Architecture

These pages tell how Enroute works inside: how it stores, replicates, and serves repositories, and how it is tested. You do not need them to run Enroute or to build on it.

Read the pages in the order below. Each one describes the system as built.

Page	Covers
Overview	Goals, the three tiers of state, protocol, and components
Consensus	One Raft log per repository, the log store, the state store, and recovery
Ledger and compute	The two roles, subscriptions, membership, checkpoints, and ownership
Storage	Layers, object ids, the storage service, the buffer pool, and tiered reads
Write path	Push, durability without waiting for the object store, and hooked pushes
Read path	Fetch planning and streaming, node-private views, and the node catalog
Serving	What the compute exposes: smart HTTP, hooks, and the API
Lifecycle	Create, delete, incarnations, and reclaiming files
Maintenance	Compaction, the layer sweep, and garbage collection
Jepsen tests	The consistency that Git clients see under faults, and how the suite checks it