openclaw - 💡(How to fix) Fix openclaw devices list times out under live-write races in device store

Symptom

openclaw devices list command times out at 10 seconds (max wait). Gateway logs show JsonFileReadError on paired.json read path.

Root Cause

The device store's atomic-file reader appears to have a race condition with the live-write path. When the reader's mtime/inode guard detects the file changing during the read operation, it rejects the operation as potentially corrupt. This is a safety measure, but it causes the CLI request to hang indefinitely until timeout.

The error occurs even after cleanup of stale .tmp orphans, suggesting the race is in the canonical write path, not from orphans.

Environment

• Gateway: 2026.5.18 (50a2481)
• Device store: ~/.openclaw/devices/paired.json (~31 KB, healthy structure)
• OS: Linux Mint 21.3
• Node: 24.15.0

Observed Error

{
  "error": "JsonFileReadError",
  "filePath": "~/.openclaw/devices/paired.json",
  "cause": "File changed during read"
}

Context

• Plugin: device-pair enabled with active config
• Pattern: reproducible whenever openclaw devices list is called while device writes may be in flight
• Impact: openclaw devices approve and related device management commands are blocked

Suggested Fix

Investigate the write path of the device store:

• Current: appears to use atomic replacement (rename temp file → final file) with mtime/inode guards on the read side
• Candidate improvements:
  • Copy-on-write snapshot: let readers get a stable snapshot while writers work on a shadow copy
  • Read-while-locked: hold a lock during write, snapshot before releasing, allow readers to access the snapshot
  • Backoff + retry: add exponential backoff to the reader when mtime/inode change is detected, rather than failing immediately

References

Related: issue tracking stale .tmp orphans (Issue #84750)