Tools » teuthology

The current use case for --upload is to run a suite and upload the results when it completes. It turns out this is not what we really need. Instead we want to run a suite over and over again until we figure out what's wrong and during these iterations the output is mostly noise. We don't know before running a suite if that's what we will want to upload. We know that only afterwards and most of the time a single suite is not enough.

We want to upload a proof of success or a proof of failure, for the record or for a reviewer to see. For the proof of success the most common case is that the aggregation of a series of run makes for a successful run for a given ceph sh1. For a proof of failure, it is the aggregation of a set of jobs whose failure is considered meaningful.