Scaling redirects to infinity on Vercel

This post details the engineering effort to scale Vercel's redirect system to millions of entries. It introduces a new architecture involving sharded Bloom filters and binary search over sorted keys within shards to optimize lookup performance and reduce memory usage. Key technical contributions include the shard structure (header, base64 encoded Bloom filter, JSON object of redirects), the lookup path leveraging Bloom filters for fast negative lookups and binary search for positive lookups to avoid full JSON parsing, and the resulting latency improvements and elimination of CPU-bound spikes.