fix node removal

[pranavcracker]

Description

remove_node procedure fails immediately with KME0006 when the control-plane node being removed is offline or powered off not reachable.

• check_nodes_accessibility must only validate nodes that are staying in the cluster, not the ones being removed.

Solution

Replaced make_group_from_roles(['control-plane', 'balancer']) with get_unchanged_nodes().having_roles(['control-plane', 'balancer']) in check_nodes_accessibility(). get_unchanged_nodes() returns all nodes minus those being added or removed,

How to apply

NA

Test Cases

TestCase 1

Test Configuration:

• OS: Ubuntu 22.04
• Inventory: FullHA (3 control-planes, 2 balancers, 3 workers)

Steps:

1. Power off one control-plane node.
2. Add the node to procedure.yaml under nodes.
3. Run kubemarine remove_node procedure.yaml.

Results:

Before	After
Fails immediately with KME0006: Nodes ['x.x.x.x'] are not reachable	Procedure proceeds; etcd member removed manually from surviving control-plane, node deleted from cluster

Checklist

• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• There is no breaking changes, or migration patch is provided
• Integration CI passed
• Unit tests. If Yes list of new/changed tests with brief description
• There is no merge conflicts

Unit tests

• test_remove_control_plane_if_offline — new test that marks a control-plane node as offline, sets it as the node to remove, and asserts no KME0006 is raised during run_tasks.

[theboringstuff]

I am not sure what this test does. I tried to revert above changes and test still passes, so it does not check new changes. If this test is not relevant, I suggest to remove it so that we have less tests to maintain

[pranavcracker]

I think you are right. After deep investigation I confirmed both make_group_from_roles and get_unchanged_nodes().having_roles() always produce the same group at LIGHT stage — calculate_nodegroups excludes the removed node from nodes['all'] before this check runs.

I had tested this issue on miniha cluster earlier. I will investigate it more deeper on FullHA cluster.

[alexarefev]

From my prospective it's quite weird case when the control-plane node is unreachable and it's needed to remove it from the cluster.