autobuild failure due to python replication tests - why ?

[Jeremy Allison]

On Thu, Aug 04, 2016 at 11:48:23AM +1200, Andrew Bartlett wrote:
> On Wed, 2016-08-03 at 15:46 +0200, Stefan Metzmacher wrote:
> > Hi Andrew,
> > 
> > > On a positive note, I appear to have an env and cut-down test that
> > > runs
> > > in 30min that consistently fails.  I have early positive vibes for
> > > this
> > > set of patches:
> > > 
> > > http://git.catalyst.net.nz/gw?p=samba.git;a=shortlog;h=refs/heads/v
> > > 4-5-test-drs-fix
> > 
> > http://git.catalyst.net.nz/gw?p=samba.git;a=commitdiff;h=77b85f3ccd36
> > ba54c58a3d9347c05a6e768fddd9
> > Can you catch the attid mismatch for linked attributes in
> > dsdb_replicated_objects_convert() please?
> > 
> > I think that's also that place where we should convert the remote
> > attids to
> > the local once.
> 
> I understand the principle.  Currently, the server-supplied data
> structures are simply attached to the end of the transaction. 
> 
> If you don't mind, I want to use those patches as-is, because that is
> what I am gathering some data to prove, that is I don't want to further
> rework the DRS code until we have this stable.
> 
> Sadly the data isn't looking good, so we may be back to square one. 
> 
> > > I'll know more when I see the full set of builds back, and then try
> > > again just to be sure.  I'll of course file bugs for back-porting
> > > per
> > > my last mail, unless you would like to take that honour :-)
> > 
> > I still don't understand how this could explain the timeouts.
> 
> These are not timeouts, but un-called callbacks.  The callback to send
> the IRPC is never fired, as far as I can tell.  That then appears to be
> a timeout to the caller. 

FYI, after our conversation yesterday I'm still trying to track
this down haven't found any missing callback yet. My guess is
that a completion function is never finishing.