{"id":13392,"date":"2019-09-08T17:46:10","date_gmt":"2019-09-08T15:46:10","guid":{"rendered":"https:\/\/pi3g.com\/?p=13392"},"modified":"2019-09-08T17:46:10","modified_gmt":"2019-09-08T15:46:10","slug":"paho-mqtt-client-javascript-envoy-vernemq-debugging-disconnects","status":"publish","type":"post","link":"https:\/\/pi3g.com\/de\/paho-mqtt-client-javascript-envoy-vernemq-debugging-disconnects\/","title":{"rendered":"Paho MQTT Client (Javascript), Envoy, VerneMQ: Fehlersuche bei Verbindungsabbr\u00fcchen"},"content":{"rendered":"

After deploying picockpit to my server, a bug was observed: the web frontend would disconnect in very regular intervals. <\/p>\n

This had not been present on the local development environment. Users of picockpit reported the same bug. <\/p>\n

I have debugged it today, and applied a hotfix. It turns out it was a timing issue. <\/p>\n

TL;DR how to fix this<\/h1>\n

set up a shorter keepAliveInterval for Paho. 60 seconds is the default, but with network latencies this will lead to these timeouts.<\/p>\n

Here is my code for the connection (inside pcp-code.js which is loaded by the webpage):<\/p>\n

\n

\/* https:\/\/www.eclipse.org\/paho\/files\/jsdoc\/Paho.MQTT.Client.html<\/a> *\/
\nfunction MQTTconnect_inner(params){
     mqtt = new Paho.Client(params.uri, params.clientid); \/* setting not required parameters as JS does not allow to pass named positional parameters *\/
     mqtt.onConnectionLost = mqtt_onConnectionLost;
     mqtt.onMessageDelivered = mqtt_onMessageDelivered;
     mqtt.onMessageArrived = mqtt_onMessageArrived;
     mqtt.onConnected = mqtt_onConnect_proper;<\/p>\n

    var options = {
         timeout: 3, \/* if conection not successfull within 3 seconds, it is deemed to have failed *\/
         onSuccess: mqtt_onConnect_connect,
         onFailure: mqtt_onConnectFailure_connect,
         userName: params.userName,
         password: params.password,
         keepAliveInterval: 30,
         useSSL: true,
         reconnect: false,
         mqttVersion: 4
     };
     mqtt.connect(options);   
\n}<\/p>\n<\/blockquote>\n

\"image\"<\/a><\/p>\n

(reproduced as image as well)<\/p>\n

Debugging steps<\/h1>\n

Timing<\/h2>\n

I sat down with my mobile phone to time the timeouts, and see whether there was a pattern there.<\/p>\n

\"image\"<\/a><\/p>\n

In my implementation, Paho is reconnected by surrounding JavaScript code every 10 seconds on timeout. <\/p>\n

\"image\"<\/a><\/p>\n

We see the Paho javascript MQTT client loosing the connection every minute or every two minutes. This is already a strong indication that this is systematic, possibly a forced disconnect by the server.<\/p>\n

JavaScript console<\/h2>\n

\"image\"<\/a><\/p>\n

The error message I see in the console is:<\/p>\n

WebSocket is already in CLOSING or CLOSED state.<\/strong><\/p>\n

In retrospect, this is possibly Paho trying to send the PINGREQ but failing just by a couple of seconds or possibly microseconds, as the server has disconnected the socket.<\/p>\n

NB: For superior compatibility with company firewalls, etc. I run Paho through websockets. <\/p>\n

Verification with other browser<\/h2>\n

I have seen this issue with Chrome (my main browser) and Firefox. Therefore it is not due to a browser problem.<\/p>\n

(I was led to believe that it might be to some browser issues by this website: https:\/\/github.com\/socketio\/socket.io\/issues\/3259<\/a>)<\/p>\n

Debugging VerneMQ<\/h2>\n

see also:<\/b><\/p>\n