We measure memory at the container (cgroup) level, not at the individual process level. That figure includes everything inside the container: your Node process RSS, file page cache, shared libraries, kernel data structures, and any child processes. It will always be higher than what process.memoryUsage().rss reports because Node only sees its own heap and resident pages. The deploy-time spike is the build step running inside the container before the final image is ready, which can pull in significantly more memory temporarily. The Railpack issue you linked is the right place to track that.