Memory leaking

[frct1]

Hi @schroffl i've detected memory leak when we recreating connection and for example sending hostinfo command

let Helpers=require('./helper');
let helper=new Helpers('memorytest');
let async=require('async');
let runloop=function () {
    async.forEachOf(credentials,(instance,index,callback)=>{
        helper.connectAndAuth(instance)
            .then(connection=>{
                return {query:connection,info:query.send('hostinfo')};
            })
            .then((object)=>{
                object.query.send('logout');//logout
                object.query.sock.destroy();//destroy connection because limit in 200 connections per instance
                callback();
            });
    },function () {
        console.log(`OK`);
        runloop();
    })
}
runloop();

Method connectAndAuth from helper class

connectAndAuth(credentials){
        return new Promise(((resolve, reject) => {
            let connect = new TeamspeakQuery.Raw({host:credentials.SQHost || credentials.ip, port:credentials.SQPort||credentials.query,localAddress: 'address'}, );
            connect.sock.on('error', function (err) {
                reject(err.code);
            });
            connect.sock.on('connect',()=>{
                connect.send('login', credentials.SQLogin||credentials.username, credentials.SQPassword||credentials.password)
                    .then(function () {
                        connect.ip = credentials.ip;
                        connect.name = credentials.name;
                        connect.throttle.set('enable', false);
                        resolve(connect);
                    }).catch(function (error) {
                        reject(error);
                        console.log(error);
                });
            });
        }))
    }

Already tried query.sock.removeAllListeners(), but it doesn't matter for memory using
Currently same thing is working on my API for job and memory growing so fast due to big count of request per minute.
P.S* Memory leak doesn't depend from command which we send to server query, that can be only simple connection with login command

[schroffl]

Can you try to replace the logout command in the first snippet with quit? The latter should actually cause the server to close the connection, which then leaves the TeamspeakQuery instance open to garbage collection (as long as you don't store any other references).

I'm not entirely sure though, so if you could run the tests and tell me if it helped I'd be really thankful :)

[frct1]

Replaced to quit command, but no effect

[frct1]

Memory using after 30k+ connections and login command

[schroffl]

I will check it out next week, since I won't be able to work on it until then.
I think I will finally implement #19 while analyzing/fixing this problem, because any resources can be freed in a proper disconnect method.

[frct1]

Yeah that's very needed.
Api app (in cluster mode 4 instances) after multiple days take around 3-4 GB of RAM lol

[frct1]

Hi, any changes or forecasts by troubleshooting?

[schroffl]

Hey, I just published Version 3.1.0 which adds a disconnect method as you requested in #19.
Can you try to call it like this instead of manually destroying the socket:

helper.connectAndAuth(instance)
  .then(connection => {
    return {
      query: connection,
      info: query.send('hostinfo')
    };
  })
  .then(object => object.query.disconnect())
  .then(() => callback());

Let's hope this fixes it :)

[frct1]

Hi @schroffl.
I've tested update: now it works better.
After 10+k connections and login command memory usage doesn't growing up.
Now deployed new api version, and provide more info after 3 days

[schroffl]

Alright, thanks for your feedback :)

[frct1]

Hi, there is still some leaking. But growing not so fast, by steps (i dont know what the reason).
Before update it is growing by every second. Now memory growing 1 time by N minutes
Tried with 100k+ connections

[schroffl]

The Garbage Collector doesn‘t have to kick in if there is no memory pressure. Are you sure that it is actually a leak? How much memory does it use in the end?

[frct1]

At this moment 866k connections.
Stats in htop

Stats in pm2
App name │ id │ mode │ pid │ status │ restart │ uptime │ cpu │ mem
memorytest │ 35 │ fork │ 4344 │ online │ 6 │ 74m │ 0% │ 85.3 MB

[frct1]

1400k connections 85MB of memory

[schroffl]

Hey, sorry for taking so long to finally answer. How does the memory consumption look now?

[frct1]

Now all is great. All services works perfect without 5+GB memory using even in peak cases :D
p.s: Thank you for update 😘

[schroffl]

I will aim to run the memory profiler at some point in the next few weeks and if no suspicious characteristics come up I'd consider this resolved.

[schroffl]

I'm going to close this issue now. I have to admit that I didn't run the profiler, but since no other complaints came up I feel confident to consider this resolved.