Add Retry to Stream Errors

packages/backend/package.json

@@ -40,7 +40,7 @@
     "moment": "^2.24.0",
     "mysql": "^2.17.1",
     "node-cron": "^4.2.1",
-    "openai": "^4.103.0",
+    "openai": "^6.44.0",
     "prom-client": "^15.1.3",
     "sentence-splitter": "^5.0.0",
     "sentiment": "^5.0.2",

packages/backend/src/lib/resilientOpenAIClient.spec.ts

@@ -129,6 +129,28 @@ describe('ResilientOpenAIClient', () => {
       expect(result).toEqual(makeResponse('ok after 429'));
     });
 
+    it('retries on FetchError with Premature close and succeeds', async () => {
+      const transientFetchError = Object.assign(
+        new Error('Invalid response body while trying to fetch https://api.openai.com/v1/responses: Premature close'),
+        {
+          name: 'FetchError',
+        },
+      );
+      const createMock = vi
+        .fn()
+        .mockRejectedValueOnce(transientFetchError)
+        .mockResolvedValue(makeResponse('ok after premature close'));
+
+      const client = new ResilientOpenAIClient(makeUnderlying(createMock), fastConfig(), registry);
+
+      const resultPromise = client.responses.create({ model: 'gpt-4o', input: 'hi' });
+      await vi.runAllTimersAsync();
+      const result = await resultPromise;
+
+      expect(createMock).toHaveBeenCalledTimes(2);
+      expect(result).toEqual(makeResponse('ok after premature close'));
+    });
+
     it('honors Retry-After header on 429 response', async () => {
       const retryAfterSeconds = 2;
       const rateLimitError = Object.assign(new Error('Rate limited'), {

packages/backend/src/lib/resilientOpenAIClient.ts

@@ -360,33 +360,22 @@ export class ResilientOpenAIClient implements OpenAIClientLike {
   // ---------------------------------------------------------------------------
 
   private isRetriable(error: unknown): boolean {
-    if (!(error instanceof Error)) {
-      return false;
-    }
-
     // Timeout errors are retriable
     if (error instanceof ResilientOpenAIError && error.code === 'TIMEOUT') {
       return true;
     }
 
-    // Network / connection errors
-    if (
-      error.message.includes('ECONNRESET') ||
-      error.message.includes('ECONNREFUSED') ||
-      error.message.includes('ETIMEDOUT') ||
-      error.message.includes('socket hang up') ||
-      error.message.includes('fetch failed') ||
-      error.message.includes('network')
-    ) {
-      return true;
-    }
-
     // OpenAI HTTP status codes: 429 (rate-limit) and 5xx are retriable
     const statusCode = this.getStatusCode(error);
     if (statusCode === 429 || (statusCode !== undefined && statusCode >= 500)) {
       return true;
     }
 
+    // Network / transport failures are typically transient and should be retried.
+    if (this.hasRetriableNetworkSignal(error)) {
+      return true;
+    }
+
     return false;
   }
 
@@ -398,6 +387,51 @@ export class ResilientOpenAIClient implements OpenAIClientLike {
     return undefined;
   }
 
+  private hasRetriableNetworkSignal(error: unknown): boolean {
+    const queue: unknown[] = [error];
+    const visited = new Set<unknown>();
+
+    while (queue.length > 0) {
+      const current = queue.shift();
+      if (typeof current !== 'object' || current === null || visited.has(current)) {
+        continue;
+      }
+
+      visited.add(current);
+
+      const message = Reflect.get(current, 'message');
+      const name = Reflect.get(current, 'name');
+      const code = Reflect.get(current, 'code');
+
+      const normalizedSignals = [message, name, code]
+        .filter((value): value is string => typeof value === 'string')
+        .map((value) => value.toLowerCase())
+        .join(' ');
+
+      if (
+        normalizedSignals.includes('econnreset') ||
+        normalizedSignals.includes('econnrefused') ||
+        normalizedSignals.includes('etimedout') ||
+        normalizedSignals.includes('socket hang up') ||
+        normalizedSignals.includes('fetch failed') ||
+        normalizedSignals.includes('fetcherror') ||
+        normalizedSignals.includes('premature close') ||
+        normalizedSignals.includes('und_err_socket') ||
+        normalizedSignals.includes('network error') ||
+        normalizedSignals.includes('networkerror')
+      ) {
+        return true;
+      }
+
+      const cause = Reflect.get(current, 'cause');
+      if (cause !== undefined) {
+        queue.push(cause);
+      }
+    }
+
+    return false;
+  }
+
   // ---------------------------------------------------------------------------
   // Backoff with full jitter; respects Retry-After header on 429
   // ---------------------------------------------------------------------------