#22143: Fix memory leak in WhatsApp channel reconnection loop

## Summary The openclaw-gateway (Node.js) grows unbounded during WhatsApp channel reconnection cycles — observed 2.3GB to 43.2GB in ~16 hours, consuming 67% of a 64GB box before OOMing with `FATAL ERROR: Reached heap limit Allocation failed - JavaScript heap out of memory`. Three leaks compound during reconnection: **Leak 1: Debouncer never cleaned up on close** (`src/auto-reply/inbound-debounce.ts`) - `createInboundDebouncer()` returns only `{ enqueue, flushKey }` — no teardown - Internal `buffers` Map holds message items with pending `setTimeout` callbacks - Each item captures `sock` WebSocket references via closures (`sendComposing`, `reply`, `sendMedia`) - When `monitorWebInbox.close()` is called on reconnect, these timers and references persist **Leak 2: groupMetaCache never cleared on close** (`src/web/inbound/monitor.ts`) - `groupMetaCache = new Map()` accumulates group metadata entries - Never cleared in the `close()` handler — expired-TTL entries still occupy memory **Leak 3: Outer-scope Maps accumulate across reconnections** (`src/web/auto-reply/monitor.ts`) - `groupHistories` and `groupMemberNames` Maps are created outside the `while(true)` reconnection loop - They grow across every reconnection cycle and are never bounded or cleared ## Changes - **`src/auto-reply/inbound-debounce.ts`**: Add `flushAll()` and `close()` methods to `createInboundDebouncer` return value. `close()` cancels all pending timers and drops buffer references. - **`src/web/inbound/monitor.ts`**: Call `debouncer.close()` and `groupMetaCache.clear()` in the `monitorWebInbox` close handler before closing the WebSocket. - **`src/web/auto-reply/monitor.ts`**: Clear `groupHistories` and `groupMemberNames` at the top of each reconnection loop iteration. ## Test plan - [ ] Verify gateway starts without errors after applying changes - [ ] Monitor `process.memoryUsage().heapUsed` across multiple reconnection cycles — should drop back after each reconnect - [ ] Confirm no `FATAL ERROR: Reached heap limit` crashes over 24+ hours of operation - [ ] Run existing test suite (`pnpm test`) <!-- greptile_comment --> <h3>Greptile Summary</h3> Fixes three distinct memory leaks in WhatsApp channel reconnection that caused unbounded heap growth (2.3GB to 43.2GB in 16 hours): - **Debouncer cleanup**: Added `close()` method to cancel pending timers and clear buffer references that captured WebSocket closures (`sendComposing`, `reply`, `sendMedia`) - **Group metadata cache**: Clear `groupMetaCache` Map on connection close to release accumulated group metadata entries - **Reconnection loop state**: Clear `groupHistories` and `groupMemberNames` Maps at the start of each reconnection cycle All three fixes correctly address the root causes described in the PR. The debouncer's `close()` method appropriately drops buffered messages during reconnection rather than attempting to flush them with a stale WebSocket connection. The `flushAll()` method was added but isn't currently used in the close path - this appears intentional as flushing would fail with a closed socket. <h3>Confidence Score: 4/5</h3> - This PR is safe to merge with low risk - fixes critical memory leaks without introducing new behavior - The fixes correctly address three specific memory leaks with surgical changes. The implementation is sound: timers are properly cancelled, Maps are cleared at appropriate lifecycle points, and the changes follow existing patterns. Score is 4 (not 5) because the `flushAll()` method is added but unused, and other channels using `createInboundDebouncer` may have similar leaks that aren't addressed in this PR. - No files require special attention - all changes are straightforward cleanup operations _{Last reviewed commit: 35b3681} <!-- greptile_other_comments_section --> _{(4/5) You can add custom instructions or style guidelines for the agent [here](https://app.greptile.com/review/github)!} <!-- /greptile_comment -->