Chorus: A Fast Single-Writer WAL Built on Google Cloud Storage

Chorus is a single-writer write-ahead log built directly on top of Google Cloud Storage. Using three zonal Rapid buckets plus one regional bucket, it delivers a durable, zone-fault-tolerant log without needing Kafka, etcd, or extra servers—the buckets themselves are the log.

Building cheap, scalable storage on object stores usually costs write latency. Chorus closes this gap by replicating each record across a strict majority of zonal buckets (typically three) and committing once any two confirm durability. Its control plane lives in a single regional GCS object managed via compare-and-swap, kept off the commit path so commits run at zonal-bucket speed.

For individually committed 4 KiB records, Chorus achieves a 1.71 ms median and 2.74 ms p99 latency—both faster than the median of Google Cloud's synchronously replicated regional block storage. The system is backed by formal methods: a P-language model-checked protocol, a Rust client tested via deterministic simulation, and implementation traces verified against the model using PObserve—giving engineers both performance and correctness guarantees.