From 9e4aad973f87e83b26ae4d25f269f72e04a08b8c Mon Sep 17 00:00:00 2001
From: ValueOn AG <p.motsch@valueon.ch>
Date: Thu, 19 Feb 2026 00:46:41 +0100
Subject: [PATCH] fix: WS auto-reconnect, reduce keepalive to 15s, downgrade
 ChatDOM to debug

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 src/bot/chatProcedure.ts |  2 +-
 src/bot/orchestrator.ts  | 56 ++++++++++++++++++++++++++++++++--------
 2 files changed, 46 insertions(+), 12 deletions(-)

diff --git a/src/bot/chatProcedure.ts b/src/bot/chatProcedure.ts
index 1e8e8a4..44c44f5 100644
--- a/src/bot/chatProcedure.ts
+++ b/src/bot/chatProcedure.ts
@@ -134,7 +134,7 @@ export class ChatProcedure {
         children: number;
         html: string;
       }) => {
-        this._logger.info(`ChatDOM: <${info.tag} data-tid="${info.tid}"> children=${info.children}, text="${info.text.substring(0, 120)}"`);
+        this._logger.debug(`ChatDOM: <${info.tag} data-tid="${info.tid}"> children=${info.children}, text="${info.text.substring(0, 120)}"`);
       });
     } catch {
       // Already exposed
diff --git a/src/bot/orchestrator.ts b/src/bot/orchestrator.ts
index a1b725c..dd47c3e 100644
--- a/src/bot/orchestrator.ts
+++ b/src/bot/orchestrator.ts
@@ -511,9 +511,9 @@ export class BotOrchestrator {
           // Connection might be closing
         }
       }
-    }, 30000);
+    }, 15000);
 
-    this._logger.info('Keepalive started (30s interval)');
+    this._logger.info('Keepalive started (15s interval)');
   }
 
   /**
@@ -527,22 +527,32 @@ export class BotOrchestrator {
     }
   }
 
+  private _wsReconnectAttempts: number = 0;
+  private _wsMaxReconnectAttempts: number = 10;
+  private _wsReconnecting: boolean = false;
+
   /**
    * Connect to the Gateway WebSocket for this session.
    */
   private async _connectToGateway(): Promise<void> {
-    // gatewayWsUrl is the full WebSocket URL provided by the Gateway
-    // It already includes instanceId and sessionId
     const wsUrl = this._options.gatewayWsUrl;
     this._logger.info(`Connecting to Gateway: ${wsUrl}`);
 
-    // Derive HTTP base URL from WebSocket URL for fallback
     this._httpBaseUrl = wsUrl
       .replace('wss://', 'https://')
       .replace('ws://', 'http://')
       .replace(/\/bot\/ws\/.*$/, '');
 
-    return new Promise((resolve, reject) => {
+    return this._createWsConnection(wsUrl, true);
+  }
+
+  /**
+   * Create (or recreate) the WebSocket connection.
+   * On initial connect, `isInitial` = true and the promise resolves/rejects.
+   * On reconnect, the promise resolves immediately (fire-and-forget).
+   */
+  private _createWsConnection(wsUrl: string, isInitial: boolean): Promise<void> {
+    return new Promise((resolve) => {
       this._gatewayWs = new WebSocket(wsUrl);
 
       const wsTimeout = setTimeout(() => {
@@ -551,7 +561,7 @@ export class BotOrchestrator {
           this._useHttpFallback = true;
           this._gatewayWs?.close();
           this._gatewayWs = null;
-          resolve(); // Continue with HTTP fallback instead of failing
+          resolve();
         }
       }, 10000);
 
@@ -559,6 +569,8 @@ export class BotOrchestrator {
         clearTimeout(wsTimeout);
         this._logger.info('Connected to Gateway via WebSocket');
         this._useHttpFallback = false;
+        this._wsReconnectAttempts = 0;
+        this._wsReconnecting = false;
         resolve();
       });
 
@@ -570,23 +582,45 @@ export class BotOrchestrator {
 
       this._gatewayWs.on('close', (code, reason) => {
         this._logger.warn(`Gateway WebSocket closed: ${code} - ${reason}`);
-        if (!this._isShuttingDown && !this._useHttpFallback) {
-          this._logger.info('Switching to HTTP fallback for transcript delivery');
+        if (!this._isShuttingDown) {
           this._useHttpFallback = true;
+          this._scheduleReconnect(wsUrl);
         }
       });
 
       this._gatewayWs.on('error', (error) => {
         clearTimeout(wsTimeout);
         this._logger.error('Gateway WebSocket error:', error);
-        this._logger.info('Switching to HTTP fallback for transcript delivery');
         this._useHttpFallback = true;
         this._gatewayWs = null;
-        resolve(); // Continue with HTTP fallback
+        if (isInitial) resolve();
       });
     });
   }
 
+  /**
+   * Schedule a WebSocket reconnection with exponential backoff.
+   */
+  private _scheduleReconnect(wsUrl: string): void {
+    if (this._isShuttingDown || this._wsReconnecting) return;
+    if (this._wsReconnectAttempts >= this._wsMaxReconnectAttempts) {
+      this._logger.warn(`WebSocket reconnect limit reached (${this._wsMaxReconnectAttempts}), staying on HTTP fallback`);
+      return;
+    }
+
+    this._wsReconnecting = true;
+    this._wsReconnectAttempts++;
+    const delayMs = Math.min(2000 * Math.pow(1.5, this._wsReconnectAttempts - 1), 30000);
+    this._logger.info(`WebSocket reconnect attempt ${this._wsReconnectAttempts}/${this._wsMaxReconnectAttempts} in ${(delayMs / 1000).toFixed(1)}s`);
+
+    setTimeout(() => {
+      if (this._isShuttingDown) return;
+      this._createWsConnection(wsUrl, false).catch((err) => {
+        this._logger.error('WebSocket reconnect failed:', err);
+      });
+    }, delayMs);
+  }
+
   /**
    * Handle incoming messages from the Gateway.
    * Async operations are awaited to ensure proper error handling