When some tests are known to be slow (minutes of runtime) and get started late in a run which has a tight deadline, then we can end up with those slow tests not finishing in time. Even with such a deadline it's less efficient to run them late because the overall runtime is higher (lot's of activity initially, then just waiting for a few slow tests at the end).

Both can be avoided by prioritizing the slow jobs such that they get started early. Then their execution overlaps with other tests and the run can finish sooner.

The tentative proposal on Slack was to annotate nodes with a priority score (= number) and sort them based on that score. In Kubernetes, we can extend our WithSlow helper to set that score to a non-default value which ensures that such tests get run early. @onsi wanted to work on this. This issue is just for tracking purposes.