The primary requirements for adding a builder to the list of builders for a given branch are:
1. it must build the branch without errors so as avoid introducing
unrelated build failures on unrelated patch submissions
2. the builder must be fast enough to keep up with the other
builders from a throughput perspective.
Adding multiple builders can increase overall throughput at the cost of
increased latency across all builders. This is a result of the
increased network load resulting from multiple builders fetching the
repo in response to a build notification.
If multiple builders for the same platform are going to be added, it is
very important that the machines that are hosting them be identically
configured. The opportunity for a patchset to pass or fail should not
be the result of which builder it randomly assigned to.
Jeffrey Altman
On 8/16/2013 1:05 PM, Jason Edgecombe wrote:
> As I recall, the main requirement for a gerrit-triggered buildslave is
> that the build must complete within a certain time, say 10-15 minutes.
>
> If we had multiple RHEL5 buildslaves and could increase throughput,
> would that be acceptable for being triggered by gerrit? Is latency the
> deciding factor?
>
> I'm concerned about RHEL5 being neglected because it doesn't appear in
> gerrit.
>
> Is there much advantage to having multiple buildslaves if the RHEL5
> builder isn't triggered by Gerrit.
>
> I'm open to other options to reduce bit rot on platforms without a
> gerrit-triggered build.
>
> Thanks,
> Jason
> _______________________________________________
> OpenAFS-devel mailing list
> [email protected]
> https://lists.openafs.org/mailman/listinfo/openafs-devel
smime.p7s
Description: S/MIME Cryptographic Signature
