Tools » teuthology

Error handling mechanisms tend to work "most of the time", and the best way to find their limitations is to subject systems to continuous failures (with many thousands of times the frequency that real systems would experience). This tends to very effectively shake out problems.

We can start out with simple rotations of monitor and osd failures, noting recovery (with health commands), recording read and write latencies, and griping when they are too long.
(we'll get much fancier later)