Sentinel
A distributed message queue written from scratch in Go to understand how a system like Kafka actually works underneath the API. Not a wrapper; the storage engine, consensus layer, and wire protocol are all hand-rolled.
what it is
At the bottom is an LSM-tree. Writes land in an in-memory skip-list memtable. When a memtable fills it flushes to an immutable SSTable on disk. Every record is CRC32-checksummed so a flipped bit gets caught on read. A write-ahead log fronts the memtable for durability: process death mid-flush replays cleanly. Leveled compaction keeps read amplification bounded.
On top sits a gRPC wire protocol with Kafka-style primitives: topics, partitions, consumer groups that track offsets.
raft, and how to test it
A single broker is not interesting. The hard part is agreeing on write order under failures, so Sentinel implements Raft from scratch: leader election, log replication, and split-brain prevention.
Consensus bugs only show up under specific interleavings of delays and drops you cannot reliably reproduce against real sockets. The cluster runs against a deterministic in-memory network simulator. A failing run is a reproducible run.
status
Learning project, not running production traffic. Code at GitHub.