Towards efficient and secure data sharing between co-located serverless containers in Kubernetes

The serverless computing paradigm is widely recognized as a practical solution for highly elastic compute and data-intensive workloads in the Cloud. Emerging serverless Cloud services based on container technologies provide higher degrees of flexibility for adopting new applications. However, managing temporary data in serverless environments remains a challenge. The stateless nature of serverless computing requires reliance on disaggregated storage, leading to latency issues due to data movements and performance degradation. Existing caching and temporary data store approaches present limitations in handling large datasets or impose additional infrastructure costs and management complexities.

To address these challenges, we propose the inclusion of GEDS (Generic Ephemeral Data Store) in serverless Kubernetes architectures. The objective is to leverage node locality using the ephemeral host file system allocated for each serverless container, enabling applications to effectively store and share temporary data across concurrent and successive invocations. This article explores the limitations and opportunities for effective data sharing between serverless co-located containers in Kubernetes deployments. Our findings demonstrate how by using memory-mapped files and file descriptor passing through shared volume mounts provides effective storage resources sharing between containers efficiently and securely.