Shri Javadekar
Shri Javadekar
Can you check if the LaunchConfig/LaunchTemplate created by the instance-manager? Does it show the spot price being set?
There was some code added to handle this. However, there will always be a "time-of-check-to-time-of-use" window. See https://github.com/keikoproj/upgrade-manager/blob/40b08738c104caa50dcd0f8b82588a07e45f7446/controllers/rollingupgrade_controller.go#L544 and https://github.com/keikoproj/upgrade-manager/blob/40b08738c104caa50dcd0f8b82588a07e45f7446/controllers/rollingupgrade_controller.go#L1027 There were several tests done when the functionality was added:...
This approach looks great to me..
Does the node's status or some annotation/label indicate that the node was found to be in bad shape by the node problem detector? Having another script is okay, but if...
Deleting all rolling-upgrades upon controller restarts could be dangerous. This should be done at least via a config option. If a particular rolling-upgrade object is causing the controller to crash...
Commit 631afe7 from @alantang888 addresses this. We need to do a new release of minion-manager that will include this fix.
AWS is moving towards using [LaunchTemplates](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/ec2-launch-templates.html) instead of LaunchConfigurations. LaunchTemplates allow specifying multiple instance types and thereby reducing the chances running out of capacity one anyone of them. Suggestions welcome;...
@souzabrunoftw , currently, the mechanism for determining whether to use on-demand or spot instances is pretty basic. Happy to accept PRs about to make the threshold configurable.
Another option might be to do `kubectl drain --dry-run` and see if that call succeeds. If there are PDBs that might get violated, the dry-run would fail (need to test...
One other option would be to label the node (or add some annotation) when it is about to be drained and terminated. Most Kubernetes aware tools allow for configuration based...