Samba 4 testsuite analysis

[Matthias Dieter Wallnöfer]

Well, okay - for the first two tests (pwdLastSet) the problem lies in 
the delay:

(torture/rpc/samr.c)
>        if (torture_setting_bool(tctx, "samba3", false) ||
>             torture_setting_bool(tctx, "samba4", false)) {
>                 delay = 999999;
>                 torture_comment(tctx, "Samba3 has second granularity, 
> setting delay to: %d\n",
>                         delay);
>         }
Probably this can be somewhat decremented. I don't exactly remember who 
introduced that delay (gd - was it you?) - I've only adopted it for s4 
as well - since with the default one it didn't work at all.

Matthias


Jelmer Vernooij wrote:
> Hi Matthias,
>
> On Thu, 2010-09-23 at 10:19 +0200, Matthias Dieter Wallnöfer wrote:
>    
>> nice that you share my opinion - which I've stated already. We really
>> need to do some performance improvements on "make test".
>> Have I understood it correctly that part of the problem lies in the
>> "fsync"s in the LDB transaction code? Probably we could add some bypass
>> code for them when run under "make test".
>>      
> I'm not sure yet if that will help. This analysis only tells us which
> tests are slow, not why they are slow.
>
> Cheers,
>
> Jelmer
>
>    
>> Jelmer Vernooij wrote:
>>      
>>> The Samba 4 test suite is taking increasingly long to run - on my
>>> machine it takes almost an hour.
>>>
>>> Now that we've got subunit in place I've done some analysis to see what
>>> tests exactly is causing the test suite to take so long, and hopefully
>>> what we can do about it.
>>>
>>> The full list of (6000) tests sorted by run time can be found at:
>>>
>>> http://samba.org/~jelmer/samba4-test-times.ods
>>>
>>> This document was created by running "subunit-ls --times
>>> --no-passthrough" on st/subunit and sorting the resulting data.
>>>
>>> Hopefully we can improve the run time of some of the slowest tests
>>> and/or split them up a bit further.
>>>
>>> Cheers,
>>>
>>> Jelmer
>>>
>>>        
>>      
>