Bluesky's one-hash bloom filter slashes deploy times from 90 to 15 minutes

One Hash Bloom Filters

Jim Calabro explains how Bluesky's backend uses bloom filters to skip millions of database queries per second, then dives into a one-hash variant that splits the bit array into pairwise coprime partitions. His Go implementation, gloom, has run in production for nine months and powers the Jetstream sync service. Cache-line alignment, a single MUL instruction for block selection, and sharded atomic writes keep lookups to one cache miss.