Summary

Fixes #24279

openclaw gateway restart always times out because the health check compares the port-listener PID (child process) directly against the systemd MainPID (supervisor process). Since the gateway runs as a two-process tree (supervisor → child that binds the port), these are structurally always different, causing:

1. ownsPort is always false → healthy is always false
2. The child PID is always classified as stale → gets killed
3. The CLI waits 60s for health to become true, which never happens

Changes

• Add isDescendantOf(childPid, ancestorPid) that walks /proc/<pid>/status PPid field to check parent-child relationship
• Update ownsPort check: accept port listeners that are descendants of MainPID
• Update staleGatewayPids filter: descendants of MainPID are not stale
• Add unit tests for isDescendantOf

Fallback behavior

On non-Linux systems (macOS, Windows) where /proc is unavailable, isDescendantOf returns false gracefully, preserving the existing direct-PID comparison behavior.

Testing

• src/cli/daemon-cli/restart-health.test.ts — unit tests for isDescendantOf (7 cases)
• Existing lifecycle.test.ts tests are unaffected (they mock the restart-health module)

AI-assisted: Yes (Claude). Prompts/session logs available on request.

Greptile Summary

Fixed openclaw gateway restart timeout by adding process tree awareness to health checks. The gateway runs as supervisor → child process, where the child binds the port. Previously, health checks compared port listener PID directly against supervisor PID, causing perpetual "unhealthy" status.

Changes:

• Added isDescendantOf() to walk /proc/<pid>/status and check parent-child relationships
• Updated ownsPort check to accept descendants of MainPID as valid owners
• Updated staleGatewayPids filter to exclude descendants of MainPID
• Added 7 unit tests covering direct children, grandchildren, error cases, and edge cases
• Graceful fallback on non-Linux systems (returns false, preserves existing behavior)

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The fix directly addresses the reported issue with a clean implementation. The isDescendantOf function has proper safeguards (visited set to prevent loops, PID <= 1 boundary check, error handling), comprehensive test coverage (7 test cases including edge cases), and graceful degradation on non-Linux systems. The changes are surgical - only modifying the PID comparison logic without touching unrelated code.
• No files require special attention

_{Last reviewed commit: 133421b}