Autossh reconnection issue with sish

[manprint]

Hi @antoniomika,

I think I found an issue when using sish with autossh.
After starting sish with autossh with the following command, for example:

autossh -M 0 -p 2222 -R pico16-a:80:localhost:8004 sish.mydomain.link

When you want to stop the tunnel, since autossh automatically reconnects the tunnel, you have to press CTRL+C several times in rapid succession to stop sish session and exit autossh.

if you try to reconnect to the same domain sish says:

Warning: remote port forwarding failed for listen port 80
The subdomain pico16-a.sish.mydomain.link is unavailable.

In order to reuse the blocked domain, sish must be restarted
the problem is probably due to the fact that sish is very quick to reconnect.
I managed to replicate the issue several time.

A possible solution to mitigate the problem could be a parameter to delay the ssh connection when ssh is invoked? (--ssh-connection-delay=5s)

If you need to do further tests I can help you.

Best,
Fabio.

[manprint]

Hi,
I did further tests and when what I described above happens, this is in the sish server logs:

2024/06/17 - 18:58:35 | Accepted SSH connection for: 172.18.0.3:53570
2024/06/17 - 18:58:35 | HTTP forwarding started: http://test.dev.synclabapps.link -> /tmp/172.18.0.3_53564_80332338001 for client: 172.18.0.3:53564
2024/06/17 - 18:58:35 | HTTPS forwarding started: https://test.dev.synclabapps.link -> /tmp/172.18.0.3_53564_80332338001 for client: 172.18.0.3:53564
2024/06/17 - 18:58:35 | Error replying to port forwarding request: read tcp 172.18.0.2:2222->172.18.0.3:53564: use of closed network connection
2024/06/17 - 18:58:36 | EOF
2024/06/17 - 18:58:51 | Accepted SSH connection for: 172.18.0.3:35654
2024/06/17 - 18:58:51 | Login attempt: 172.18.0.3:35654, user fabio key: ssh-rsa (REDACTED)
2024/06/17 - 18:58:51 | Error setting up HTTPListener: error assigning requested subdomain to tunnel

Step to reproduce the issue:

1. start tunnel with autossh -M0 -p 2222 -R test:80:localhost:8080 dev.mydomain.link
2. wait for 30+ seconds for the gate time (AUTOSSH_GATETIME=30 (default)) to finish
3. repeatedly press or hold CTRL+C to stop sish session and autossh

I hope it helps you.
Best,
Fabio

[zk-steve]

sish will close unused connections (include closed ones) by default, unless you mess something up with your configuration.

Otherwise, you can try to set:

force-requested-subdomains = true,

[antoniomika]

Agree with setting force-requested-subdomains.

I’m still looking into this as I’m wondering if maybe it has something to do with autossh’s monitor port.

[synclab-srl]

Hi,
I tried the setting you recommended, but it didn't work. The problem persists.
I attach a video of the test case in which the problem is highlighted.
I hope it can be of use to you.

sish_issue.mp4

Best,
Fabio

[synclab-srl]

Hi @antoniomika

In the past weeks I did some tests and with the help of a developer friend of mine, we found a way to solve the issue.

We added, to line 64 of sshmuxer/channel.go and to line 87 of utils/conn.go a sleep of 3 seconds.

In this way, the issue disappears: reconnections and stop/start tunnel with autossh work correctly.

I attach the modified files.

I did not make a pull request, because I think the fix is ​​dirty, surely you will have a better and more elegant way to solve the problem :D

Best,
Fabio.

modified_files_go.zip

[llapthorn]

Hello,
Just to confirm we see same behaviour on our side.
It can also be triggered when autossh have to reconnect due to a disconnection.
When the error "The alias xyz is unavailable", the only fix we found at the moment is to restart sish.
Regards,
Léo