Converge main: telemetry auth wiring + absorb 202 reconcile

crates/gitforge-api/src/middleware.rs

@@ -19,7 +19,7 @@ const PUBLIC_PATHS: &[&str] = &["/health", "/metrics", "/swagger-ui", "/api-docs
 /// instance the aggregate counters leak activity to anonymous parties, so
 /// the default is authenticated access. `/health` stays public either way
 /// — a liveness probe carries no instance data (F5).
-fn telemetry_is_public(raw: Option<&str>) -> bool {
+pub(crate) fn telemetry_is_public(raw: Option<&str>) -> bool {
     matches!(raw, Some(v) if v.trim() == "1")
 }
 

crates/gitforge-api/src/server.rs

@@ -103,19 +103,25 @@ impl ApiServer {
                 .allow_headers(Any)
         };
 
-        // Public routes (no auth required)
+        // Public routes (no auth required). Telemetry surfaces are added
+        // only when the operator opted in; otherwise they ride behind the
+        // auth middleware via the protected router (F5).
+        let telemetry_public = crate::middleware::telemetry_is_public(
+            std::env::var("GITFORGE_PUBLIC_TELEMETRY").ok().as_deref(),
+        );
+        let metrics_router = Router::new().route("/metrics", get(metrics_handler));
         let health = Router::new().route("/health", get(health_check));
-        let metrics = Router::new().route("/metrics", get(metrics_handler));
         let swagger = Router::new()
             .route("/swagger-ui", get(crate::openapi::swagger_ui))
             .route("/api-docs/openapi.json", get(crate::openapi::openapi_spec));
-        let dashboard = crate::dashboard::dashboard_routes();
 
         let mut public_routes = Router::new();
         public_routes = public_routes.merge(health);
-        public_routes = public_routes.merge(metrics);
-        public_routes = public_routes.merge(dashboard);
         public_routes = public_routes.merge(swagger);
+        if telemetry_public {
+            let dashboard = crate::dashboard::dashboard_routes();
+            public_routes = public_routes.merge(metrics_router.clone()).merge(dashboard);
+        }
         // Runner registration is unauthenticated by design (the runner
         // proves itself with its registration token inside the handler), so
         // it is the other public write surface and gets its own rate-limit
@@ -159,7 +165,8 @@ impl ApiServer {
             .merge(ssh_key_routes())
             .merge(artifact_routes())
             .merge(webhook_routes())
-            .merge(review_routes())
+            .merge(review_routes());
+        let protected_routes = protected_routes
             // Authenticate once at the protected-route boundary. Individual
             // handlers may still apply resource authorization, but token
             // parsing and validation must not depend on every handler
@@ -168,16 +175,30 @@ impl ApiServer {
             .layer(Extension(Arc::new(auth.clone())))
             .layer(Extension(pool_arc.clone()));
 
-        // Metrics layer for automatic request recording
+        // Telemetry surfaces stay at top level either way; when the
+        // operator has not opted into public access they are merged in
+        // behind their own auth-middleware layer (F5). They cannot ride
+        // the /api-nested protected router — nesting would move them to
+        // /api/metrics and leave the canonical paths unrouted.
         let metrics_layer = MetricsLayer::new(metrics_arc.clone());
-
         let app = public_routes
             .layer(cors)
             .layer(metrics_layer)
             .layer(Extension(pool_arc.clone()))
             .layer(Extension(metrics_arc))
             .merge(auth_routes)
             .nest("/api", protected_routes);
+        let app = if telemetry_public {
+            app
+        } else {
+            let dashboard = crate::dashboard::dashboard_routes();
+            let telemetry = metrics_router
+                .merge(dashboard)
+                .layer(middleware::from_fn(crate::middleware::auth_middleware))
+                .layer(Extension(Arc::new(auth.clone())))
+                .layer(Extension(pool_arc));
+            app.merge(telemetry)
+        };
 
         Self {
             router: app,

crates/gitforge-api/tests/integration.rs

@@ -26,6 +26,12 @@ use gitforge_storage::{Artifact, ArtifactStore};
 use std::sync::Arc;
 use tower::ServiceExt;
 
+/// Serializes the tests that mutate GITFORGE_PUBLIC_TELEMETRY. The
+/// decision is read at router-build time from the process env, which is
+/// shared by every test thread in this binary — without the lock the
+/// default-posture tests race the opt-in test and randomly invert.
+static TELEMETRY_ENV_LOCK: std::sync::Mutex<()> = std::sync::Mutex::new(());
+
 async fn webhook_fixture(
     config: serde_json::Value,
 ) -> (
@@ -142,9 +148,16 @@ async fn test_health_check_endpoint() {
 
 #[tokio::test]
 async fn test_metrics_endpoint() {
+    // Opt-in telemetry: with GITFORGE_PUBLIC_TELEMETRY unset, /metrics is
+    // served behind the auth middleware (F5) and must reject anonymous
+    // access. The wiring boundary (route placement, not just the
+    // middleware decision) is what this test pins.
     let pool = Pool::memory().await.unwrap();
+    let _guard = TELEMETRY_ENV_LOCK.lock().unwrap();
+    unsafe { std::env::remove_var("GITFORGE_PUBLIC_TELEMETRY") };
     let server = ApiServer::new("test-secret", pool);
     let app = server.into_router();
+    drop(_guard);
 
     // Request metrics
     let response = app
@@ -157,7 +170,7 @@ async fn test_metrics_endpoint() {
         .await
         .unwrap();
 
-    assert_eq!(response.status(), StatusCode::OK);
+    assert_eq!(response.status(), StatusCode::UNAUTHORIZED);
 }
 
 #[tokio::test]
@@ -675,9 +688,36 @@ async fn test_delete_nonexistent_repo() {
 
 #[tokio::test]
 async fn test_dashboard_endpoint() {
+    // Default posture: authenticated only (F5 wiring boundary).
+    let pool = Pool::memory().await.unwrap();
+    let _guard = TELEMETRY_ENV_LOCK.lock().unwrap();
+    unsafe { std::env::remove_var("GITFORGE_PUBLIC_TELEMETRY") };
+    let server = ApiServer::new("test-secret", pool);
+    let app = server.into_router();
+    drop(_guard);
+
+    let response = app
+        .oneshot(
+            Request::builder()
+                .uri("/dashboard")
+                .body(Body::empty())
+                .unwrap(),
+        )
+        .await
+        .unwrap();
+
+    assert_eq!(response.status(), StatusCode::UNAUTHORIZED);
+}
+
+#[tokio::test]
+async fn test_dashboard_endpoint_public_opt_in() {
+    // Operator opt-in keeps both telemetry surfaces public.
     let pool = Pool::memory().await.unwrap();
+    let _guard = TELEMETRY_ENV_LOCK.lock().unwrap();
+    unsafe { std::env::set_var("GITFORGE_PUBLIC_TELEMETRY", "1") };
     let server = ApiServer::new("test-secret", pool);
     let app = server.into_router();
+    drop(_guard);
 
     let response = app
         .oneshot(
@@ -690,6 +730,7 @@ async fn test_dashboard_endpoint() {
         .unwrap();
 
     assert_eq!(response.status(), StatusCode::OK);
+    unsafe { std::env::remove_var("GITFORGE_PUBLIC_TELEMETRY") };
 }
 
 #[tokio::test]