Apache Druid vs runc

The 1.6.0 cycle opened on October 6 with a first release candidate that is mostly cleanup: deprecated libcontainer APIs removed, Go 1.26 now required, urfave/cli moved from v1 to v3, and the binary about 15% smaller on amd64. The changes with operational weight are that cpuAffinity and NUMA memoryPolicy no longer stop at 1024 CPUs or nodes, and that a container asking for a read-only rootfs without its own mount namespace is now refused. Eleven days earlier 1.5.2 worked around a kernel bug, present since v6.17, that corrupted memory inside runc and caused random crashes on cgroup v2. It also stopped runc exec --cgroup from escaping into a sibling cgroup.

The project keeps converting silent behaviour into explicit behaviour. Read-only requests that were ignored now fail, a umask that was dropped is now honoured, and poststart hooks moved to conform to the runtime spec. The same maintainers are shrinking the surface underneath, dropping the deprecated API and the high-level cilium/ebpf path, which took roughly 7.5% off the 1.5.2 binary. Stricter behaviour plus a smaller dependency set is what a component under every container host would be expected to do between majors.

1.5.0's notes pointed to a 1.6.0 in late October 2026, and the feed so far shows only rc.1, so further release candidates or the final are the next entries to expect. The refused read-only configurations are the change most likely to surface as user reports during testing.