Nodens: Enabling Resource Efficient and Fast QoS Recovery of Dynamic Microservice Applications in Datacenters

Current microservice applications always meet with load and call graph dynamics. These dynamics can easily lead to inappropriate resource allocation for microservices, and further lead to Quality-of-Service (QoS) violations of applications. However, current microservice management works are incapable to handle these dynamics, mainly due to the execution blocking effect among microservices. We therefore propose Nodens, a runtime system that enables fast QoS recovery of the dynamic microservice application, while maintaining the efficiency of the resource usage. Nodens comprises a trafficbased load monitor, a blocking-aware load updater, and a resource-efficient query drainer. The load monitor periodically checks microservices' network bandwidth usage and predicts the monitored loads based on it. The load updater updates the actual "to-be-processed" load of each microservice to enable fast resource adjustment. The query drainer allocates just-enough excessive resources for microservices to drain the queued queries, which can ensure the QoS recovery time target. Our experiments show that Nodens can reduce the QoS recovery time by 12.1X with only the excessive resource usage of 6.1% on average, compared to the state-of-the-art microservice management systems.