[BUG] FBNS MQTT connection randomly disconnecting

[qwer-lives]

What broke?

I'm not sure if it's a bug, or this is expected.

I'm trying to listen to push notifications in a long lived worker. What I observe is the MQTT socket closing randomly after a while (it seems to happen around every hour but it's not consistent at all. I have tried sending pings every 10 minutes and also every 60 seconds - it doesn't seem to make any difference.

Is this expected given the nature of MQTT? If it is, should this reconnection happen internally within the aiograpi Client, or is it expected for the library user to implement their own reconnection logic?

Also, I noticed after the socket is broken calling fbns_disconnect sometimes leaves the client in a broken state: the disconnect also fails, so client.fbns is not set to None. I attached the code I have to work around this but it's not very nice.

Affected area

other

Minimal reproduction code

cl = Client()
// init client
try:
    while True:
        if not cl.fbns or not cl.fbns.connected:
            try:
                await cl.fbns_connect()
                cl.fbns_on("push", on_fbns_push)
            except Exception as e:
                log.error(f"FBNS connection error: {e}")
                await asyncio.sleep(10)
                continue
        try:
            await cl.fbns_read_once()
        except Exception as e:
            err_msg = str(e).lower()
            if "timed out" in err_msg or "timeout" in err_msg:
                continue
            log.error(f"FBNS read error: {e}")
            try:
                await cl.fbns_disconnect()
            except Exception as de:
                log.debug(f"FBNS disconnection error: {de}")
            finally:
                if cl.fbns:
                    with contextlib.suppress(Exception):
                        cl.fbns.connected = False
                        await asyncio.to_thread(cl.fbns.transport.disconnect)
                    cl.fbns = None
            await asyncio.sleep(10)
except asyncio.CancelledError:
    with contextlib.suppress(Exception):
        await cl.fbns_disconnect()
    raise
except Exception as e:
    log.exception(f"FBNS loop error: {e}")
    await asyncio.sleep(10)

Full traceback

No traceback

Observed result

MQTT socket randomly disconnects.

Expected result

Either the connection is stable, or reconnection is handled internally.

aiograpi version

1.10.26

Python version

3.14.5

Operating system

7.0.12-1-cachyos

Login method

username/password

Proxy in use

no proxy

Can you reproduce on current main?

yes

Raw API response or payload sample

Additional context

No response

[qwer-lives]

After testing it some more, it seems very inconsistent - sometimes the connection breaks after 30 min, sometimes it stays connected for 12h. I am regularly pinging every 60 seconds, and reconnecting immediately after the connection break seems to work (the account got flagged for suspicious behaviour once though).

[subzeroid]

Thanks for the detailed report. I reproduced the broken-state part from the code path and opened a fix in #400.

What the PR fixes:

• if FBNS transport read/send fails after the socket is already broken, FbnsClient.connected is marked False;
• fbns_disconnect() now treats the MQTT DISCONNECT packet as best-effort cleanup and still closes the transport;
• Client.fbns is cleared in a finally block, so a failed low-level disconnect does not leave the high-level client stuck.

This does not claim the MQTT socket itself will stay connected forever. For long-lived workers, remote/network disconnects are still possible and the safe pattern is to reconnect after read/ping failures. A separate high-level auto-reconnect/listen helper may still be useful, but #400 fixes the state corruption that made user-side reconnect logic awkward.

[subzeroid]

Thanks for the report. This is released now.

• Fixed in aiograpi 1.11.2: https://github.com/subzeroid/aiograpi/releases/tag/1.11.2 / https://pypi.org/project/aiograpi/1.11.2/
• Mirrored in instagrapi 2.17.2: https://github.com/subzeroid/instagrapi/releases/tag/2.17.2 / https://pypi.org/project/instagrapi/2.17.2/
• PRs: Fix FBNS cleanup after socket failures #400 and Fix FBNS cleanup after socket failures instagrapi#2715

What changed: FBNS now marks the client disconnected after MQTT transport send/read failures, and fbns_disconnect() clears broken FBNS state even if low-level socket cleanup fails. That should let long-running workers reconnect cleanly instead of getting stuck behind a stale/broken socket object.

Private live smoke: session-backed account access worked, and direct FBNS MQTT connect/register/ping/disconnect worked. The account HTTP proxies from the test pool returned 407 Proxy Authentication Required for MQTT CONNECT, so the successful MQTT smoke used direct transport. This release does not try to promise that an MQTT socket will stay connected forever; workers still need a reconnect loop for normal network/server disconnects.

[subzeroid]

Closing as released in aiograpi 1.11.2 and mirrored in instagrapi 2.17.2.

The fix covers stale/broken FBNS socket state after MQTT transport failures so callers can reconnect cleanly. Long-running consumers should still keep their reconnect loop for normal network/server disconnects.