The test failure can manifest on both `ThreeNodeReplGlobalStateRestartTest` and `ThreeNodeDistGlobalStateRestartTest` tests.

The issue happens when the extraneous node delays the join request until the cluster recovers from the state. It receives a new topology concurrently with the join request. We can safely ignore the topology updates while waiting for the node to join. We only need updates for the stable topology to identify the cluster has finished the recovery process.