FULL mode problem!

[mhzabet]

i started a full mode using my own VPS, its works for few hours correctly, after it on full mode its not works but app script still works...
i dont know why?! i have multiple Deployment ID's and every things ok :/

[dazzling-no-more]

Send the logs.

[mhzabet]

now i can only use it on browser when use it as full mode in Mac or Windows but on android no problem with full mode...

[therealaleph]

Reviewed via Anthropic Claude.

@mhzabet — "works for few hours then full mode dies, apps_script still works" + "works on Android Full mode but not Mac/Windows" is a useful pattern. The split between platforms is the smoking gun.

Likely candidates given Android-works-but-desktop-doesn't:

1. Mac/Windows desktop firewall blocking outbound from a specific PID after some idle period. Both OSes have stateful firewall rules that can flap. Test: Settings → Network/Firewall → check rules.
2. Wi-Fi router NAT table aging out the long-lived connection. Tunnel-node's connections from your Mac/Windows go through your router; some consumer routers age out idle NAT entries after 3-4 hours and the connection fails silently next packet.
3. Mac/Windows DNS cache going stale on your VPS hostname. scutil --dns on Mac, ipconfig /flushdns on Windows.

Diagnostic to narrow it down: when full mode "stops working" on desktop:

• Run mhrv-rs from a fresh terminal — does it fail immediately on first request?
• curl -v https://your-vps-domain/health from Mac/Windows when failing — does it succeed? If yes, mhrv-rs is the issue. If no, the Mac/Windows-to-VPS network path is the issue.
• Restart mhrv-rs only (not VPS, not Apps Script). Does that fix it temporarily?

If restart mhrv-rs fixes it, it's a long-lived connection state bug in the client — possibly related to the h2 multiplexed connection going half-stale and not being torn down. PR #903 (per-session pipelining) reworks that path significantly; might be worth testing the PR branch.

If curl from Mac/Windows fails too, it's network-layer and not mhrv-rs's problem.

Could you do that 3-step diagnostic next time it fails and post the result? Leaving open as the tracking issue.

[mirbaagheri]

therealaleph @