On Mon, Nov 14, 2011 at 5:18 AM, Alexander Sack <asac@linaro.org> wrote:

Problem is that with CI we cannot really hold off submitting jobs because that happens ad-hoc when a build finishes.

Agree, this isn't always easy, but neat to see that infrastructure has a workaround for it already.

I understand the need to have a clean lab pipe. However, for that we need to work on a mechanism that allows you
to put submitted jobs into a queue that isn't processed etc.

Would also be useful to put jobs there during maintenance so the frontend doesn't fail submitting it's job.

There are a few issues to resolve around this. It would have to be a separate piece running somewhere else (like a cloud instance). To make it more reliable, we'd really need to run more than one of these as backups for one another so that when/if we need to reboot it, it could be phased. It would also lack the ability to tell you the job number, since it can't maintain a connection and give you that information, so there would be no direct way to reference the job without coming up with another unique identifier to go by.

A simpler approach in the short term that's been requested already, is the ability to resubmit a failed job. This covers more situations than the one described here.

Thanks,
Paul Larson