Description

When starting the gateway with --replace, a race condition can leave multiple gateway instances running simultaneously. This triggers Telegram (and likely other platform) polling conflicts and causes the bot to become unresponsive.

Steps to Reproduce

1. Start the gateway normally (e.g. via launchd/systemd)
2. A second instance starts with --replace (e.g. manual restart or service restart overlap)
3. Both processes remain alive simultaneously

Actual Behavior

Multiple processes run at once (observed PIDs 548, 4101, and 4188 all alive simultaneously). Repeated errors in logs:

WARNING gateway.platforms.telegram: [Telegram] Telegram polling conflict (1/3), will retry in 10s.
Error: Conflict: terminated by other getUpdates request; make sure that only one bot instance is running

Expected Behavior

The old process should be fully terminated before the new one starts polling.

Root Cause

In start_gateway() (gateway/run.py), the new process writes its PID to the PID file before the old process has exited. A racing second --replace invocation then reads its own PID from the file (instead of the old process PID), so it skips the termination step and both instances run.

Environment

• Platform: macOS (darwin)
• Triggered by launchd auto-restart overlapping with a manual gateway run --replace

Suggested Fix

Write the new PID to the PID file only after the old process has been confirmed dead, or use a separate lock file that is held for the duration of the transition.