GRPCRAFTGOLANG 2024
Network Control Plane
130 pods, 75GiB service, one config store
Owned a critical control-plane service at Amazon — a distributed config store that fans out updates to ~130 pods of a downstream data plane. Built on a custom Raft implementation in Go with strong consistency guarantees and low p99 publish latency.
The interesting work was operational: drift detection between followers, hands-free leader election under network partitions, and a control surface that lets oncall reason about cluster state without paging the team that built it.