winbind: Remove wbint_QueryUserList? (was: Re: Trying winbind flapping tests on the Catalyst cloud)

[Jeremy Allison]

On Tue, Jan 10, 2017 at 12:33:47PM +1300, Andrew Bartlett wrote:
> On Tue, 2017-01-10 at 08:17 +1300, Andrew Bartlett wrote:
> > On Mon, 2017-01-09 at 11:09 -0800, Jeremy Allison wrote:
> > > On Tue, Jan 10, 2017 at 07:31:24AM +1300, Andrew Bartlett wrote:
> > > > On Mon, 2017-01-09 at 20:57 +1300, Andrew Bartlett wrote:
> > > > 
> > > > > I'm running an autobuild on each commit, with a mask for the
> > > > > ntlm_auth
> > > > > issues, on the Catalyst cloud.  We were seeing it pretty
> > > > > reliably. 
> > > > > 
> > > > > While one per commit might not be enough, I hope in aggregate
> > > > > the
> > > > > 25
> > > > > builds might tell something.  It is just as likely that I'll
> > > > > get
> > > > > 25
> > > > > unrelated failures, but I'll let you know if any kind of
> > > > > meaningful
> > > > > pattern emerges.
> > > > 
> > > > A pattern certainly does emerge.  It appears quite reasonably
> > > > that
> > > > the
> > > > errors come from this patch or one nearby:
> > > 
> > > Well it's got to be one nearby, as that specific patch
> > > just removes dead code no longer called.
> > > 
> > > The query user list patchsets are:
> > > 
> > > 81e5770aeebccfe6c65a40a5ac0e9e3a7b4c5d60
> > > 479ce28fd7dd54a6ae76fbbe3cd0a870738d87c0
> > > 67c0696761dedb748b1e4dc02531acbbf5ff95ca
> > > 480c9581a13afc08b20e80d2ff8a45ac8d7f18d3
> > > 
> > > so can you concentrate on looking at these ?
> > 
> > I'll fire off four builds on each of these and see if I can lock this
> > down to a particular commit.  In terms of engineering I'll focus on
> > the
> > other flapping tests that have come up, to try and reduce the noise
> > closer to my area of expertise, but I hope this data is helpful to
> > the
> > others also looking at it!
> 
> I'm also testing a revert up to and including "winbind: Remove
> wbint_QueryUserList" (67c0696761dedb748b1e4dc02531acbbf5ff95ca).  I
> know this is 'just' removing code, but it is the earliest patch I've
> had fail in this way.
> 
> So far I've not seen 479ce28fd7dd54a6ae76fbbe3cd0a870738d87c0 fail with
> this error code, so there may be something quite fishy going on.

Yeah, just to confirm what we chatted about on the phone, at
this point a debug level 10 log client and server side from
the wbinfo -u that works and fails is what we need now. And
depending on what that shows we might need valgrinds :-).