Bug type

Regression (worked before, now fails)

Summary

After upgrading from v2026.3.11 to v2026.3.24, the gateway crashes every ~35 minutes due to Discord's health-monitor detecting stale sockets and triggering a reconnection path that throws an uncaught exception. Zero crashes occurred across 5+ days on v2026.3.11. On v2026.3.24, 16 crashes occurred in a single day.

Steps to reproduce

1. Install v2026.3.24 with Discord channel enabled (single guild, allowlist-only)
2. Gateway runs normally for ~30-35 minutes
3. Discord health-monitor detects a stale WebSocket (no events within staleSocketMinutes, default 30)
4. Health-monitor calls stopChannel() → triggers onAbort()
5. onAbort() sets gateway.options.reconnect = { maxAttempts: 0 } then calls gateway.disconnect()
6. WebSocket closes with code 1005 ("No Status Received")
7. handleClose(1005) → handleReconnectionAttempt() → checks reconnectAttempts(0) >= maxAttempts(0) → true
8. Emits new Error("Max reconnect attempts (0) reached after code 1005")
9. Error is uncaught → entire Node.js process crashes
10. systemd restarts → cycle repeats every ~35 minutes

Expected behavior

Health-monitor should gracefully restart the Discord channel without crashing the gateway process.

Actual behavior

The onAbort handler sets maxAttempts: 0 before disconnecting. The WebSocket close handler then fires and immediately triggers the max-attempts error path (0 >= 0 is true), emitting an uncaught exception that crashes the entire Node.js process.

OpenClaw version

2026.3.24 (upgraded from 2026.3.11)

Operating system

Ubuntu 24.04 LTS (Linux 6.18.7 x64)

Install method

npm global

Model

anthropic/claude-opus-4-6 / anthropic/claude-sonnet-4-6

Provider / routing chain

openclaw -> anthropic (direct)

Additional provider/model setup details

Bug is in Discord WebSocket lifecycle management, not model-specific.

Logs, screenshots, and evidence

[health-monitor] [discord:default] health-monitor: restarting (reason: stale-socket)
[openclaw] Uncaught exception: Error: Max reconnect attempts (0) reached after code 1005
    at SafeGatewayPlugin.handleReconnectionAttempt (provider-CAlWEl41.js:3318:47)
    at SafeGatewayPlugin.handleClose (provider-CAlWEl41.js:3364:8)
    at WebSocket.<anonymous> (provider-CAlWEl41.js:3307:9)

Root cause in provider-CAlWEl41.js:

Line 6952 — onAbort sets: gateway.options.reconnect = { maxAttempts: 0 };

Lines 3316-3318 — Reconnection handler checks:

const { maxAttempts = 5 } = this.options.reconnect ?? {};
if (this.reconnectAttempts >= maxAttempts) {
    this.emitter.emit("error", new Error(`Max reconnect attempts (${maxAttempts}) reached...`));

Crash frequency data:

• Mar 17-24 (v2026.3.11): 0 crashes across 5+ days
• Mar 25 (v2026.3.24): 16 crashes in one day, every ~35 min

Impact and severity

High — Gateway crashes every ~35 minutes. All running subagent sessions are disrupted or killed. Subagent completion announce-back fails after restart ("Outbound not configured for channel: telegram"). Long-running subagent tasks (30-75 min) have near-zero chance of completing.

Additional information

Suggested fixes:

1. (Preferred) Set a flag to suppress the close handler rather than manipulating maxAttempts — lifecycleStopping already exists on line 6944, add a check in handleClose
2. Set maxAttempts to a sentinel value that handleReconnectionAttempt treats as "intentional shutdown, don't emit error"
3. Catch the error in the health-monitor's restart flow so it doesn't propagate as uncaught

Workaround: Disable Discord (channels.discord.enabled: false).

Note: Also observed a secondary issue — with Discord channel disabled but Discord plugin still enabled (plugins.entries.discord.enabled: true), message-action-discovery still tries to resolve the Discord token SecretRef, causing a separate crash ("Unhandled promise rejection: channels.discord.token: unresolved SecretRef"). Both the channel AND plugin must be disabled as workaround.