Less Provisioning: A Hybrid Resource Scaling Engine for Long-Running Services With Tail Latency Guarantees