Tools » teuthology

This was inspired by #10112. It was raised that refusing to schedule entire runs because one of the required distros is missing packages is not ideal.

Initially I thought it might be good enough to schedule those jobs whose packages were already present, and email a warning for the others. Yuri then added that it would be nice to have the doomed jobs still present in pulpito, so that we could see what we were missing. Of course, since we still use beanstalkd and rely on it for job_id generation, that wouldn't work out so well.

But what if, instead of dropping the jobs we know will fail, we schedule them anyway - and then make them fail far earlier than they do currently?

It would mean adding similar checks to, most likely, task.internal - probably even using the functions in teuthology.suite.