Define machines deletion policies.

[kubernetes-sigs]

Define machines deletion policies.

[rsdcastro]

From @p0lyn0mial on February 28, 2018 16:58

/cc @rsdcastro

[rsdcastro]

From @mvladev on March 7, 2018 10:5

If I would manually remove one Machine from a set of machines I would use the order bellow to determine which to remove:

1. Machine needs to be preferably removed - marked with annotation or some other means by human / cluster autoscaler (down-sizing scenario) / other controller
2. Health of the Machine - unhealthy machines first
3. If all else fails: select one at random

[ingvagabund]

@rsdcastro does it make sense to define the delete policies on the level of cloud provider itself? In my case, I am not able to delete an AWS machine directly. Instead, we stop an instance and append a specific suffix. Later on, a bot that periodically checks all AWS instances removes all instances that have the suffix. This way, I can't accidentally delete machines and this approach also helps to make permission more regular so only a small group of people can actually delete instances.

So I can imagine there can be more implementation of "how to delete an instance".

[mvladev]

@ingvagabund the physical (API call to the cloud provider) deletion of machines is not part of this issue and it's an implementation detail in your machine controller. It's about which Machine resource in K8S to delete during scale down.

Ideally the scale subresource should support this functionality, but at the moment it only has a number of desired replicas.

[ingvagabund]

@mvladev I put the physical deletion of machines here since this was the closest issue reported to that. You are right it's an implementation detail. Do you think it's worth opening a new issue and starting the conversation on that topic there?

[mvladev]

@ingvagabund sounds great!

[ingvagabund]

Do you think it's worth opening a new issue and starting the conversation on that topic there?

#446

[timothysc]

Closing in favor of #75