[BUG] Linux Clients Frequently Disconnect v1.78.1

[MeCJay12]

Is this a support request?

• This is not a support request

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

I have 23 clients connected to a Headscale server. Some of the clients have no issues but many of the clients will disconnect and go offline after some time. Some clients it happens immediately after connecting and others is happens after some time. The issue seems to be 1.78.1 on Linux.

Ubuntu 22.04.5: 4 clients (1.78.1), disconnect after medium to long time
Debian 12: 5 clients (1.78.1), disconnect after medium to long time
Debian 11/Pi 4: 1 client (1.78.1), disconnect after medium to long time

Android: 2 clients (1.78.3), maybe issue but hard to tell with sleep and wake
Docker (on above Ubuntu): 4 clients (1.78.3), no issue
Windows 11 24H2: 1 client (1.78.1), no issue
Synology: 2 clients (1.58.2), no issue

I tried running tailscale bugreport on my broken nodes but I got back BUG-NO-LOGS-NO-SUPPORT-this-node-has-had-its-logging-disabled and I'm unsure how to change it. I also noticed that re-running tailscale login would reconnect the client for a time but running down && up did not.

Expected Behavior

Clients should connect and, as long as they stay online, stay connected to Headscale.

Steps To Reproduce

I followed the Tailscale install guides for each client then ran tailscale login --login-server https://hs.example.com

Environment

- OS: Ubuntu 22.04.5, Debian 11/12
- Headscale version: 0.4.1
- Tailscale version: 1.78.1

Runtime environment

• Headscale is behind a (reverse) proxy
• Headscale runs in a container

Anything else?

config.yaml.txt
Debian11.json
Debian12.json
Ubuntu.json

[MeCJay12]

Tailscale Bug

[kradalby]

@MeCJay12 Have you tried using Tailscale SaaS and still have this problem? If not, please do not file issues with Tailscale, there is no point for us to put work on them if it is a problem with Headscale.

[MeCJay12]

@kradalby While I have not used Tailscale SaaS, I opened a bug with both because it seems like a bug in the Linux Tailscale client.

[kradalby]

I opened a bug with both because it seems like a bug in the Linux Tailscale client.

If you have not tested it with Tailscale, it might only be a Headscale until you have tested that, and we try to maintain a good relationship with them by not driving more work than needed towards them.

The bare minimum would be to first establish if it happens with both, and then you can escalate it to Tailscale.

Generating noise in their issue tracker that they cannot debug because it is not connected to their infra is putting us in a bad light.