new Kerberos tests in Samba3

[Andrew Bartlett]

On Wed, 2011-04-06 at 10:16 +0200, Stefan (metze) Metzmacher wrote:
> Am 06.04.2011 04:40, schrieb Andrew Bartlett:
> > On Wed, 2011-04-06 at 06:45 +1000, Andrew Bartlett wrote:
> >> On Tue, 2011-04-05 at 16:31 +0200, Guenther Deschner wrote:
> >>> Hi Andrew,
> >>>
> >>> On 04/04/2011 01:32 PM, Andrew Bartlett wrote:
> >>>> The branch, master has been updated
> >>>>         via  a3ef974 s3-rpc_server Remove comment, yes the key is correct.
> >>>>         via  77e6716 s3-auth consolidate create_local_token() into make_server_info_krb5()
> >>>>         via  841d0bc s3-selftest Remove more instances of /tmp in test_smbclient_s3.sh
> >>>>         via  6351dee s3-selftest Add testing of kerberos login
> >>>>         via  55134c9 s4-credentials Add a command line hook to set the kerberos credentials cache
> >>>>         via  ffb6003 s3-selftest Disable log rotation in 'make test'
> >>>>        from  513574a talloc - some documentation changes
> >>>>
> >>>> http://gitweb.samba.org/?p=samba.git;a=shortlog;h=master
> >>>>
> >>>>
> >>>
> >>> Wow, that is great stuff! Finally we have some security=ads testing on 
> >>> the buildfarm! the only problem is that the ktest tests fail quite 
> >>> often, I tried to nail down the problem using
> >>>
> >>> SMBD_OPTIONS=-d10 make test TESTS=lookupsids FAIL_IMMEDIATELY=1 VALGRIND=1
> >>>
> >>> but the failure seem not to happen when running under valgrind, do you 
> >>> have any other idea ? If we don't catch it, we should mark the kerberos 
> >>> tests knownfail for now probably.
> >>
> >> Can you give me what details you have of the failure?
> >>
> >> I ran it here often, and got it past autobuild, so I'll need some
> >> clues.  
> >>
> >> There is only one thing I came across in my testing: a failure to
> >> contact a newly running server in the test environment.  This change
> >> requires that it contact the server by name, not IP, and so if nmbd
> >> isn't fast enough, or smbd hasn't really started in time, then we could
> >> have problems.  
> >>
> >> Perhaps that's the issue?  Samba4 puts some 'nmblookup' calls in the
> >> server starting code to ensure things are up and going in it's selftest.
> > 
> > I don't fully understand why netbios name lookups are not reliable in
> > make test, but I've pushed a change to autobuild that uses the same
> > 'fake dns host file' that Samba4 uses.  This should avoid the lookups
> > and make this reliable.
> > 
> > I've also found and fixed a bug where the previous machine trust account
> > password was not honoured. 
> > 
> 
> It seems it's still flakey, see
> http://git.samba.org/autobuild.flakey/2011-04-06-0824/samba3.stdout

Yeah, I saw that.  This one is annoying, as it did actually go away in
my test - but that's exactly what a flaky test is I suppose. 

I'll dig into this again tomorrow.  I suspect the fake DNS code isn't
doing so well with just a short name, but this has supprised me enough
times that I can't be totally sure. 

The fact that in make test NetBIOS name lookups are not reliable or that
nmbd can't start honouring name lookups after 10 secs is also worrying. 

Andrew Bartlett

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