Tailscale authentication is lost by updating the NanoKVM application

[Bengt]

When I have Tailscale connected and update the NanoKVM application via the web interface, Tailscale is no longer connected:

# tailscale status
failed to connect to local tailscaled; it doesn't appear to be running

When I reauthenticate using the wizard in the web interface (settings / Tailscale), Tailscale works again.

[nwgat]

i also ran into this issue

[jasonafriedman]

I am running into the issue where authentication is being lost without updating after a couple hours. I am using nanokvm 2.1.1. Looking at dmesg it looks like there is an out of memory error.
[98979.301911] oom-kill:constraint=CONSTRAINT_NONE,nodemask=(null),task=tailscaled,pid=3684,uid=0 [98979.301964] Out of memory: Killed process 3684 (tailscaled) total-vm:1314096kB, anon-rss:62728kB, file-rss:92kB, shmem-rss:0kB, UID:0 pgtables:336kB oom_score_adj:0

[alexoughton]

I'm having the same issues with both the authentication and the oom-kill. It looks like it's important to keep tailscale at version 1.70.0 and not update it to 1.76.6 in order to avoid this issue. I was also able to downgrade back to 1.70.0 on a KVM experiencing this problem and it went away.

[mindstorms6]

@alexoughton how did you force Tailscale back down to 1.70.0?

[alexoughton]

edit for anyone newly coming to this: I instead recommend following @jasonafriedman 's simpler solution below, as this is more steps than were necessary.

Original response:

@mindstorms6 Some of these steps may not have been necessary, but this is what I had done while investigating the problem:

1. I moved /etc/init.d/S98tailscaled away to ~ to disable it, and rebooted the KVM.
2. I moved /usr/bin/tailscale to /usr/bin/tailscale-old, and moved /usr/sbin/tailscaled to /usr/sbin/tailscaled-old.
3. I went to the tailscale option in the web interface. It told me it wasn't installed, so I hit the "install" button. This shows an error but...
4. ...the correct versions of /usr/bin/tailscale and /usr/sbin/tailscaled were now in place. I moved /etc/init.d/S98tailscaled back into place and ran /etc/init.d/S98tailscaled restart.
5. Back in the web interface, I was now able to re-register to tailscale, and the admin interface for that showed I was back at 1.70.0.

As I said, this may have been more steps than actually needed, but stumbling through this as above got me into the right place.

[mindstorms6]

@alexoughton you're the best - that also worked for me - thank you!

[jasonafriedman]

For me removing the old packages and reinstalling through webui worked.
rm /usr/bin/tailscale and rm /usr/sbin/tailscaled

[crustovsky]

This worked @alexoughton thanks, there seems to be mistake in step 2 though, i think it should state to "and moved /usr/sbin/tailscaled to /usr/sbin/tailscaled-old"

[alexoughton]

Thanks @crustovsky,

Yes, that's right there's a mistake there which I've now updated. In any case though, following @jasonafriedman 's simpler solution is recommended here. I'll edit my comment for that.

[alexoughton]

Note that you will also need to run tailscale set --auto-update=false to prevent auto-updates from occurring again until this is fixed.

[ironicbadger]

@alexoughton where does the fix need to come from? sipeed side? or Tailscale?

[alexoughton]

@ironicbadger I haven't at all investigated why tailscale is getting oom-killed, only found this workaround. I'd say sipeed need to do that initial investigation, and then feed back to tailscale if it's just a memory leak on their side.

[ironicbadger]

Lmk as I am a Tailscale employee