GitHub Issue Report: Gateway Socket Connection Leak

Bug Title

Gateway WebSocket connections leak - CLOSE_WAIT/FIN_WAIT_2 zombie connections accumulate and cause crashes

Environment

• OpenClaw version: 2026.3.24
• Node.js version: v24.14.0
• Platform: Windows 10 (x64)
• Gateway port: 18789

Problem Description

The Gateway process accumulates zombie socket connections in CLOSE_WAIT and FIN_WAIT_2 states over time. These connections are not properly closed by the Gateway, leading to resource leak and eventual crashes.

Evidence

Connection State Analysis (observed on 2026-03-28)

Zombie connection ratio: 66% of non-listening connections

Crash Pattern

Gateway crashes at irregular intervals (1-6 hours), requiring manual or automated restart.

Crash times observed on 2026-03-28:

• 00:28, 01:22-01:23, 07:53, 09:00, 10:09, 13:06

Memory Behavior

• Gateway memory fluctuates between 600-1500 MB
• After restart: ~600 MB
• After 1-2 hours: ~800-1400 MB
• Not a memory leak per se, but connection accumulation correlates with crashes

Root Cause Analysis

This is a socket connection management bug in the Gateway:

1. WebSocket clients (main OpenClaw process, PID 1424) request connection close
2. Gateway WebSocket handler does not properly respond to close events
3. Connections remain in CLOSE_WAIT/FIN_WAIT_2 states
4. Zombie connections accumulate over time
5. Eventually causes Gateway instability and crashes

Impact

• Gateway crashes unpredictably (1-6 hour intervals)
• Requires external watchdog process to restart
• Users experience service interruption until restart completes

Reproduction Steps

1. Start OpenClaw Gateway: openclaw gateway --port 18789
2. Connect multiple clients (Discord, Feishu, etc.)
3. Monitor connections: netstat -ano | findstr 18789
4. Observe CLOSE_WAIT/FIN_WAIT_2 connections accumulating over time
5. Gateway eventually crashes (no error output to stderr)

Expected Behavior

Gateway should properly close WebSocket connections when:

• Client requests close
• Connection timeout occurs
• Keep-alive fails

Workaround (Implemented)

1. Gateway Guardian watchdog process (auto-restart within 30 seconds)
2. Daily scheduled restart at 04:00 AM to clear zombie connections
3. Error logging via stderr redirect: gateway.cmd 2>> gateway-error.log

Suggested Fix

Review WebSocket connection lifecycle handling in:

• gateway-runtime-DkLKThnW.js
• server-C8VdPOMv.js
• WebSocket close event handlers

Ensure proper socket cleanup on:

• Client-initiated close
• Server-initiated close
• Connection timeout/keep-alive failure

Additional Information

• No error output captured in stderr during crashes
• System has ample memory (16GB, 7GB free), not memory exhaustion
• Node.js heap limit: 4.28GB (not exceeded)

Reporter: OpenClaw user
Date: 2026-03-28