Publishing freeze on connection lost

[panter-dsd]

Version: master
If the option options.AutoReconnect is false then options.MessageChannelDepth is set to 0 here. So, we have a channel with zero length. When we get into the line

paho.mqtt.golang/client.go

Line 602 in 0d81c29

c.obound <- &PacketAndToken{p: pub, t: token}

and connection is lost (channel reader has already exited), we're stuck at this line and never leave the Publish function.

[ChIoT-Tech]

I believe that the fix introduces a new issue; if WriteTimeout is set to 0 (the default) then the "case <-time.After(c.options.WriteTimeout):" triggers immediately. This sometimes (approx 10% in my test code) ends up getting selected before the write to c.obound meaning that the publish message token is flagged as being in error state incorrectly (and the published message is lost despite AutoReconenct and persistence being active).

While setting WriteTimeout to a non-zero value would fix this the default value is currently is 0 (supposed to mean no timeout) and when used with AutoReconenct and Persistence these values make sense as the data will be written at some point.

The simplest solution might be:

if c.options.WriteTimeout == 0 {
	c.obound <- &PacketAndToken{p: pub, t: token}
} else {
	select {
	case c.obound <- &PacketAndToken{p: pub, t: token}:
	case <-time.After(c.options.WriteTimeout):
		token.setError(errors.New("publish was broken by timeout"))
	}
}

[panter-dsd]

@ChIoT-Tech sorry, I didn't notice this. Fixed. Better would be to using contexts but this will change API.

[MattBrittan]

Closing this because the change has been merged.