CockroachDB engineers faced a significant challenge: adding vector search to their distributed database while maintaining their core architectural principles. When existing solutions failed to meet their requirements, they built C-SPANN, a novel vector indexing system. This decision came after evaluating dozens of algorithms that couldn't satisfy their strict constraints for distributed, scalable vector search without compromising transactional consistency.

The team established six strict requirements: no central coordinator, minimal memory usage, real-time updates, no hot spots, sharding compatibility, and incremental updates. Most popular vector indexes like HNSW couldn't meet these demands. Traditional solutions required either single-node deployments, large in-memory structures, or separate systems entirely, making them fundamentally incompatible with CockroachDB's distributed SQL architecture.

C-SPANN combines ideas from Microsoft's SPANN, SPFresh, and Google's ScaNN projects, creating a hierarchical K-means tree structure. This design allows for parallel processing across partitions while maintaining low latency. The wide, shallow tree structure enables efficient searching through billions of vectors without excessive memory usage or network hops, solving the distributed vector search problem within a transactional database framework.