At Cirrus I built the market data path that every chart, alert and automated strategy depends on. I wrote most of it; a few colleagues contributed to the shared services. The code is private, so this is a description, not a link.
One Rust service bridges broker WebSockets to NATS and forwards only the fields that changed. A second merges those changes into Redis, keeps a replayable log of every tick for the day, and builds 1-minute candles by replaying that log, so the candles are deterministic and gaps can be found and refilled. It parses and merges a tick in about 1 microsecond, and a standby takes over within seconds if the active node fails. On top of it, a trigger engine evaluates user conditions on every tick with a 99th-percentile latency of 2 ms or better in production. It replaced an older Python version, which I ran side by side in shadow mode and compared output for output before switching over.
0 likes · 0 views · Nov 22, 2026
Market data pipeline and trigger engine in Rust on NATS JetStream and Redis. About 1 microsecond per tick, triggers evaluated at 2 ms p99 in production. I desig