Stop or hibernate on-demand instances instead of terminating them.

[cristim]

Github issue

Issue type

• Feature Idea

Summary

Stop or hibernate each on-demand instance instead of terminating it, and then restart/resume it when a spot instance from the same AZ is being terminated.

This way we'll be able to quickly recover in case the spot instances inadvertently caused any sort of downtime, and also hopefully don't lose any capacity when the spot instances are being terminated.

[cristim]

We need to see how to do this in a way that doesn't cause us to leak these hibernated instances

[umitseremet]

During stopping the normal node after creating spot instead of him, which function is being applied? is that a graceful-shutdown or power-off effect?

[cristim]

@umitseremet currently we just use the usual API calls to detach and terminate instances in Autoscaling groups. This is a graceful termination much like when scaling in.

[umitseremet]

Thanks @cristim

[cristim]

@umitseremet no problem.

If you have further questions that are unrelated to any of the existing issues please raise them as new issues or just come and ask them on gitter