fix: handle Bedrock 503 streaming errors with timeout detection (#92)

- Add 15s streaming timeout to detect mid-stream stalls (e.g., Bedrock 503)
- Add stop() call to allow user retry after timeout
- Add streamingError state for timeout-detected errors
- Improve server-side error logging for empty usage detection
- Add user-friendly error messages for ServiceUnavailable and Throttling errors

app/api/chat/route.ts

@@ -154,9 +154,28 @@ ${lastMessageText}
     messages: allMessages,
     ...(providerOptions && { providerOptions }),
     ...(headers && { headers }),
-    onFinish: ({ usage, providerMetadata }) => {
-      console.log('[Cache] Full providerMetadata:', JSON.stringify(providerMetadata, null, 2));
-      console.log('[Cache] Usage:', JSON.stringify(usage, null, 2));
+    onFinish: ({ usage, providerMetadata, finishReason, text, toolCalls }) => {
+      // Detect potential mid-stream failures (e.g., Bedrock 503 ServiceUnavailableException)
+      // When this happens, usage is empty and providerMetadata is undefined
+      const hasUsage = usage && Object.keys(usage).length > 0;
+      if (!hasUsage) {
+        console.error('[Stream Error] Empty usage detected - possible Bedrock 503 or mid-stream failure');
+        console.error('[Stream Error] finishReason:', finishReason);
+        console.error('[Stream Error] text received:', text?.substring(0, 200) || '(none)');
+        console.error('[Stream Error] toolCalls:', toolCalls?.length || 0);
+        // Log the user's last message for debugging
+        const lastUserMsg = enhancedMessages.filter(m => m.role === 'user').pop();
+        if (lastUserMsg) {
+          const content = lastUserMsg.content;
+          const preview = Array.isArray(content)
+            ? (content.find((c) => c.type === 'text') as { type: 'text'; text: string } | undefined)?.text?.substring(0, 100)
+            : String(content).substring(0, 100);
+          console.error('[Stream Error] Last user message preview:', preview);
+        }
+      } else {
+        console.log('[Cache] Full providerMetadata:', JSON.stringify(providerMetadata, null, 2));
+        console.log('[Cache] Usage:', JSON.stringify(usage, null, 2));
+      }
     },
     tools: {
       // Client-side tool that will be executed on the client
@@ -232,6 +251,23 @@ IMPORTANT: Keep edits concise:
         ? error.message
         : JSON.stringify(error);
 
+    // Check for Bedrock service errors (503, throttling, etc.)
+    if (errorString.includes('ServiceUnavailable') ||
+      errorString.includes('503') ||
+      errorString.includes('temporarily unavailable')) {
+      console.error('[Bedrock Error] ServiceUnavailableException:', errorString);
+      return 'The AI service is temporarily unavailable. Please try again in a few seconds.';
+    }
+
+    // Check for throttling errors
+    if (errorString.includes('ThrottlingException') ||
+      errorString.includes('rate limit') ||
+      errorString.includes('too many requests') ||
+      errorString.includes('429')) {
+      console.error('[Bedrock Error] ThrottlingException:', errorString);
+      return 'Too many requests. Please wait a moment and try again.';
+    }
+
     // Check for image not supported error (e.g., DeepSeek models)
     if (errorString.includes('image_url') ||
       errorString.includes('unknown variant') ||

components/chat-panel.tsx

@@ -61,6 +61,7 @@ export default function ChatPanel({
     const [files, setFiles] = useState<File[]>([]);
     const [showHistory, setShowHistory] = useState(false);
     const [input, setInput] = useState("");
+    const [streamingError, setStreamingError] = useState<Error | null>(null);
 
     // Store XML snapshots for each user message (keyed by message index)
     const xmlSnapshotsRef = useRef<Map<number, string>>(new Map());
@@ -71,7 +72,7 @@ export default function ChatPanel({
         chartXMLRef.current = chartXML;
     }, [chartXML]);
 
-    const { messages, sendMessage, addToolOutput, status, error, setMessages } =
+    const { messages, sendMessage, addToolResult, status, error, setMessages, stop } =
         useChat({
             transport: new DefaultChatTransport({
                 api: "/api/chat",
@@ -83,13 +84,13 @@ export default function ChatPanel({
                     const validationError = validateMxCellStructure(xml);
 
                     if (validationError) {
-                        addToolOutput({
+                        addToolResult({
                             tool: "display_diagram",
                             toolCallId: toolCall.toolCallId,
                             output: validationError,
                         });
                     } else {
-                        addToolOutput({
+                        addToolResult({
                             tool: "display_diagram",
                             toolCallId: toolCall.toolCallId,
                             output: "Successfully displayed the diagram.",
@@ -122,7 +123,7 @@ export default function ChatPanel({
 
                         onDisplayChart(editedXml);
 
-                        addToolOutput({
+                        addToolResult({
                             tool: "edit_diagram",
                             toolCallId: toolCall.toolCallId,
                             output: `Successfully applied ${edits.length} edit(s) to the diagram.`,
@@ -136,7 +137,7 @@ export default function ChatPanel({
                                 ? error.message
                                 : String(error);
 
-                        addToolOutput({
+                        addToolResult({
                             tool: "edit_diagram",
                             toolCallId: toolCall.toolCallId,
                             output: `Edit failed: ${errorMessage}
@@ -153,9 +154,64 @@ Please retry with an adjusted search pattern or use display_diagram if retries a
             },
             onError: (error) => {
                 console.error("Chat error:", error);
+                setStreamingError(error);
             },
         });
 
+    // Streaming timeout detection - detects when stream stalls mid-response (e.g., Bedrock 503)
+    // This catches cases where onError doesn't fire because headers were already sent
+    const lastMessageCountRef = useRef(0);
+    const lastMessagePartsRef = useRef(0);
+
+    useEffect(() => {
+        // Clear streaming error and reset refs when status changes to ready
+        if (status === "ready") {
+            setStreamingError(null);
+            lastMessageCountRef.current = 0;
+            lastMessagePartsRef.current = 0;
+            return;
+        }
+
+        if (status !== "streaming") return;
+
+        const STALL_TIMEOUT_MS = 15000; // 15 seconds without any update
+
+        // Capture current state BEFORE setting timeout
+        // This way we compare against values at the time timeout was set
+        const currentPartsCount = messages.reduce(
+            (acc, msg) => acc + (msg.parts?.length || 0),
+            0
+        );
+        const capturedMessageCount = messages.length;
+        const capturedPartsCount = currentPartsCount;
+
+        // Update refs immediately so next effect run has fresh values
+        lastMessageCountRef.current = messages.length;
+        lastMessagePartsRef.current = currentPartsCount;
+
+        const timeoutId = setTimeout(() => {
+            // Re-count parts at timeout time
+            const newPartsCount = messages.reduce(
+                (acc, msg) => acc + (msg.parts?.length || 0),
+                0
+            );
+
+            // If no change since timeout was set, stream has stalled
+            if (
+                messages.length === capturedMessageCount &&
+                newPartsCount === capturedPartsCount
+            ) {
+                console.error("[Streaming Timeout] No activity for 15s - forcing error state");
+                setStreamingError(
+                    new Error("Connection lost. The AI service may be temporarily unavailable. Please try again.")
+                );
+                stop(); // Allow user to retry by transitioning status to "ready"
+            }
+        }, STALL_TIMEOUT_MS);
+
+        return () => clearTimeout(timeoutId);
+    }, [status, messages, stop]);
+
     const messagesEndRef = useRef<HTMLDivElement>(null);
 
     useEffect(() => {
@@ -167,8 +223,11 @@ Please retry with an adjusted search pattern or use display_diagram if retries a
 
     const onFormSubmit = async (e: React.FormEvent<HTMLFormElement>) => {
         e.preventDefault();
-        const isProcessing = status === "streaming" || status === "submitted";
+        // Allow retry if there's a streaming error (workaround for stop() not transitioning status)
+        const isProcessing = (status === "streaming" || status === "submitted") && !streamingError;
         if (input.trim() && !isProcessing) {
+            // Clear any previous streaming error before starting new request
+            setStreamingError(null);
             try {
                 let chartXml = await onFetchChart();
                 chartXml = formatXML(chartXml);
@@ -415,7 +474,7 @@ Please retry with an adjusted search pattern or use display_diagram if retries a
             <main className="flex-1 overflow-hidden">
                 <ChatMessageDisplay
                     messages={messages}
-                    error={error}
+                    error={error || streamingError}
                     setInput={setInput}
                     setFiles={handleFileChange}
                     onRegenerate={handleRegenerate}