Bug: DuplicateOpenClawRuntimeError on gateway restart blocks plugin registration

[wacke1]

Bug Description

On gateway restart, memos-local-plugin fails to acquire its runtime lock during the brief window between SIGTERM and process death, causing DuplicateOpenClawRuntimeError to abort register(). All hooks and tools become unavailable until next restart.

Reproduction

1. Gateway running with plugin loaded
2. Run systemctl --user restart openclaw-gateway or openclaw gateway restart
3. Check logs:

[plugins] memos-local-plugin failed during register: DuplicateOpenClawRuntimeError: memos-local OpenClaw runtime is already active (pid=XXXX)

4. memos_search / memos_get tools unavailable, before_prompt_build hook not triggered

Root Cause

In runtime-lock.js, acquireOpenClawRuntimeLock() loop:

if (owner && pidIsAlive(owner.pid)) {
    throw new DuplicateOpenClawRuntimeError(lockDir, owner);  // BLOCKS here
}
// Only clears lock when !owner (owner is null)
if (!owner && !lockLooksStale(lockDir, now(), unwrittenOwnerStaleMs)) {
    throw new DuplicateOpenClawRuntimeError(lockDir, null);
}
fs.rmSync(lockDir, { recursive: true, force: true });

After SIGTERM, pidIsAlive(old_pid) returns true during the 100-500ms grace period before OS reclaims the PID. The lock-clear path only handles !owner, so it never triggers when owner exists and pid is alive.

Proposed Fix

Add retry with backoff when owner pid is alive:

if (owner && pidIsAlive(owner.pid)) {
    await new Promise(r => setTimeout(r, 200));
    continue;  // retry
}

Environment

• OpenClaw: 2026.7.1-2
• memos-local-plugin: 2.0.16 (npm) / 2.0.0-beta.5 (registered)
• Node.js: 24.18.0
• OS: Linux 7.0.14-6-pve (Proxmox)

[wacke1]

Update: Gateway 已优化。将 MemoryHigh 从 3G 提高到 4G，MemoryMax 从 8G 降到 5G。当前已稳定运行超过 1 小时，重启频率从 24h 10 次降到 0 次。DuplicateOpenClawRuntimeError 复现频率大幅降低，但需继续观察。如有更多复现日志请提供。