RisingWave does not rely on the concept of replicas for high availability. Instead, it automatically handles failures by recovering from the last checkpoint stored in S3, which can be done in milliseconds. Additionally, RisingWave supports a Dedicated Serving Cluster feature, allowing the deployment of compute nodes dedicated to serving queries without running streaming. This setup can serve online queries without interruption, with potential increased latency during cache warm-up after recovery, but without errors.