[OpenAFS-devel] Re: adding "make check" to build slaves

[Jason Edgecombe]

On 04/09/2012 07:34 PM, Steven Jenkins wrote:
> On Mon, Apr 9, 2012 at 7:27 PM, Jeffrey Altman
> <jaltman@secure-endpoints.com>  wrote:
>>
>> On Monday, April 09, 2012 7:13:13 PM, Andrew Deason wrote:
>>
>>> Or at least make them not fail the build. 'make check || true' perhaps
>>> would not cause verification to fail, but we could still see what
>>> happens when it runs.
>>
>> Who is going to look at the output?
>>
>> In the last two weeks we have twice hit the point where the slowest
>> builders are so far behind the fastest builders that by the time the
>> build reports finally get submitted to the coordinator none of the
>> build logs are still available for the fastest builders.
>>
>> If the builders are not going to fail the build for a test failure I do
>> not see the point of running them for every build as part of patchset
>> verification.
>>
>> In my opinion what would be reasonable at this point would be
>> to have a separate set of builders that run 'make check' on a daily
>> basis.  However, I do not want them in the current builder pool
>> because failures that result in deadlock will bring the entire buildbot
>>   verification system to a halt.
>>
>>
> I think that's a very good suggestion.
>
> I don't know what the slowest buildbot systems are, but I do have some
> good Intel hardware if that would help (sorry, no IRIX/MIPS AIX, or
> Sparc).  Someone ping me directly, and I'll be happy to get that into
> the pool.  And if a few systems would be useful to run make check on,
> I'm happy to have those in a separate pool.
>

Email me privately about contributing a build slave.

I've removed the opensuse-i386 builder from gerrit. It was the slowest 
builder at 40 minutes. The next slowest is ~25 minutes. That should 
speed things up.

At the moment, the debian-i386 and freebsd82-amd64 builders are running 
"make check" with warnings instead of failing.

We already have multiple pools of builders, which aren't mutually 
exclusive. We have two main pools: the gerrit-triggered pool 
(debian-i386, freebsd82-amd64,opensuse-x86_64, win2008 builders), and 
the scheduled builders (the rest). Some branches only build on certain 
builders.

We can split the builders and slaves however we like. The main issue is 
that it would be nice to run "make check" on as many platforms as 
possible, but we need more slaves, possibly double, if we want the same 
platform coverage and a completely isolated testing pool.

Jason