running wintest issues

[Sean Dague]

On 10/05/2011 03:56 AM, Andrew Bartlett wrote:
> On Wed, 2011-10-05 at 17:49 +1100, tridge at samba.org wrote:
>> Hi Sean,
>>
>>   >  It appears to just hang, waiting for something to happen. I'll do
>>   >  another run today to get more details.
>>
>> you may want to look in the c:\windows\debug directory and see if
>> dcpromo gives any useful output there
>>
>>   >  I'm wondering if long term it would make more sense to put a little
>>   >  python xml-rpc daemon in the windows guests to send the command stream
>>   >  in, instead of using the telnet / expect interface.
>>
>> I quite like the telnet approach, because it works on all windows
>> versions without having to worry about additional packages. We were
>> trying to keep the windows install very simple, and I think it's
>> better that we keep that simplicity if possible.
>
> The key word here is 'works'.  It certainly is the most sensitive/flaky
> part of wintest, and the more we can do to get that step automated (can
> use other tools to get administrator into the right groups and telnet
> started before we start telnet?) the better off we will be.

Works is a very relative term, it only seems to work in *very* specific 
windows configurations. I've had the telnet approach fail in 2 ways 
during setup that I didn't expect.

1) In Windows 7 Professional, it only works if you enable the disabled 
built in Administrator account and give it a password. Any other user 
that is in Administrators group (including the user you are prompted to 
create on install) is not allowed to log in via telnet (that was about 2 
days worth of discovery).

2) It only works in Windows 2008 if you've not changed the Shell to 
power shell (and only use cmd.exe). That took a bit to discover.

It is also sensitive to telnet settings on the client. If those have 
been changed because of working with other systems that needed different 
line modes, it may no longer work.

The point being that it's very fiddly to a lot of the environment that 
isn't checked for. This fiddliness I'm sure contributes to it not being 
run regularly by many folks.

If we want to stay with the telnet based approach, there at least needs 
to be *a lot* of extra environment checks to ensure the systems are 
ready to be tested, as they need to be in a very specific state before 
the tests can be run and mean anything.

	-Sean

-- 
Sean Dague
IBM Linux Technology Center
email: sldague at us.ibm.com
alt-email: sdague at linux.vnet.ibm.com