$ show flashbots/mev-boost-relay
This is the Flashbots MEV-Boost Relay, a production-grade Go service that mediates between Ethereum block proposers (validators) and block builders. It implements the proposer-builder separation (PBS) specification, providing REST APIs for builder block submission (with simulation), validator registration, and proposer block retrieval (getHeader/getPayload). The relay is the critical infrastructure that enables validators to outsource block construction to specialized builders, maximizing MEV extraction while maintaining trustless verification. The architecture is a multi-service system with three independently scalable components: an API server (builder, proposer, data, and internal endpoints), a housekeeper background service (validator updates, proposer duties, metrics), and a website service (public status page). It uses PostgreSQL for persistent storage, Redis for low-latency bid caching, optional Memcached for redundant payload storage, and connects to one or more beacon nodes (consensus layer clients) for chain data. Block simulation is performed via an external validation node with a concurrency rate limiter. Key engineering highlights include a multi-beacon-client with failover and least-used ordering, a datastore fallback chain (Redis → Memcached → PostgreSQL), streaming JSON decoding for large validator sets, SSZ/JSON content negotiation with version detection across Ethereum forks (Capella through Fulu), and extensive Prometheus/OpenTelemetry metrics. The codebase is well-tested with integration tests, mock databases, and test data for multiple fork versions. It is actively maintained with recent commits, CI/CD pipelines, and a clear migration path for database schema changes.
The LLM identified 4 files worth highlighting for reuse:
Implements a multi-beacon-node client with automatic failover, least-used ordering for heavy calls, and concurrent publish with first-success semantics. The client tracks node health via a sync.Map of errors and uses a round-robin-like selection for read operations. It is self-contained, depends only on the beaconclient package interfaces, and could be extracted as a generic multi-endpoint client with minimal changes.
A concurrency limiter for block simulation requests using sync.Cond and an atomic counter. Supports fast-track slots (high-value proposers) that bypass the queue, and a configurable maximum concurrency. The implementation is clean, thread-safe, and decoupled from the rest of the API service. It demonstrates a reusable pattern for rate-limiting external RPC calls with priority queuing.
Provides a streaming JSON decoder for the beacon node's state validators endpoint, avoiding buffering hundreds of megabytes of JSON in memory. It uses a feature flag with a fallback to the standard decoder. The pattern of streaming large API responses with a custom decoder is reusable for any JSON array response where memory is a concern.
Implements a multi-tier retrieval chain for execution payloads: Redis (primary cache), Memcached (secondary cache), and PostgreSQL (persistent fallback). Each tier is tried in order with error handling and metrics. The pattern is a clean abstraction for building resilient data access layers with multiple backends, and the interface could be reused for other cache hierarchies.