fix(models): handle watcher errors, close on shutdown, rewarm after invalidate

sjf · sjf · commit fc49e5914731 · 2026-05-21T21:05:07.000-07:00
Addresses three ClawSweeper findings on the fs-watcher commit:

- [P1] auth-profile watcher now handles chokidar 'error' events (logs +
  closes once) mirroring the gateway config-reload pattern. Without
  this, an unhandled error from chokidar can crash the gateway.

- [P2] auth-profile watcher handle is pushed into postReadySidecars so
  stopPostReadySidecarsAfterCloseStarted closes it on gateway shutdown.

- [P2] auth-failure and file-change invalidation paths now schedule a
  background rewarm (with a 'reason=' log line). Without this, the next
  /models call after an invalidation paid the slow per-provider path
  until the next reload. The warmer's existing generation counter
  handles concurrent rewarms safely.
diff --git a/src/agents/auth-profiles-watcher.ts b/src/agents/auth-profiles-watcher.ts
@@ -13,9 +13,14 @@ export type AuthProfilesWatcherHandle = {
   stop: () => Promise<void>;
 };
 
+type WatcherLog = {
+  warn: (msg: string) => void;
+};
+
 export function watchAuthProfilesForChanges(params: {
   cfg: OpenClawConfig;
   onChange: () => void;
+  log?: WatcherLog;
 }): AuthProfilesWatcherHandle {
   const watchPaths = listAgentIds(params.cfg).map((agentId) =>
     path.join(resolveAgentDir(params.cfg, agentId), "auth-profiles.json"),
@@ -25,16 +30,26 @@ export function watchAuthProfilesForChanges(params: {
     awaitWriteFinish: { stabilityThreshold: 200, pollInterval: 50 },
     usePolling: Boolean(process.env.VITEST),
   });
+  let closed = false;
   watcher.on("all", () => {
     try {
       params.onChange();
     } catch {
       // onChange errors must not crash the watcher.
     }
   });
+  watcher.on("error", (err) => {
+    if (closed) {
+      return;
+    }
+    closed = true;
+    params.log?.warn(`auth-profile watcher error: ${String(err)}`);
+    void watcher.close().catch(() => {});
+  });
   return {
     stop: async () => {
-      await watcher.close();
+      closed = true;
+      await watcher.close().catch(() => {});
     },
   };
 }
diff --git a/src/gateway/server-startup-post-attach.ts b/src/gateway/server-startup-post-attach.ts
@@ -1019,18 +1019,42 @@ export async function startGatewayPostAttachRuntime(
     });
 
   void sidecarsPromise
-    .then(async () => {
+    .then(async (sidecarsResult) => {
       if (params.minimalTestGateway) {
         return;
       }
       const { clearCurrentProviderAuthState, warmCurrentProviderAuthState } =
         await import("../agents/model-provider-auth.js");
       const { setAuthProfileFailureHook } = await import("../agents/auth-profiles.js");
       const { watchAuthProfilesForChanges } = await import("../agents/auth-profiles-watcher.js");
-      setAuthProfileFailureHook(() => clearCurrentProviderAuthState());
-      watchAuthProfilesForChanges({
+      const scheduleAuthMapRewarm = (reason: string) => {
+        const startMs = Date.now();
+        void warmCurrentProviderAuthState(params.cfgAtStart)
+          .then(() => {
+            params.log.info(
+              `provider auth state re-warmed (${reason}) in ${Date.now() - startMs}ms`,
+            );
+          })
+          .catch((err) => {
+            params.log.warn(`provider auth state rewarm failed: ${String(err)}`);
+          });
+      };
+      setAuthProfileFailureHook(() => {
+        clearCurrentProviderAuthState();
+        scheduleAuthMapRewarm("auth-profile-failure");
+      });
+      const authProfilesWatcher = watchAuthProfilesForChanges({
         cfg: params.cfgAtStart,
-        onChange: () => clearCurrentProviderAuthState(),
+        onChange: () => {
+          clearCurrentProviderAuthState();
+          scheduleAuthMapRewarm("auth-profiles.json change");
+        },
+        log: params.log,
+      });
+      sidecarsResult.postReadySidecars.push({
+        stop: () => {
+          void authProfilesWatcher.stop();
+        },
       });
       const startMs = Date.now();
       await warmCurrentProviderAuthState(params.cfgAtStart);
