Stateful Container Migration employing Checkpoint-based Restoration for Orchestrated Container Clusters

摘要

Recently, container-leveraged cloud computing technology is being adopted in a wide range of ICT areas. For a more effective management of diversified containers over multiple machine nodes, several container orchestrators including Kubernetes have been utilized in production cloud environments. Due to reliability reasons, stateful containers in an orchestrated container cluster are sometimes required to be migrated to other clusters. Typically, a storage-based migration approach is commonly adopted despite its considerable downtime, which also requires the application-level implementation of storage components for state save and restore. Thus, in this paper, we propose a checkpoint-based approach of stateful container migration for orchestrated container clusters. By checkpointing and restoring in-memory states of processes that are running in the containers with CRIU (Checkpoint/Restore in Userspace) feature, container migration can be realized without the application-level implementation of storage components. We then verify the proposed approach at a PoC(Proof-of-Concept) level by managing the associated migration procedure with the container orchestrator.