A Shared KV Cache for vLLM: Reuse, Failover, and Elastic ScalingMengmei YeChen Wanget al.2026Kubecon + CloudNativeCon NA 2026Talk