|
| 1 | +# Coordinator Proxy |
| 2 | + |
| 3 | +The **Coordinator Proxy** is a standalone HTTP reverse-proxy application that sits between prover clients and multiple upstream Scroll L2 Coordinators. It exposes the same REST API surface as a real coordinator, allowing provers to connect transparently without knowing that requests are being load-balanced across a pool of backend coordinators. |
| 4 | + |
| 5 | +--- |
| 6 | + |
| 7 | +## Table of Contents |
| 8 | + |
| 9 | +- [Overview](#overview) |
| 10 | +- [Architecture](#architecture) |
| 11 | +- [Key Components](#key-components) |
| 12 | +- [Data Structures](#data-structures) |
| 13 | +- [API Endpoints](#api-endpoints) |
| 14 | +- [Authentication Flow](#authentication-flow) |
| 15 | +- [Task Routing](#task-routing) |
| 16 | +- [Configuration](#configuration) |
| 17 | +- [Usage](#usage) |
| 18 | +- [Design Decisions](#design-decisions) |
| 19 | + |
| 20 | +--- |
| 21 | + |
| 22 | +## Overview |
| 23 | + |
| 24 | +In a production environment, a single coordinator may not be sufficient to serve a large fleet of provers. The coordinator proxy addresses this by: |
| 25 | + |
| 26 | +- **Multiplexing** a single prover connection across multiple upstream coordinators. |
| 27 | +- **Authenticating** provers locally using the same JWT challenge-response mechanism as a real coordinator. |
| 28 | +- **Maintaining per-upstream, per-prover sessions** (including bearer tokens) so each coordinator sees the prover as a direct client. |
| 29 | +- **Routing `get_task` and `submit_proof` requests** intelligently across upstreams with sticky priority and random fallback. |
| 30 | + |
| 31 | +Because the proxy exposes the standard coordinator API (`/coordinator/v1/...`), existing prover SDKs require no changes to work through the proxy. |
| 32 | + |
| 33 | +--- |
| 34 | + |
| 35 | +## Architecture |
| 36 | + |
| 37 | +``` |
| 38 | +┌─────────────┐ ┌─────────────────────┐ ┌─────────────────┐ |
| 39 | +│ Prover SDK │─────>│ Coordinator Proxy │─────>│ Coordinator A │ |
| 40 | +└─────────────┘ │ (port 8590) │ └─────────────────┘ |
| 41 | + │ │ ┌─────────────────┐ |
| 42 | + │ • Local Auth │─────>│ Coordinator B │ |
| 43 | + │ • Session Manager │ └─────────────────┘ |
| 44 | + │ • Task Router │ ┌─────────────────┐ |
| 45 | + │ • Token Cache │─────>│ Coordinator C │ |
| 46 | + └─────────────────────┘ └─────────────────┘ |
| 47 | +``` |
| 48 | + |
| 49 | +### Entry Point |
| 50 | + |
| 51 | +| File | Role | |
| 52 | +|------|------| |
| 53 | +| `coordinator/cmd/proxy/main.go` | Thin wrapper that invokes the CLI application. | |
| 54 | +| `coordinator/cmd/proxy/app/app.go` | Bootstraps the proxy: parses `ProxyConfig`, optionally initializes the database, and starts the HTTP server with graceful shutdown. | |
| 55 | +| `coordinator/cmd/proxy/app/flags.go` | Defines HTTP server flags (`--http`, `--http.addr`, `--http.port`). Default port is `8590`. | |
| 56 | + |
| 57 | +--- |
| 58 | + |
| 59 | +## Key Components |
| 60 | + |
| 61 | +### 1. Auth Controller (`auth.go`) |
| 62 | + |
| 63 | +Handles the `/login` endpoint locally, then fans out a **proxy-login** to every upstream coordinator asynchronously. |
| 64 | + |
| 65 | +- Runs standard challenge-response validation (reuses existing coordinator auth logic). |
| 66 | +- On success, spawns goroutines to call `proxy_login` on each upstream. |
| 67 | +- Stores returned upstream tokens in the prover's session. |
| 68 | + |
| 69 | +### 2. GetTask Controller (`get_task.go`) |
| 70 | + |
| 71 | +Routes prover task requests to the most appropriate upstream. |
| 72 | + |
| 73 | +- First tries the **priority upstream** (sticky routing from a previous successful task assignment). |
| 74 | +- If the priority upstream has no task, shuffles the remaining upstreams randomly and tries each until one returns a task. |
| 75 | +- Prefixes the returned `taskID` with the upstream name (`upstream:taskID`) so that `submit_proof` can route correctly. |
| 76 | + |
| 77 | +### 3. SubmitProof Controller (`submit_proof.go`) |
| 78 | + |
| 79 | +Forwards proof submissions to the correct upstream coordinator. |
| 80 | + |
| 81 | +- Parses the `upstream:taskID` prefix to determine the target coordinator. |
| 82 | +- Uses the cached upstream token from the prover session. |
| 83 | +- On success, clears the priority upstream (task is complete). |
| 84 | + |
| 85 | +### 4. Client Manager (`client_manager.go`) |
| 86 | + |
| 87 | +Manages the proxy's own identity and bearer token for each upstream. |
| 88 | + |
| 89 | +- Keeps a cached `upClient` with a background re-login goroutine. |
| 90 | +- If the cached token expires or becomes invalid, `Reset()` clears it so a fresh login is attempted. |
| 91 | + |
| 92 | +### 5. Prover Session Manager (`prover_session.go`) |
| 93 | + |
| 94 | +Maintains in-memory (and optionally DB-backed) sessions for every prover public key. |
| 95 | + |
| 96 | +- Each session holds a map of `upstream -> loginToken`. |
| 97 | +- Transparently refreshes expired tokens. |
| 98 | +- Implements a session size limit; when exceeded, the old session map is rotated to a deprecated map rather than deleted immediately. |
| 99 | + |
| 100 | +### 6. Priority Upstream Manager |
| 101 | + |
| 102 | +Stores sticky routing hints (`publicKey -> upstreamName`). |
| 103 | + |
| 104 | +- When a prover successfully receives a task from an upstream, that upstream becomes the priority for the next `get_task` call. |
| 105 | +- Can be persisted to the database so routing preferences survive restarts. |
| 106 | + |
| 107 | +### 7. HTTP Client (`client.go`) |
| 108 | + |
| 109 | +Low-level HTTP client (`upClient`) that implements the `ProxyCli` and `ProverCli` interfaces. |
| 110 | + |
| 111 | +- `ProxyCli`: used by the proxy itself to log in to an upstream. |
| 112 | +- `ProverCli`: used to impersonate a prover when calling `get_task` or `submit_proof` on an upstream. |
| 113 | + |
| 114 | +--- |
| 115 | + |
| 116 | +## Data Structures |
| 117 | + |
| 118 | +### Configuration |
| 119 | + |
| 120 | +```go |
| 121 | +// ProxyConfig — top-level configuration |
| 122 | +type ProxyConfig struct { |
| 123 | + ProxyManager *ProxyManager `json:"proxy_manager"` |
| 124 | + ProxyName string `json:"proxy_name"` |
| 125 | + Coordinators map[string]*UpStream `json:"coordinators"` |
| 126 | +} |
| 127 | + |
| 128 | +// ProxyManager — auth, verifier, client identity, optional DB |
| 129 | +type ProxyManager struct { |
| 130 | + Verifier *VerifierConfig `json:"verifier"` // minimum prover version |
| 131 | + Client *ProxyClient `json:"proxy_cli"` // proxy's own identity |
| 132 | + Auth *Auth `json:"auth"` // JWT secret & expiry |
| 133 | + DB *database.Config `json:"db,omitempty"` |
| 134 | +} |
| 135 | + |
| 136 | +// ProxyClient — identity the proxy uses to authenticate with upstreams |
| 137 | +type ProxyClient struct { |
| 138 | + ProxyName string `json:"proxy_name"` |
| 139 | + ProxyVersion string `json:"proxy_version,omitempty"` |
| 140 | + Secret string `json:"secret,omitempty"` |
| 141 | +} |
| 142 | + |
| 143 | +// UpStream — per-coordinator connection settings |
| 144 | +type UpStream struct { |
| 145 | + BaseUrl string `json:"base_url"` |
| 146 | + RetryCount uint `json:"retry_count"` |
| 147 | + RetryWaitTime uint `json:"retry_wait_time_sec"` |
| 148 | + ConnectionTimeoutSec uint `json:"connection_timeout_sec"` |
| 149 | + CompatibileMode bool `json:"compatible_mode,omitempty"` |
| 150 | +} |
| 151 | +``` |
| 152 | + |
| 153 | +### Runtime Data Structures |
| 154 | + |
| 155 | +```go |
| 156 | +// Client interface — abstracts per-upstream access |
| 157 | +type Client interface { |
| 158 | + Client(string) ProverCli // token-bound prover client |
| 159 | + ClientAsProxy(context.Context) ProxyCli // proxy's own authenticated client |
| 160 | + Name() string |
| 161 | +} |
| 162 | + |
| 163 | +// upClient — actual HTTP implementation |
| 164 | +type upClient struct { |
| 165 | + httpClient *http.Client |
| 166 | + baseURL string |
| 167 | + loginToken string |
| 168 | + compatibileMode bool |
| 169 | + resetFromMgr func() |
| 170 | +} |
| 171 | + |
| 172 | +// ProverManager — registry of active prover sessions |
| 173 | +type ProverManager struct { |
| 174 | + data map[string]*proverSession |
| 175 | + willDeprecatedData map[string]*proverSession |
| 176 | + sizeLimit int |
| 177 | + persistent *proverDataPersist |
| 178 | + // ... prometheus metrics |
| 179 | +} |
| 180 | + |
| 181 | +// proverSession — per-prover tokens across all upstreams |
| 182 | +type proverSession struct { |
| 183 | + persistent *proverDataPersist |
| 184 | + proverToken map[string]loginToken |
| 185 | + completionCtx context.Context |
| 186 | +} |
| 187 | + |
| 188 | +// loginToken — upstream token with a monotonic phase |
| 189 | +type loginToken struct { |
| 190 | + token string |
| 191 | + phase uint |
| 192 | +} |
| 193 | +``` |
| 194 | + |
| 195 | +### Database Models (Optional Persistence) |
| 196 | + |
| 197 | +When a database is configured, the proxy persists the following tables: |
| 198 | + |
| 199 | +| Table | Columns | Purpose | |
| 200 | +|-------|---------|---------| |
| 201 | +| `prover_sessions` | `public_key`, `upstream`, `up_token`, `expired` | Stores upstream tokens per prover so restarts do not force re-login. | |
| 202 | +| `priority_upstream` | `public_key`, `upstream` | Restores sticky routing preferences after restart. | |
| 203 | + |
| 204 | +--- |
| 205 | + |
| 206 | +## API Endpoints |
| 207 | + |
| 208 | +### Exposed API (Prover → Proxy) |
| 209 | + |
| 210 | +All endpoints are mounted under `/coordinator/v1/`. |
| 211 | + |
| 212 | +| Method | Path | Auth | Description | |
| 213 | +|--------|------|------|-------------| |
| 214 | +| `GET` | `/challenge` | None | Returns a challenge token for the prover to sign. | |
| 215 | +| `POST` | `/login` | Challenge | Validates the prover signature, then fans out `proxy_login` to all upstreams. | |
| 216 | +| `POST` | `/get_task` | JWT | Routes a task request to the best available upstream. | |
| 217 | +| `POST` | `/submit_proof` | JWT | Forwards the proof to the upstream that issued the task. | |
| 218 | + |
| 219 | +### Upstream API Calls (Proxy → Coordinator) |
| 220 | + |
| 221 | +| Method | Path | Purpose | |
| 222 | +|--------|------|---------| |
| 223 | +| `GET` | `/coordinator/v1/challenge` | Obtain a challenge token for proxy login. | |
| 224 | +| `POST` | `/coordinator/v1/login` | Proxy authenticates itself as a client. | |
| 225 | +| `POST` | `/coordinator/v1/proxy_login` | Forward prover identity to the upstream. | |
| 226 | +| `POST` | `/coordinator/v1/get_task` | Forward prover task request. | |
| 227 | +| `POST` | `/coordinator/v1/submit_proof` | Forward proof submission. | |
| 228 | + |
| 229 | +### Key Interfaces |
| 230 | + |
| 231 | +```go |
| 232 | +// ProxyCli — proxy's own client to an upstream |
| 233 | +type ProxyCli interface { |
| 234 | + Login(ctx context.Context, genLogin func(string) (*types.LoginParameter, error)) (*ctypes.Response, error) |
| 235 | + ProxyLogin(ctx context.Context, param *types.LoginParameter) (*ctypes.Response, error) |
| 236 | + Token() string |
| 237 | + Reset() |
| 238 | +} |
| 239 | + |
| 240 | +// ProverCli — prover-impersonating client to an upstream |
| 241 | +type ProverCli interface { |
| 242 | + GetTask(ctx context.Context, param *types.GetTaskParameter) (*ctypes.Response, error) |
| 243 | + SubmitProof(ctx context.Context, param *types.SubmitProofParameter) (*ctypes.Response, error) |
| 244 | +} |
| 245 | +``` |
| 246 | + |
| 247 | +--- |
| 248 | + |
| 249 | +## Authentication Flow |
| 250 | + |
| 251 | +### 1. Prover → Proxy `/login` |
| 252 | + |
| 253 | +1. Prover requests a challenge from the proxy (`GET /challenge`). |
| 254 | +2. Prover signs the challenge and sends it to `POST /login`. |
| 255 | +3. The proxy validates the signature locally using the same logic as a real coordinator. |
| 256 | +4. On success, the proxy spawns asynchronous goroutines to call `POST /proxy_login` on **every** upstream coordinator. |
| 257 | +5. Each upstream returns a bearer token specific to that prover. |
| 258 | +6. The proxy stores all upstream tokens in the prover's session and returns its own JWT to the prover. |
| 259 | + |
| 260 | +### 2. Token Refresh & Resilience |
| 261 | + |
| 262 | +Both `GetTask` and `SubmitProof` implement a retry-on-token-expiry pattern: |
| 263 | + |
| 264 | +1. Try the request with the cached upstream token. |
| 265 | +2. If the upstream returns `ErrJWTTokenExpired` or `ErrJWTCommonErr`, trigger `maintainLogin` to refresh the token. |
| 266 | +3. Retry the request once with the new token. |
| 267 | + |
| 268 | +The `ClientManager` also maintains a background login goroutine for the proxy's own identity; if the cached client fails, `Reset()` clears it so a fresh login is attempted on the next call. |
| 269 | + |
| 270 | +--- |
| 271 | + |
| 272 | +## Task Routing |
| 273 | + |
| 274 | +### Sticky Priority Routing |
| 275 | + |
| 276 | +When a prover successfully receives a task from an upstream, that upstream is recorded as the **priority upstream** for that prover. On the next `get_task` call, the proxy tries the priority upstream first. |
| 277 | + |
| 278 | +### Random Fallback |
| 279 | + |
| 280 | +If the priority upstream has no available task (or returns an error), the proxy shuffles the remaining upstreams randomly and tries each one until a task is returned or all upstreams are exhausted. |
| 281 | + |
| 282 | +### Task ID Namespacing |
| 283 | + |
| 284 | +To ensure `submit_proof` routes to the correct upstream, the proxy prefixes the task ID returned to the prover: |
| 285 | + |
| 286 | +``` |
| 287 | +upstreamName:originalTaskID |
| 288 | +``` |
| 289 | + |
| 290 | +The `SubmitProof` controller parses this prefix, extracts the upstream name, and forwards the proof to the correct coordinator. |
| 291 | + |
| 292 | +--- |
| 293 | + |
| 294 | +## Configuration |
| 295 | + |
| 296 | +Example `config_proxy.json`: |
| 297 | + |
| 298 | +```json |
| 299 | +{ |
| 300 | + "proxy_manager": { |
| 301 | + "proxy_cli": { |
| 302 | + "proxy_name": "proxy_name", |
| 303 | + "secret": "client private key" |
| 304 | + }, |
| 305 | + "auth": { |
| 306 | + "secret": "proxy secret key", |
| 307 | + "challenge_expire_duration_sec": 3600, |
| 308 | + "login_expire_duration_sec": 3600 |
| 309 | + }, |
| 310 | + "verifier": { |
| 311 | + "min_prover_version": "v4.4.45", |
| 312 | + "verifiers": [] |
| 313 | + }, |
| 314 | + "db": { |
| 315 | + "driver_name": "postgres", |
| 316 | + "dsn": "postgres://localhost/scroll?sslmode=disable", |
| 317 | + "maxOpenNum": 200, |
| 318 | + "maxIdleNum": 20 |
| 319 | + } |
| 320 | + }, |
| 321 | + "coordinators": { |
| 322 | + "sepolia": { |
| 323 | + "base_url": "http://localhost:8555", |
| 324 | + "retry_count": 10, |
| 325 | + "retry_wait_time_sec": 10, |
| 326 | + "connection_timeout_sec": 30 |
| 327 | + } |
| 328 | + } |
| 329 | +} |
| 330 | +``` |
| 331 | + |
| 332 | +### Field Reference |
| 333 | + |
| 334 | +| Field | Description | |
| 335 | +|-------|-------------| |
| 336 | +| `proxy_manager.proxy_cli.secret` | ECDSA private key material used to derive the proxy's signing key for upstream login. | |
| 337 | +| `proxy_manager.auth.secret` | JWT HMAC key for prover-to-proxy sessions. | |
| 338 | +| `proxy_manager.verifier.min_prover_version` | Minimum prover version allowed to connect. | |
| 339 | +| `proxy_manager.db` | Optional database configuration. If omitted, the proxy runs in memory-only mode (no persistence across restarts). | |
| 340 | +| `coordinators.*.base_url` | HTTP endpoint of the upstream coordinator. | |
| 341 | +| `coordinators.*.compatible_mode` | If `true`, skips `proxy_login` and uses standard login with a dummy token (for legacy coordinators). | |
| 342 | + |
| 343 | +--- |
| 344 | + |
| 345 | +## Usage |
| 346 | + |
| 347 | +### Build |
| 348 | + |
| 349 | +```bash |
| 350 | +cd coordinator |
| 351 | +make proxy |
| 352 | +``` |
| 353 | + |
| 354 | +### Run |
| 355 | + |
| 356 | +```bash |
| 357 | +./build/bin/coordinator_proxy --config conf/config_proxy.json |
| 358 | +``` |
| 359 | + |
| 360 | +### Run with custom HTTP address |
| 361 | + |
| 362 | +```bash |
| 363 | +./build/bin/coordinator_proxy --config conf/config_proxy.json --http.addr 0.0.0.0 --http.port 8590 |
| 364 | +``` |
| 365 | + |
| 366 | +### Prover Connection |
| 367 | + |
| 368 | +Provers connect to the proxy exactly as they would connect to a real coordinator: |
| 369 | + |
| 370 | +``` |
| 371 | +https://proxy.example.com/coordinator/v1 |
| 372 | +``` |
| 373 | + |
| 374 | +No SDK changes are required. |
| 375 | + |
| 376 | +--- |
| 377 | + |
| 378 | +## Design Decisions |
| 379 | + |
| 380 | +| Decision | Rationale | |
| 381 | +|----------|-----------| |
| 382 | +| **Sticky routing for tasks** | Reduces cross-coordinator state churn; a prover that received a task from upstream A is likely to get the next task from the same upstream. | |
| 383 | +| **Random load balancing as fallback** | Simple and stateless; avoids hot-spotting when one upstream runs out of tasks. | |
| 384 | +| **Task ID namespacing** | Allows the proxy to remain stateless for proof submissions; the task ID itself encodes the routing target. | |
| 385 | +| **Session size limit with rotation** | Prevents unbounded memory growth in the proxy; old sessions are moved to a deprecated map and eventually garbage-collected. | |
| 386 | +| **Phase-based token updates** | A monotonic `phase` counter on `loginToken` prevents stale concurrent login attempts from overwriting a fresher token. | |
| 387 | +| **Compatible mode** | Allows the proxy to work with older coordinators that do not support the `proxy_login` endpoint. | |
| 388 | +| **Optional DB persistence** | The proxy can run entirely in memory for simplicity, or use a database for token and routing persistence across restarts. | |
0 commit comments