link/sidecar/README.md

# uo-link sidecar

The Rust half of the bridge. It terminates the loopback link to the ServUO shard and (as it grows) exposes WebSocket + REST to the website.

```
website ──WS (live feed) / REST (queries)──►  sidecar  ──loopback TCP 127.0.0.1:7788──►  shard
                                              (this)      newline-JSON, bidirectional
```

The sidecar is the TCP **listener**; the shard dials out to it. That is what keeps the game unreachable from the website — the game exposes no port of its own. See `../docs/PLAN.md` §2.

## Run

```bash
cargo run                 # info logging
RUST_LOG=debug cargo run  # see every event, incl. pong heartbeats
```

Binds `127.0.0.1:7788` and waits for the shard to connect. Boot the shard (or it will reconnect on its own) and watch `server.hello` arrive.

## Status

| Piece | State |
|-------|-------|
| Shard link (`shard.rs`) | **done** — accepts the shard, reads events, sends commands, re-accepts on disconnect. Verified against the live shard: received `server.hello`, round-tripped a `ping`→`pong`, and reconnected after a sidecar restart. |
| WebSocket feed (`web.rs`) | **done** — `/ws` fans every shard event out to connected clients via a `broadcast`. Verified: a WS client received `ws.hello` then live `pong` events relayed from the shard. Live-only, no replay. |
| REST queries (`rpc.rs` + `web.rs`) | **done** — synchronous queries and commands, correlated to shard replies by id. Verified end-to-end against the live shard, success and error paths. |
| SQLite persistence (`store.rs`) | **done** — every live event persisted; history/economy served from the DB; profiles cached with shard-down fallback; link map. Verified: data survived a sidecar restart, and a cached profile served at 200 with the shard killed. |

**The sidecar is feature-complete.** All four pieces work end-to-end against the live shard.

The web server binds `127.0.0.1:8080` by default (`WEB_ADDR` in `main.rs`). Widen the bind and add auth before exposing it off-host.

### Routes

| Method | Path | Shard command | Reply |
|--------|------|---------------|-------|
| GET | `/health` | — | `ok` |
| GET | `/ws` | — | live event feed (WebSocket) |
| GET | `/char/{account}/{slot}` | `char.request` | `char.profile` |
| GET | `/char/serial/{serial}` | `char.request` | `char.profile` |
| GET | `/roster/{account}` | `account.roster` | `account.roster` |
| GET | `/vendors/{account}` | `vendor.snapshot` | `vendor.snapshot` |
| POST | `/link/confirm` `{code, websiteUserId}` | `link.confirm` | `link.ok` / `link.error` |
| POST | `/towncrier` `{id, lines, durationSec}` | `towncrier.add` | `towncrier.ok` / `towncrier.error` |
| DELETE | `/towncrier/{id}` | `towncrier.remove` | `towncrier.ok` / `towncrier.error` |
| GET | `/link/{account}` | — (reads store) | `{account, websiteUserId}` or 404 |
| GET | `/history?kind=&limit=` | — (reads store) | `{events: [...]}` newest first |
| GET | `/economy?limit=` | — (reads store) | `{series: [...]}` supply snapshots |

A shard `*.error` reply maps to HTTP 404 (unknown/not-found) or 400 (bad request). No shard connected → 503; no reply within 10 s → 504. `GET /char/serial/{serial}` falls back to the cached profile when the shard is unreachable, so an already-viewed character still renders during an outage.

## Design

- **`shard.rs`** — `serve()` binds the listener and accepts shard connections in a loop. Each connection splits into read/write halves: the read half parses newline-JSON into `ShardEvent { kind, value }` and forwards them; the write half drains an mpsc of command lines. `ShardHandle::send` posts a command to whichever shard is currently connected, and **drops with a warning if none is** — a website query during a shard outage should fail fast and retry, not queue behind a reconnect. Live *events* that must survive an outage are buffered by the shard, not here.
- **`web.rs`** — the website-facing HTTP surface (axum). `AppState` holds the `broadcast::Sender<String>`; each `/ws` client subscribes and forwards every event as a text frame. A client that lags past the broadcast buffer is warned and kept live (it just misses events) rather than stalling the others. This side *may* be exposed beyond loopback — it is the gatekeeper, so add auth when you do.
- **`rpc.rs`** — request/reply correlation over the one shard socket. A REST call registers a pending entry under a correlation id, sends the command, and awaits the reply (10 s timeout). The event loop routes any incoming line whose id is pending back to the waiter; everything else flows on as a live event. Recognizes three correlation fields, matching what the plugin echoes: `reqId` (queries), `code` (link), `id` (town-crier).
- **`store.rs`** — SQLite (`sqlx`). Three tables: `events` (the full live stream, append-only), `links` (account ↔ website user, mirrored from `link.ok`), `profiles` (last-known character sheet, cached from `char.profile`). History and economy read here instead of the shard; `pong` is dropped as ephemeral chatter. DB file defaults to `uo-link.db` (`DB_PATH` in `main.rs`), gitignored.
- **`main.rs`** — wires it together: the shard event loop first tries to route each line as an RPC reply; if it isn't one, the line is a live event — logged, persisted, and broadcast to WS.

## Wire protocol

Every line is one JSON object with `t` (epoch ms) and `kind`. The shard→sidecar events and sidecar→shard commands are catalogued in `../docs/PLAN.md` (§5 data catalog, §7 protocol) and were all validated end-to-end while building the plugin. Notable inbound commands the sidecar will issue: `char.request`, `account.roster`, `vendor.snapshot`, `link.confirm`, `towncrier.add`/`remove`, `ping`.