openclaw - 💡(How to fix) Fix Bug: inbound Discord messages bypass RECENT_QUEUE_MESSAGE_IDS dedup cache [1 participants]

Summary

Normal inbound Discord messages are never checked against the RECENT_QUEUE_MESSAGE_IDS dedup cache, allowing the same message to be processed multiple times (e.g. after a gateway restart or on Discord RESUME replay).

Root cause

There are two distinct dispatch paths for Discord messages:

Path A — Follow-up / queue turns (via enqueueFollowupRun):

runReplyAgent → enqueueFollowupRun → RECENT_QUEUE_MESSAGE_IDS.peek() ✅

Path B — Normal inbound messages (every message a user sends):

Discord event → debouncer.enqueue() → onFlush() → preflightDiscordMessage() → inboundWorker.enqueue(buildDiscordInboundJob(ctx))

Path B has no dedup check at any point. RECENT_QUEUE_MESSAGE_IDS is only consulted inside enqueueFollowupRun (src/auto-reply/reply/queue/enqueue.ts), which Path B never calls.

Evidence

From discord-CcCLMjHw.js (compiled dist):

• RECENT_QUEUE_MESSAGE_IDS defined at line 78640 with TTL=300s, maxSize=10000
• Checked at lines 78667/78681 inside enqueueFollowupRun only
• preflightDiscordMessage (line 132019): no dedup check before returning ctx
• createDiscordInboundWorker.enqueue() (lines 132592, 132626): jobs enqueued with no prior cache lookup
• message.id (Discord snowflake) is available in preflightDiscordMessage context throughout

Impact

• On Discord RESUME events, recent messages are replayed by the Discord gateway with the same snowflake IDs — Path B processes them again
• After a gateway restart (which clears the in-process cache), the same message can be dispatched twice if it was in-flight at restart
• Under any condition where the Discord listener fires the same event more than once, the message will be processed without duplicate suppression

Expected behavior

preflightDiscordMessage (or the onFlush handler in the debouncer) should check RECENT_QUEUE_MESSAGE_IDS by message.id before proceeding. If the message ID is already in the cache, return null (drop). If not, mark it before handing off to inboundWorker.enqueue.

Suggested fix location

src/discord/monitor/message-handler.process.ts (or equivalent pre-compiled source for preflightDiscordMessage):

// Near top of preflightDiscordMessage, after resolving messageChannelId:
const dedupKey = buildRecentMessageIdKey({ messageId: message.id, originatingChannel: discord, originatingAccountId: params.accountId, ... });
if (dedupKey && RECENT_QUEUE_MESSAGE_IDS.peek(dedupKey)) {
  logVerbose(`discord: drop duplicate inbound message ${message.id}`);
  return null;
}
// ... rest of preflight ...
// Before returning ctx:
if (dedupKey) RECENT_QUEUE_MESSAGE_IDS.check(dedupKey);
return ctx;

Alternatively, the check could be placed in the debouncer onFlush handler before calling preflightDiscordMessage.

Environment

• OpenClaw version: checked against installed dist in /usr/lib/node_modules/openclaw/dist/discord-CcCLMjHw.js
• Analyzed via static grep of compiled bundle
• Confirmed: processDiscordMessage is only ever reached via processDiscordInboundJob → createDiscordInboundWorker.enqueue → the Path B chain above