[SCM] Samba Shared Repository - branch master updated

[Matthieu Patou]

On 05/21/2012 01:23 AM, Volker Lendecke wrote:
> On Mon, May 21, 2012 at 01:17:24AM -0700, Matthieu Patou wrote:
>> On 05/21/2012 01:03 AM, Volker Lendecke wrote:
>>> On Fri, May 18, 2012 at 10:09:56AM -0700, Jeremy Allison wrote:
>>>> On Fri, May 18, 2012 at 01:13:03PM +0200, Volker Lendecke wrote:
>>>>> The branch, master has been updated
>>>>>         via  f5ca3f1 s3: Revert the serverid changes, they need more work
>>>>>        from  45082a8 s4-torture: Improve torture test boilerplate, use torture_assert()
>>>>>
>>>>> http://gitweb.samba.org/?p=samba.git;a=shortlog;h=master
>>>>>
>>>>>
>>>>> - Log -----------------------------------------------------------------
>>>>> commit f5ca3f11e47e48d195616f813b5b2c9e8255c6cd
>>>>> Author: Volker Lendecke<vl at samba.org>
>>>>> Date:   Fri May 18 09:10:02 2012 +0200
>>>>>
>>>>>      s3: Revert the serverid changes, they need more work
>>>> PLEASE DO NOT DO THIS WITHOUT CONSULTATION WITH OTHERS.
>>>> THIS IS NOT YOUR SANDPIT.
>>> To avoid this kind of thread in the future, maybe we should
>>> change autobuild so that it has to run 5 times in a row
>>> successfully before the push is made. Right now we are
>>> somewhere between 1:30h and 2:00h for a push to get through.
>>> This is completely detached from a normal workflow.
>>> Extending that to, say, 8 to 10 hours should not make a
>>> significant difference anymore.
>> -1
> Why? This would avoid a lot of the flaky tests we have seen.
Well as Luk pointed it, you propose a technical solution for solving 
communication problem. I didn't follow quite closely this thread to 
understand how it has happened but ihmo the conclusion are we have to 
better communicate.
You could say "ok, we'll try to better communicate but let's add this 5 
time in a row as a safety net". I see a couple of important issues:

* You can't rule out that even with 5 runs you won't have a intermittent 
failure and if the intermittent failure prevent my patches to go in the 
tree  I'll be pretty upset and I'm sure I won't be the only one
* My understanding of autobuild is that somehow build are serialized, 
that is to say that if you push a patchset to autobuild and 10 minutes 
after I do another push our two builds will run in parallel but if your 
build succeed then mine will be canceled, my patch set will be rebased 
on top of your and new build will be kicked creating a kind of implicit 
serialization. It means that on "busy days" (like last tuesday) it can 
take more than 3 days to see my patch in the tree.

Matthieu.


-- 
Matthieu Patou
Samba Team
http://samba.org