SIGILL (Illegal instruction) restart loop when enabling memory-lancedb-pro on older x86_64 CPU (no AVX/AVX2)

[babeloo]

Plugin Version

v1.1.0-beta.x, worked on older versions

OpenClaw Version

2026.3.8 and 2026.3.11

Bug Description

Summary

Enabling memory-lancedb-pro causes the OpenClaw gateway process to be killed by SIGILL (Illegal instruction) and systemd keeps restarting it. This only happens when the plugin is enabled. Disabling the plugin makes the gateway stable again.

This looks like a CPU instruction set compatibility issue (likely AVX/AVX2 requirements from a native dependency).

Environment

• OpenClaw: 2026.3.11
• Node.js: v22.22.0
• Arch: x86_64
• CPU: Intel(R) Celeron(R) CPU J3455 @ 1.50GHz
• CPU flags: no avx / avx2 shown in lscpu output

Actual Behavior

• Gateway enters a restart loop.
• systemctl --user status openclaw-gateway.service shows the process was killed by illegal instruction:
  • code=killed, signal=ILL
• systemctl --user show openclaw-gateway.service -p NRestarts keeps increasing.

Expected Behavior

• Plugin should either:
  - run normally on this CPU, or
  - fail fast with a clear, user-friendly error explaining CPU requirements (instead of crashing the whole gateway process with SIGILL).
• If yes, can you document the minimum CPU instruction set requirements prominently?
• Is there a non-AVX build / fallback path available?
• Could the plugin detect missing CPU features at startup and emit an explicit error, to avoid SIGILL crashing the entire
  gateway?

Steps to Reproduce

Reproduction Steps

1. Install plugin under shared plugins dir:
   • ~/.openclaw/plugins/memory-lancedb-pro
2. In the instance config (openclaw.json), enable the plugin:
   • plugins.slots.memory = "memory-lancedb-pro"
   • plugins.entries["memory-lancedb-pro"].enabled = true
   • Ensure plugins.allow includes "memory-lancedb-pro" (if using allowlist)
3. Restart the gateway service.

Control Test (Confirms It’s Plugin-Specific)

• Same machine, same OpenClaw:
  • With memory-lancedb-pro disabled (or memory slot switched back), the gateway is stable and does not restart.
  • Only enabling memory-lancedb-pro triggers SIGILL + restart loop.

Error Logs / Screenshots

Embedding Provider

None

OS / Platform

No response

[Unicellular-GKZ]

一样的问题，昨日更新3.11和v1.1.0-beta.8后，opneclaw无限重启，无法正常启动

[AliceLJY]

SIGILL 基本可以确认是 LanceDB 底层原生库（@lancedb/lancedb）依赖 AVX2 指令集，旧款 x86_64 CPU（Haswell 以前，或部分 Xeon E3/E5 系列）不支持会直接触发 Illegal Instruction。

排查确认方式：

grep -m1 avx2 /proc/cpuinfo

没有输出说明 CPU 不支持 AVX2，问题即在此。

目前没有官方 CPU 降级构建，临时绕过方案：换支持 AVX2 的机器，或在容器里用 QEMU 模拟（性能会下降）。这个限制来自 LanceDB 上游，插件侧暂时无法绕过，已记录等待上游支持。

[AliceLJY]

跟 #172 是同一个问题——LanceDB 上游原生库依赖 AVX2 指令集，旧款 CPU 不支持会 SIGILL。在 #172 统一追踪，关闭这个避免重复。