openclaw - 💡(How to fix) Fix [Bug]: WeChat channel message response significantly slower on 2026.5.x compared to earlier versions [1 comments, 2 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading to OpenClaw 2026.5.x (tested 5.12 and 5.18), the WeChat channel (@tencent-weixin/openclaw-weixin v2.4.3) has noticeably slower message response times compared to 2026.4.x and earlier 5.x versions.

The issue manifests as a significant delay between the user sending a message on WeChat and receiving the bot's reply. The model API call itself completes quickly (sub-2s for first token), but the overall end-to-end response time is much longer than expected.

This appears related to the session-resource-loader regression reported in #83943, but specifically affects the WeChat long-poll channel on Windows.

Steps to reproduce

1. Windows 11, Node.js v24.15.0
2. OpenClaw 2026.5.18 (latest stable), installed via npm
3. Model: deepseek/deepseek-v4-flash (via DeepSeek API)
4. Channel: WeChat (@tencent-weixin/openclaw-weixin v2.4.3, long-poll mode)
5. Start gateway, send a simple message via WeChat DM
6. Observe end-to-end response time

Expected behavior

Response time should be comparable to 2026.4.x — typically 3-8 seconds for a simple message with DeepSeek API (including network round-trip to API + token generation).

Actual behavior

Response time is significantly longer. From gateway logs, the model-fetch itself returns in ~700ms-1500ms (first token), but the full turn takes much longer due to internal processing overhead.

Key observations from logs:

• provider-transport-fetch shows model response in 734-1556ms (normal)
• But total turn time from inbound message to outbound reply is disproportionately long
• The session-resource-loader and session commit stages appear to add significant overhead
• Issue compounds over multiple messages in the same gateway session (similar to #83943)

OpenClaw version

2026.5.18

Operating system

Windows 11

Install method

npm

Model

deepseek/deepseek-v4-flash

Provider / routing chain

Primary: deepseek/deepseek-v4-flash (DeepSeek API, api.deepseek.com)

Additional provider/model setup details

Single agent (main), embedded mode, no sub-agents or ACP.

Logs, screenshots, and evidence

// Model fetch is fast:
[provider-transport-fetch] [model-fetch] start provider=deepseek model=deepseek-v4-flash
[provider-transport-fetch] [model-fetch] response status=200 elapsedMs=734

// But overall response perceived by user is much slower
// Diagnostic heartbeat shows active=1 waiting=0 queued=1 pattern
[diagnostic] heartbeat: webhooks=0/0/0 active=1 waiting=0 queued=1

Related issues

• #83943 — Session resource loader grows unbounded across warm turns (5.x regression)
• #57123 — Incremental/adaptive inbound message debounce (fixed debounce adds unnecessary delay)
• #55334 — sessions.json unbounded growth causes gateway OOM

Workaround

None found yet. Downgrading to 2026.4.x is not feasible as the WeChat plugin v2.4.3 requires >=2026.3.22.