mjhjmtu

I launched an autossh tunnel between a Raspberry PI and a VPS and it works. I'm testing whether it's reliable after the RPI internet connection drops and reestablishes.

What I did

• Created an ~/.ssh/config and confirmed it's valid


• Installed autossh and successfully started the tunnel defined in the above config


• Unplugged the ethernet cable from my router, keeping the RPI Wi-Fi connection active


• Checked netstat -tulpan on both machines; waited a few minutes


• Plugged the ethernet cable back into the router

What was expected to happen

autossh re-establishes a working tunnel

What actually happened

autossh re-established a tunnel, but when I tried connecting to it from the other end, it gave me:

Unable to communicate with “my-host”. Make sure the remote computer is available and the firewall is not blocking screen sharing.

More info

After I took down the upstream internet connection, netstat on both VPS and RPI respectively showed:

tcp 0 0 VPS_IP:22 RPI_IP:41444 ESTABLISHED PID/sshd: admin

and

tcp 0 0 RPI_IP:41444 VPS_IP:22 ESTABLISHED PID/ssh

That is, same matching ports, and the status ESTABLISHED, even though RPI no longer had internet. Also, netstat on the VPS showed me the port I needed as being connectable (LISTEN), although the tunnel was supposed to be offline.

Once I plugged the cable back into the router, it showed the same situation on both ends, but with different tunnel ports now, plus the one I needed stayed the same.

Few minutes later, as I was writing this post, the port I needed has disappeared from the VPS, and it was only showing me the SSH connections (now two):

tcp 0 0 VPS_IP:22 RPI_IP:41878 ESTABLISHED PID/sshd: admin

tcp 0 0 VPS_IP:22 RPI_IP:41912 ESTABLISHED PID/sshd: admin

And the RPI netstat now looks like:

tcp 0 0 RPI_IP:41878 VPS_IP:22 ESTABLISHED PID/ssh

tcp 0 0 RPI_IP:41912 VPS_IP:22 ESTABLISHED PID/ssh

So apparently it restarted a tunnel (even two) but without port forwarding. Again, initially the tunnel worked. It's been 1 hour now and nothing changed.

I've seen this but the solution appears to be based on the netstat output, which in my case looks fine, but actually isn't working.