When a private Kura runner-cache node becomes unready, cache dispatches now fall back to the public cache instead of using the dead private endpoint. Previously, the private node-port URL was treated as a hard override, so builds timed out on cache downloads until the node recovered. With this fix, a readiness heartbeat tracks whether the node-port endpoint is observable, and stale nodes are skipped within about two minutes. CI builds and customer macOS runner builds are no longer blocked by a wedged Kura node.
Hive