Discord gateway websocket flapping on WSL2 causes silent inbound message loss

[karkarl]

Title: Discord gateway websocket flapping on WSL2 causes silent inbound message loss

Environment:

• Host: WSL2 (Windows), KAREN-DEVBOX
• OS: Linux 6.18.33.2-microsoft-standard-WSL2 (x64)
• OpenClaw: CLI & Gateway 2026.6.10
• Node: v22.22.0
• Channel: Discord (direct messages)

Summary:
On WSL2, the Discord gateway websocket disconnects abnormally every ~3–5 minutes. Messages (including text + photo attachments) sent during a drop/reconnect window are silently lost — never delivered to the agent session, no error surfaced to the user.

Impact:
User sends messages that appear to send on the Discord client side but never reach the bot. No error, no retry, no notice. Confirmed via sessions_history that dropped messages have zero trace in the session record. Happened 3+ times in one session, including a batch with photo attachments.

Evidence (single day, /tmp/openclaw/openclaw-2026-07-14.log):

• 66 total websocket closures
• Breakdown: 38× code 1006 (abnormal), 4× 1000, 2× 1001
• Zero RESUME/READY/reconnect-success markers logged
• Closures roughly every 3–5 min, all day, e.g. 08:05, 08:09, 08:18, 08:22, 08:26, 08:28, 08:34…

Suspected root cause:
WSL2's virtual NIC/NAT layer drops long-lived idle connections; Discord's persistent gateway websocket is killed repeatedly. On reconnect the session appears to re-IDENTIFY rather than cleanly RESUME (no RESUME markers observed), so events missed during the gap are not replayed → permanent message loss.

Expected behavior:

• On gateway reconnect, the client should RESUME and replay missed events where possible.
• Inbound messages that fail to reach the agent should surface an error/retry rather than silently vanish.
• Consider TCP keepalive defaults / connection-health handling tuned for NAT'd environments like WSL2.

Workarounds under investigation:

• WSL2 TCP keepalive sysctls (tcp_keepalive_time, etc.)
• WSL2 networkingMode=mirrored
• wsl --update