on connection callback not triggered on reconnection to the same broker with the same mosq instance

[anthonio9]

mosquitto version: 2.0.18
system: ubuntu, kernel 5.15.0-91

Hi,

I've noticed that when running mosquitto_connect or mosquitto_connect_async on mosquitto instance that has just been disconnected from a broker, the connection callback is not triggered, even if the broker (the same broker) actually receives a connection request and reponds with a CONNACK message. This results in logs that look like below:

1704722562: New connection from 192.168.1.19:40550 on port 1883.
1704722562: New client connected from 192.168.1.19:40550 as auto-48ECE486-E347-A8CB-3E04-2E05A03E8E38 (p5, c1, k10).
1704722562: No will message specified.
1704722562: Sending CONNACK to auto-48ECE486-E347-A8CB-3E04-2E05A03E8E38 (0, 0)
1704722581: Client auto-48ECE486-E347-A8CB-3E04-2E05A03E8E38 has exceeded timeout, disconnecting.

The timeout happens because there connection callback is never triggered.

Is this the expected behavior? If yes, then it would mean that after a disconnect the mosquitto instance has to be always destroyed or reinitialized, is that the right way?

[popovec]

Please try to look at the PR #2966 .. it may be the same problem.

[anthonio9]

I don't think it's the same, but I'll give your branch a go anyway.
Your PR seems to take care of a situation where the client call mosquitto_connect_async before the async loop is started. In my case the async loop is started first, then the mosquitto_connect_async is run, however not always after this call the onConnectCallback is triggered, which means that not always the connection gets finalized.

[anthonio9]

Should a new connection be made with a fresh mosq structure, or can it be recycled? Does it have to be recycled with mosquitto_reinitialized?

[anthonio9]

Using mosquitto_reinitialized just before running the mosquitto_connect_async really helped making a reliable connection every time.