openclaw - 💡(How to fix) Fix [Bug]: Regression (previously worked, now fails)

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

On Windows, OpenClaw versions after 2026.4.23 show severe performance regressions when using the Feishu/Lark channel. A simple message such as “Hello” can take 2–5 minutes to complete. During processing, multiple subsystems stall simultaneously: Event loop delay reaches 40–145 seconds auth stage takes 45–50 seconds per request session-resource-loader takes 28–36 seconds Feishu WebSocket connections repeatedly timeout and reconnect Memory embeddings are rate limited and block progress Sessions are classified as stalled_agent_run Version 2026.4.23 does not exhibit these issues and remains fast and stable on the same machine and configuration. This appears to be a Windows-specific runtime regression introduced after 2026.4.23, likely involving event loop starvation and cascading subsystem delays.

Steps to reproduce

Install OpenClaw 2026.5.12 on Windows 11. Configure Feishu/Lark channel in WebSocket mode. Enable memory-core with default settings. Use MiniMax-M2.7 as primary model and DeepSeek as fallback. Start 5 Feishu bot accounts simultaneously. Send a simple message such as:

Hello, are you there?

Expected behavior

Response should begin within a few seconds, similar to OpenClaw 2026.4.23. Typical target timing: auth: < 5s session-resource-loader: < 3s total response startup: < 10s no WebSocket reconnect storms no stuck session diagnostics

Actual behavior

A simple greeting may take 2–5 minutes. Observed delays: auth: 45–49 seconds session-resource-loader: 28–36 seconds prep stages: 38–49 seconds event loop delay: up to 144,820 ms fetch timeout: 120–165 seconds repeated WebSocket timeout and reconnect loops The system eventually responds, but only after severe delays.

OpenClaw version

2026.5.12

Operating system

Windows 11

Install method

npm

Model

MiniMax-M2.7 DeepSeek Chat (fallback)

Provider / routing chain

openclaw -> minimax-portal/MiniMax-M2.7 openclaw -> custom-api-deepseek-com/deepseek-chat

Additional provider/model setup details

Primary model: openclaw -> minimax-portal/MiniMax-M2.7 Fallback model: openclaw -> custom-api-deepseek-com/deepseek-chat

Feishu/Lark channel is configured in WebSocket mode.

Installed extensions:

• browser
• device-pair
• memory-core
• memory-wiki
• openclaw-lark

Memory-core is enabled with default settings.

Five Feishu bot accounts are started simultaneously:

• lobster-bot
• aochuang-bot
• zhoubapi-bot
• jiaweisi-bot

No custom code modifications were made. The same machine, Node.js version, and configuration work normally in OpenClaw 2026.4.23.

Logs, screenshots, and evidence

eventLoopDelayMaxMs=144820.9
eventLoopUtilization=1
cpuCoreRatio=0.998

auth:48727ms
auth:45213ms

session-resource-loader:28247ms
session-resource-loader:36661ms

[ws] timeout of 15000ms exceeded
unable to connect to the server after trying 1 times

memory embeddings rate limited; retrying...

classification=stalled_agent_run
reason=active_work_without_progress

fetch timeout after 120000ms (actual 144877ms)
provider=minimax-portal
elapsedMs=164865
message=Request timed out

Impact and severity

Plain text Affected users: Windows users running OpenClaw versions later than 2026.4.23, especially those using the Feishu/Lark channel with memory-core enabled.

Severity: High. Basic conversations become extremely slow and daily use is impractical.

Frequency: Observed consistently in every test after upgrading to 2026.5.12.

Consequences:

• Simple messages take 2–5 minutes to complete.
• Feishu WebSocket connections repeatedly timeout and reconnect.
• Sessions are marked as stalled.
• Memory operations are delayed.
• Significant productivity loss and poor user experience.

Additional information

Last known good version: 2026.4.23 First known bad versions: 2026.4.27 and later Tested version: 2026.5.12

Observed progression:

• 2026.4.23: Stable and responsive
• 2026.4.27: Occasional timeouts begin
• 2026.4.29: Severe instability
• 2026.5.x: Consistent 2–5 minute delays

Related issues:

• #82073 Feishu/Lark channel slow responses
• #82264 2026.5.12 slower than 2026.5.7
• #83103 Feishu repeatedly re-enters start-account
• #83449 MiniMax auth takes 30–50 seconds per request

The issue appears to be a regression introduced after 2026.4.23 on Windows. The bottleneck occurs before model generation begins, suggesting event loop starvation and subsystem contention rather than a model-specific problem.

This is the same file as below. This is my error log and some configs. This error has been here since I upgraded from 4.23 — it never went away. I rewrote it clearly this time and am submitting a bug report. Hope you can find the problem.

8888 (1).txt