openclaw - 💡(How to fix) Fix [Bug]: Gateway pegs single CPU thread at 100% immediately on boot in 2026.4.26 — TUI handshake timeouts, multi-minute message latency [1 comments, 2 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading from an earlier release to 2026.4.26, openclaw-gateway immediately pins one CPU thread at 100% on boot and stays there indefinitely, while reporting gateway ready in the logs. The TUI cannot establish a WebSocket connection to the gateway (handshake timeouts), and inbound channel messages (WhatsApp, Telegram) take 30 seconds to several minutes to receive a reply, if at all. This issue also existed in 2026.04.25 and was not present on version 2026.04.21

Steps to reproduce

1. Upgrade openclaw to 2026.4.26 via npm.
2. Restart the gateway: systemctl --user restart openclaw-gateway
3. Observe: top -p $(pgrep -f openclaw-gateway) shows ~100% CPU continuously, even when idle.
4. Run openclaw tui in another shell. The TUI fails to connect with gateway closed (1000) or gateway request timeout for connect.
5. Send a message via a connected channel (WhatsApp/Telegram). Reply latency is 30+ seconds, often longer.

Expected behavior

• Gateway should idle near 0% CPU when no messages are being processed.
• TUI should connect within the WebSocket handshake window.
• Inbound channel messages should be processed in normal time (sub-second routing, latency dominated by LLM call).

Actual behavior

• Gateway should idle near 0% CPU when no messages are being processed.
• TUI should connect within the WebSocket handshake window.
• Inbound channel messages should be processed in normal time (sub-second routing, latency dominated by LLM call).

OpenClaw version

2026.04.26

Operating system

Ubuntu 24.04

Install method

npm global

Model

openai-codex/gpt-5.4

Provider / routing chain

openclaw -> openai

Additional provider/model setup details

No response

Logs, screenshots, and evidence

## Environment

- **openclaw**: `2026.4.26` (commit `be8c246`)
- **OS**: Ubuntu 24, x86_64 cloud VPS (4GB RAM, 2 vCPU)
- **Node**: `22.22.2`
- **Install method**: `npm install -g openclaw`
- **Run mode**: gateway under `systemd --user` service
- **Active plugins on boot log**: 1 in-house plugin, plus `browser`, `memory-core`, `telegram`, `whatsapp` (5 plugins)
- **Agent model**: `openai-codex/gpt-5.4` (primary), `openai-codex/gpt-5.3-codex` (fallback)

Impact and severity

Interactive use with the agent is effectively impossible. Question to answer latency is too long for any work. Sometime answer doesn't come at all.

Additional information

CPU stays pegged with no inbound traffic at all.

What I've tried

• Disabled an in-house plugin (not published to npm; moved its extension directory aside) and restarted the gateway. Gateway boot succeeds with 5 plugins instead of 6, but CPU still pegs at 100% immediately. So the issue is not specific to that plugin.
• /queue interrupt via channel — confirmed accepted, but did not change latency.
• Hard restart of the gateway — issue reproduces immediately on each restart.
• Memory / OOM check: no swap, no OOM kills in dmesg, ~2GB free buff/cache. Issue is not memory pressure.
• Background workflows that bypass the gateway (cron-driven Python pipelines making LLM API calls directly) work normally, so the VPS, network, and upstream LLM provider are all healthy.

Possibly related side observation

A new config warning appears in 4.26 that wasn't present before:

Config warnings:
- plugins.entries.workspace: plugin workspace: installed plugin index
  could not hash /home/<user>/.openclaw/workspace/.claude-plugin/plugin.json:
  ENOENT: no such file or directory

There is no .claude-plugin/plugin.json manifest in the workspace directory. Unclear if this is decorative or related to the CPU loop, but it is new in 4.26.

Workaround

Rolling back to 2026.4.21 is my next step.

Logs available

I can provide:

• Full journalctl --user -u openclaw-gateway since the upgrade
• /tmp/openclaw/openclaw-2026-04-28.log
• ~/.openclaw/openclaw.json (sanitized of secrets) if useful

Just let me know what would help most.