How do you use virtual nodes for load balancing to mitigate hotspots, and how does hashing user IDs to virtual nodes help spread the load across the cluster?

Virtual nodes (vnodes) are a lightweight abstraction that lets a single physical machine host many logical partitions. By hashing a user ID to a vnode, we distribute keys uniformly across the cluster, which mitigates hotspots that would otherwise occur if a few physical nodes received a disproportionate share of traffic. The hashing function maps the user ID to a point on a ring; each vnode occupies a contiguous segment of the ring. When a new node joins, we add its vnodes to the ring and only the keys that map to those vnodes need to be moved, reducing data churn. Similarly, when a node leaves, its vnodes are reassigned to the next healthy nodes. Because each physical node owns many vnodes, the probability that a single node becomes a bottleneck is low, and load can be balanced by adjusting the number of vnodes per machine. In practice, we use 128–256 vnodes per node, which gives a good trade‑off between granularity and metadata overhead.