Kura 0.7.4 · Hive

Kura 0.7.4 improves cluster reliability after pod rollouts and handles bursty concurrent uploads better.

Peer discovery and primary routing now rely on runtime mesh health, not only Kubernetes readiness. The headless discovery DNS name is resolved on every membership pass and each returned address is probed individually, so a pod can no longer become isolated with only itself in the ring after a rollout. The controller also requires serving state, writer lock ownership, and enough ring visibility before it keeps or selects a public primary.
File descriptor limits are now derived from the process capacity instead of hardcoded defaults, giving deployments more headroom for bursty REAPI ByteStream uploads. When the FD pool is exhausted, Kura returns a clear RESOURCE_EXHAUSTED error instead of an internal CAS persistence failure. Helm deployments now use dynamic sizing unless you explicitly override the values.