the ldap, drepl and rpc servers are) starts a task.  

The prefork model isn't ever used in practice, and the standard modal
won't collide as it always used pid,fd (so cannot have collided with a
pid of '1'). 

I'll fix up the issue with python changes being in the wrong patch, and
return to the list. 

Andrew Bartlett

-- 
Andrew Bartlett                                http://samba.org/~abartlet/
Authentication Developer, Samba Team           http://samba.org




More information about the samba-technical mailing list