[OpenAFS-devel] Re: [OpenAFS release-team] release-team jabber meeting August the 28th

Jason Edgecombe jason@rampaginggeek.com
Sat, 14 Sep 2013 19:19:08 -0400


On 09/11/2013 10:00 AM, Derek Atkins wrote:
> Jason Edgecombe <jason@rampaginggeek.com> writes:
>
>> On 09/10/2013 10:58 AM, Andrew Deason wrote:
>>> On Tue, 10 Sep 2013 10:43:29 -0400
>>> Derek Atkins <warlord@MIT.EDU> wrote:
>>>
>>>>> A build slave can be behind a NAT but if so it must be administered
>>>>> by someone with NAT access.
>>>> Why?  Does the master need to be able to TCP-Connect() to the slave?
>>> No, the slave connects to the master. You (or someone behind the NAT)
>>> just need to handle upgrading or fixing buildbot itself (or anything
>>> else on the box) if it breaks or needs upgrading or whatever.
>>>
>>>>> The goal for a build slave is that it can perform a git checkout and
>>>>> build the tree in under 20 minutes.   The network/cpu/disk-space
>>>>> requirements are those necessary to achieve that goal.
>>>> Alas, that doesn't give me a good estimate of required network or cpu.
>>>> For disk space I'm guessing the slave doesn't keep many builds around?
>>> No, it doesn't keep any builds after they're done. You just need the
>>> disk space for a single build, the git repo, and maybe a little bit for
>>> logs and buildbot overhead. The latter I believe is very small compared
>>> to any of the other items.
>>>
>> The entire disk partition that contains the buildbot master and slave
>> is 6GB with ~3GB used.
> Well, in that case, I have a VM server that I'm willing to leverage for
> a few Fedora and/or CentOS slaves if you can use them.  I'm happy to
> host and administer the slaves.  (Beyond just a few slaves might tax the
> hardware too much, affecting the other VMs; the host is currently
> memory-bound and extending the RAM is way expensive because it requires
> 8 sticks of DDR2).
>
> -derek
Hi Derek,

We don't have a Fedora build slave, so that would be welcome. Would your 
VM server be able to handle multiple slaves building at the same time? 
In addition to the performance impact, I think that the group was 
divided on whether dissimilar buildslaves would be acceptable.

Here is a guide for setting up a buildslave:
http://openafs-wiki.stanford.edu/BuildbotSlaveHowto/

Here is a puppet module for installing buildbot, but it's only for 
Debian/Ubuntu at the moment:
https://github.com/edgester/puppet-module-buildbot

I'll have to configure any new slaves on the master before that can join 
the build pool.

Let me know if you have any questions or problems.

Jason