MaxSpotInstanceCountExceeded in GL tests due to GPU

[DavidGOrtega]

We have a difficult scenario having MaxSpotInstanceCountExceeded errors. Probably reuse will solve this

[0x2b3bfa0]

If we enable the --reuse option, unit tests won't [always] cover the runner creation process. Do we want that?

[casperdcl]

@shcheklein found that there are multiple instances running for about a week. We need:

• [ ] automatic warning (email etc.) from AWS if instances run for more than e.g. 30min
• [ ] automatic shutdown of instances (timeout) by AWS
• [ ] figure out why CML didn't cleanly terminate the instances

--reuse will only hide the problem without solving it

[DavidGOrtega]

related to #680

This might be related to #678 after seeying the logs seems that the chrono is not working properly

[casperdcl]

also https://github.com/cloud-custodian/cloud-custodian (@dberenbaum suggestion)

[dacbd]

We haven't seen this in a while, closing for now.