fix(ai): avoid stack overflow saving base64 request records (#7280)

packages/service/core/ai/record/controller.ts

@@ -8,21 +8,33 @@ export const createLLMRequestId = () => {
 };
 };
 
 
 const base64OmittedPlaceholder = '[base64 omitted]';
 const base64OmittedPlaceholder = '[base64 omitted]';
-const dataUrlBase64Regex = /data:([^,]*;base64,)([A-Za-z0-9+/=_-]{32,})/g;
-const fullBase64Regex = /^[A-Za-z0-9+/=_-]+$/;
 const base64LikeKeySet = new Set(['base64', 'data']);
 const base64LikeKeySet = new Set(['base64', 'data']);
+const fullBase64Regex = /^[A-Za-z0-9+/=_-]+$/;
+const dataUrlPrefix = 'data:';
+const base64UrlMarker = ';base64,';
+const rawBase64OmitLength = 256;
 
 
 const isBase64LikeString = (value: string) => {
 const isBase64LikeString = (value: string) => {
   const text = value.trim();
   const text = value.trim();
 
 
-  return text.length >= 256 && fullBase64Regex.test(text);
+  return text.length >= rawBase64OmitLength && fullBase64Regex.test(text);
+};
+
+/**
+ * 清洗整个字段值形式的 data URL。
+ * 大视频会生成超长 data URL，使用正则 replace 容易触发 V8 栈溢出。
+ */
+const sanitizeDataUrlBase64 = (value: string) => {
+  if (!value.startsWith(dataUrlPrefix)) return value;
+
+  const markerIndex = value.indexOf(base64UrlMarker);
+  if (markerIndex === -1) return value;
+
+  return `${value.slice(0, markerIndex + base64UrlMarker.length)}${base64OmittedPlaceholder}`;
 };
 };
 
 
 const sanitizeString = (value: string, key?: string) => {
 const sanitizeString = (value: string, key?: string) => {
-  const sanitizedDataUrl = value.replace(
-    dataUrlBase64Regex,
-    (_match, prefix: string) => `data:${prefix}${base64OmittedPlaceholder}`
-  );
+  const sanitizedDataUrl = sanitizeDataUrlBase64(value);
 
 
   if (base64LikeKeySet.has(key?.toLowerCase() ?? '') && isBase64LikeString(sanitizedDataUrl)) {
   if (base64LikeKeySet.has(key?.toLowerCase() ?? '') && isBase64LikeString(sanitizedDataUrl)) {
     return base64OmittedPlaceholder;
     return base64OmittedPlaceholder;

packages/service/test/core/ai/record/controller.test.ts

@@ -72,19 +72,42 @@ describe('sanitizeLLMRequestRecordPayload', () => {
       text: 'plain text should stay'
       text: 'plain text should stay'
     });
     });
   });
   });
+});
 
 
-  it('redacts base64 embedded in a longer string', () => {
-    const payload = {
-      markdown: `![image](data:image/jpeg;base64,${createBase64()})`
-    };
+describe('LLM request record team isolation', () => {
+  it('saves records with very large video data urls after redacting base64 payloads', async () => {
+    const largeVideoBase64 = createBase64(10_000_000);
 
 
-    expect(sanitizeLLMRequestRecordPayload(payload)).toEqual({
-      markdown: '![image](data:image/jpeg;base64,[base64 omitted])'
+    await saveLLMRequestRecord({
+      teamId: '507f1f77bcf86cd799439011',
+      requestId: 'large_video_request',
+      body: {
+        messages: [
+          {
+            role: 'user',
+            content: [
+              { type: 'text', text: 'analyze this video' },
+              {
+                type: 'video_url',
+                video_url: {
+                  url: `data:video/mp4;base64,${largeVideoBase64}`
+                }
+              }
+            ]
+          }
+        ]
+      },
+      response: { answerText: 'done' }
     });
     });
+
+    const record = await getLLMRequestRecord('large_video_request', '507f1f77bcf86cd799439011');
+
+    expect(record?.body.messages[0].content[1].video_url.url).toBe(
+      'data:video/mp4;base64,[base64 omitted]'
+    );
+    expect(record?.response).toEqual({ answerText: 'done' });
   });
   });
-});
 
 
-describe('LLM request record team isolation', () => {
   it('saves records with teamId and only reads them from the same team', async () => {
   it('saves records with teamId and only reads them from the same team', async () => {
     await saveLLMRequestRecord({
     await saveLLMRequestRecord({
       teamId: '507f1f77bcf86cd799439011',
       teamId: '507f1f77bcf86cd799439011',