[sig-cloud-provider] Hybrid cloud native support.

[sergelogvinov]

What would you like to be added?

CLI global flag --hybrid-cloud=true

Why is this needed?

It's possible to run multiple CCMs in one cluster, but it requires a deep understanding of how the cloud provider works. In some cases, we need to add a few checks or extra logic on the CCM side.

The node-controller functions properly; only one CCM can find and initialize the node, while others skip it with an error or log message.

With the node-lifecycle-controller, we need to incorporate more checks into each CCM because we cannot determine if an instance exists or not. For instance, Azure-CCM always responds that the instance exists and is not switched off if the ProviderID is not Azure. The CCM responsible for this cloud can make the correct decision. I believe these checks are better made on the cloud provider side.

When a CCM starts, it registers the name of the cloud using 'cloudprovider.RegisterCloudProvider'. This name usually exists in the ProviderID cloud-provider-name://. For example, we can add the CLI global flag --hybrid-cloud=true, which will use this name from RegisterCloudProvider to check the prefix in the ProviderID. If it does not match, the functions InstanceExists and InstanceShutdown will not be called.

/sig cloud-provider

[aryasoni98]

I would like to work on it . Can you please assign me this issue .

[sergelogvinov]

I would like to work on it . Can you please assign me this issue .

Thank you @aryasoni98, but we need to wait triage first. I am not core developer, i cannot say is it fit or not.

[elmiko]

we are discussing this in the sig cloud provider meeting today.

i'm personally very interested in this idea. i think hybrid cloud usage will be important in the future. i would like to see a KEP describing how this will work and also so that we can have a discussion about the design.

from our sig meeting:

• should we have a new interface in the cloud provider to help with identifying ownership of a node in question
• would be nice if we could label a node object with some cloud provider type info, this would help to get away from the provider identification problem

it sounds like we have consensus in the meeting about wanting a KEP, if there are any questions about getting started please reach out in the sig-cloud-provider slack channel or to @bridgetkromhout or@elmiko directly.

[elmiko]

/triage accepted