Client idles and stops serving requests

[Saturnix]

(Thanks for making this awesome piece of software!)

I have a little problem: I run my media server on Windows, but I'm behind NAT. I also have a Linux VPS, which I use to expose my local machine thanks to your script.

Linux runs in server mode, Windows runs in client mode.

The way I run the client is simply opening a cmd.exe tab, and run the script as in the example in the docs

Python27\python.exe natsrv.py --mode client --secret pass123 --local 192.168.1.117:8096 --admin example.com:80

However, if I leave cmd.exe open for a while, example.com:80 eventually stops serving requests. There are no error messages or anything.

I know the error is on the client because if I simply close cmd.exe, reopen it and rereun the client it starts working again.

This happens even after just a few minutes of inactivity (less than 20mins).

Have any idea what might be causing this?

Many thanks again!

[rofl0r]

thanks for your report. have you tried running the client on a linux machine already? i wonder whether this is a bug in the windows version of python.
trying python3 might also be an option.

[Saturnix]

Hi, thanks for the answer. Still have to try running the client on Linux: will try and let you know here.

Python 3 doesn’t work, I thought it wasn’t supported so I downgraded to 2.7. Will post you the precise error I get on Python 3 on a separate issue asap.

EDIT: done. Opened a separate issue for Win.

[meetcshah19]

@rofl0r Same issue for me. I am running both client and server on linux.

[rofl0r]

@rofl0r Same issue for me. I am running both client and server on linux.

sigh. maybe using python for this project was a bad choice. anyway, maybe you can help debugging to find out what's wrong. for example does netstat output give some hint about the state of connections ?

[meetcshah19]

Yup thats what I have been trying. It looks like once the socket connection between the client and the server breaks (maybe due to unstable internet) it is unable to establish the connection again.

[rofl0r]

the way the program currently works is that
0) establishing control channel connection

1. it establishes an idle data connection to the server
2. waits for a request
3. as soon as request is served a new conn is intantiated (i.e. goto 1)
   this is done so there's no latency for establishing the conn after client connect.
   can you figure out whether the control connection or the data connection gets interrupted?

[meetcshah19]

It seems like the data connection is getting interrupted.

[ariririos]

Same issue here between two enterprise ethernet connections (MIT local and AWS lightsail remote) so I doubt it's anything to do with the actual internet connection. Maybe some sort of keep-alive isn't being properly set?

[rofl0r]

that's possible. maybe the "prepare a connection in advance" thing wasn't so smart after all. do you feel capable of changing the code so the data connection is only done after a client connects?

[ariririos]

that's possible. maybe the "prepare a connection in advance" thing wasn't so smart after all. do you feel capable of changing the code so the data connection is only done after a client connects?

Yes I think so! I'll take a look this weekend.

[ariririos]

that's possible. maybe the "prepare a connection in advance" thing wasn't so smart after all. do you feel capable of changing the code so the data connection is only done after a client connects?

The socket code here is pretty far above what I'm familiar with, and I don't want to break anything, so I don't think I'll be able to make the necessary edits to resolve this issue. Sorry about this!

[schtritoff]

Got a working solution for the problem with client idle connection dropping by using systemd unit and watchdog script.

Client systemd unit (for example in /etc/systemd/system/nat-tunnel-01.service)

[Unit]
Description=nat-tunnel for ssh access
After=network.target

[Service]
Type=simple
WatchdogSec=20
NotifyAccess=all
Restart=always
RestartSec=60
Environment=WATCHDOG_USEC=2000000
ExecStart=/usr/bin/python3 /root/nat-tunnel/natsrv.py --mode client --secret $MYSECRET --local localhost:22 --admin $MYVPSHOST:ADMINPORT
ExecStartPost=/root/nat-tunnel/watchdog.sh

[Install]
WantedBy=multi-user.target

Similar could be used for server side but without parameters for watchdog script (ExecStartPost, WatchdogSec, Environment, NotifyAccess).

watchdog.sh contents

#!/usr/bin/env bash

# src: https://www.medo64.com/2019/01/systemd-watchdog-for-any-service/

watchdog() {
    while(true); do

        # src: https://stackoverflow.com/a/19866239
        TIMEOUT=`timeout 1 bash -c 'cat < /dev/null > /dev/tcp/$MYVPSHOST/$PUBLICPORT'`
        if [ "$?" -eq 0 ]; then
            #echo yeah
            /bin/systemd-notify WATCHDOG=1;
            sleep $(($WATCHDOG_USEC / 2000000))
        else
            #echo no
            sleep 1
        fi
    done
}

watchdog &

Now it works reliably for me, on restart or any other occasion - the service is always available in my case. Maybe this could be somewhere in docs / readme for this project.

[rofl0r]

Maybe this could be somewhere in docs / readme for this project.

your solution is not universal - it depends on systemd, which i despise.
i'd rather find out what's going wrong and fix the bug than documenting bug workarounds.

[schtritoff]

Regarding SystemD workarround - works for me but not for everyone, I agree. Most popular linux distros have systemd out of the box so this comment might be of value for some users.

On topic - it could be that some router (local gateway or ISP) is dropping/closing connections because there is no activity. I didn't test local (same subnet) client-server variant. If it work in the same subnet it could be that some 3rd party is closing inactive connections. Maybe some heartbeat data flow is needed to keep connection alive.

[rofl0r]

Maybe some heartbeat data flow is needed to keep connection alive.

this could only mitigate the problem to some degree. the fundamental issue here is that an existing connection is rendered infunctional (disconnected?) but the code fails to detect that event properly.

having an strace dump available for when this happens would help a lot in figuring out what happens effectively.