Abstract
This paper proposes a new solution to handle the service downtime and performance degradation issues posed by traditional rolling updates when maintaining distributed in-memory key-value stores (IMKVS). The authors propose a live patching system for Redis Cluster which proactively manages software updates by modifying process memory directly without restarting nodes, while leveraging the cluster-internal gossip protocol for automatic patch distribution. The authors present a set of experiments comparing the efficacy of live patching (LP) to state-of-the-art rolling updates (CP) on a range of cluster sizes and memory states up to 200~GiB. We find that on the hardware specified (including Intel Xeon and AMD EPYC configurations) we were able to reproduce the key results and plots shown in the paper, illustrating the robust nature of these experiments and confirming that live patching achieves constant-time updates regardless of data size. Furthermore, the authors provided well-written shell scripts and easy-to-follow instructions within a QEMU virtual machine environment to reproduce this material.

This publication has 1 reference indexed in Scilit: