Distributed systems
Relay
A durable, fault-tolerant distributed job queue in Rust. Expiring leases with fencing tokens, retries with backoff, delayed work, dead letters and bounded concurrency, on a single transactional SQLite writer. Delivery is at-least-once, and the failure semantics are documented and tested instead of hidden.
View on GitHub →- 9,792/sLease + ACK throughput, every state change durably committed
- 270 µsp99 lease + ACK latency
- +129%Throughput after a profile-driven scheduler fix (4,284 to 9,792 jobs/s)
- 7.4 MiBPeak memory with 64 worker threads
Measured on an Apple M5 with 5,000 jobs, SQLite WAL and synchronous=FULL. About 1.9k lines of Rust, with tests for competing claims, stale-token writes, lost ACKs and 1,000-job crash recovery.