Time Box: 1 sprint

Adaptable topology is a cluster-topology mode that adjusts cluster control-plane and infrastructure behavior based on the current number of control-plane and worker nodes. The goal of this spike is to update the enhancement proposal with the answers to the following questions.

Research Questions

• If we run a learner on a second control-plane node and the voter fails, can quorum restore promote the learner? Or can only former voters be restored with quorum?
• Can CEO enforce etcd safety for 1↔3 node transitions? DualReplica offloads this to pacemaker. Do we need something similar?
• Can we guarantee both learner promotions happen together (or neither happens)?
• What other mechanisms are needed to prevent unsafe etcd transitions?

Involved Teams

• Control Plane (cluster-etcd-operator)

Acceptance Criteria

• Determine learner promotion capabilities during voter failures
• Document etcd safety enforcement requirements for 1↔3 node transitions
• Design atomic learner promotion mechanism
• Identify additional safety mechanisms needed
• Update enhancement proposal with findings and recommendations

Additional Context

• Enhancement Proposal: https://github.com/openshift/enhancements/pull/1905
• Critical for preventing etcd data loss during topology transitions
• Related to risk mitigation: "Risk: etcd Data Loss If Transitions Are Not Atomic"
• May require coordination with Control Plane team