Kubernetes v1.36 introduces server-side sharded list and watch to optimize controller performance in high-scale environments. By moving filtering logic to the API server, controller replicas only process events relevant to their specific hash range, reducing CPU, memory, and network overhead.
Key points
Client-side sharding is inefficient because every replica processes the full event stream, wasting resources on discarded objects.
Server-side sharding offloads filtering to the API server, ensuring replicas receive only data within their assigned hash range.
Deterministic FNV-1a hashing on metadata fields (UID or namespace) allows consistent partitioning across multiple API server instances.
The `shardInfo` field in list responses allows controllers to verify if the server successfully applied the requested filter.
Controllers should maintain fallback logic to perform client-side filtering if the server does not support or honor the requested shard selector.