Tools » teuthology

there's a lot of teardown time just wasted on tests that run on a cluster of VMs that will be unlocked/destroyed anyway. IWBNI there were a way to avoid this wasted work.

Some ideas:

1) if the cluster is all VMs, just don't call exit() by default
1a) maybe call it if the task defines an attribute/decorates in a way that says it wants to

2) allow tasks to decide if they should do work in exit by passing an "all VMs" flag, maybe, or make sure it's easy for them to ask. This could combine with 1a) above, so that most tasks don't do anything, but of the few that do, they could avoid the bulk of the 'reset' work (and maybe just do the "finish timing and log" type work)

Perhaps there are other approaches. This could reduce cycle time for VM tests by a lot.