Likely duplicate of #42740 — same root cause and the same fix. Consolidating there; flagging for maintainer review.

✅ Verified — signal-initiated teardown skips gateway_state persistence

Reviewed the full diff in gateway/run.py.

• Flag propagation: runner._was_signal_initiated_shutdown = True is set in shutdown_signal_handler before calling runner.stop(), ensuring the flag is available in _stop_impl before the state-write decision.
• Conditional logic: getattr(self, "_was_signal_initiated_shutdown", False) defaults to False, so existing explicit-stop behavior is unchanged — only signal-originated teardowns skip the state write.
• Container boot compatibility: By not overwriting gateway_state on signal teardown, the previous state (typically "running") is preserved. container_boot.py reads this on next boot and auto-starts the gateway, fixing the container-restart issue (Gateway does not auto-start after container restart/upgrade — signal-initiated shutdown persists gateway_state=stopped #42675).
• Note: alt-glitch flagged this as likely duplicate of fix(gateway): preserve running state on signal shutdown #42740. Both PRs address the same issue with different approaches — fix(gateway): preserve running state on signal shutdown #42740 actively writes "running" state, while this PR skips writing entirely. Both are valid; this approach is more conservative (preserves whatever state was there).

The fix is correct. No issues found.

Superseded by #43236 (merged), which fixes #42675 using the existing planned-stop marker primitive instead of inferring intent from the signal.

The concern with skipping the stopped persist on any signal is that it treats every unmarked SIGTERM identically — but an OOM-kill or external kill under systemd Restart=always is also unmarked, and the marker-based classification distinguishes those from an operator stop reliably (see the analysis in #42517). #43236 also had to persist running rather than merely skip the write, because the mid-shutdown draining marker isn't in _AUTOSTART_STATES either — a real container-restart E2E surfaced that. Thanks for surfacing and pushing on this!