Environment

• OpenClaw 2026.4.15 (041266a)
• Node v24.14.1
• Ubuntu 24.04.3 LTS, aarch64 (Hetzner CAX21)
• User-level systemd unit at ~/.config/systemd/user/openclaw-gateway.service

Observed

On every gateway restart, the bonjour advertiser reliably hangs in the announcing state for ~14–15 seconds. A watchdog then kicks in, logs the stall, and restarts the advertiser. This happens on every restart (observed ~8 distinct restarts in 6 hours, 100% repro).

Log excerpt

[bonjour] watchdog detected non-announced service; attempting re-advertise (gateway fqdn=openclaw-01._openclaw-gw._tcp.local. host=openclaw.local. port=18789 state=probing)
[bonjour] restarting advertiser (service stuck in announcing for 14774ms (... state=announcing))

Stall durations across 8 restarts: 14466, 14774, 14802, 15019, 15127, 15708 ms — consistent ~15s.

Impact

• Gateway readiness is delayed by ~20–30s on every restart (the stall + the restart itself).
• Non-fatal; once the advertiser is restarted, it comes up normally.
• May be correlated with the [ws] handshake timeout + [telegram] native-approvals failed to start cluster that fires at the same timestamps — see companion issue (link coming in comment after this is filed).

Hypothesis

Looks like an async initialization race where the advertiser attempts announcing before whatever it depends on (mDNS socket? state machine event?) is ready, then times out. The watchdog is a band-aid — the underlying "why does announcing hang for exactly ~15s?" probably needs addressing.

Repro

1. systemctl --user restart openclaw-gateway on any Linux host with user-level systemd.
2. journalctl --user -u openclaw-gateway -f | grep bonjour — the stall shows up within ~15s of startup.

Happy to collect more diagnostic data if useful (strace, lsof on the mDNS socket, etc.).