session lifetime unde tmux + ssh

[carlesoctav]

Continuation of #160.
I don't think this is true. I held a colab ssh session open from an always-on VM, and Colab purged the runtime after ~2 hours — while from my own laptop I can keep sessions alive for 4–8 hours.

One thing I noticed: there is no ping loop running in the newest CLI version, though I can still see keep-alive requests in the .log on my laptop (left over from an older version, I assume). Any idea why the behavior differs? @EvanWiederspan

[carlesoctav]

Well, this happens when I trigger some kernel action, like colab exec, on a dead kernel (not really sure why, maybe due to inactivity). An action on a dead kernel purges the local sessions, even though the SSH connection is still active. Not performing any action on the kernel doesn’t trigger the session purge.

[carlesoctav]

So there’s still some discrepancy between the Jupyter kernel lifetime and the VM lifetime. I can see that the VM is still running and consuming my credits even though the kernel is dead and unrecoverable. However, an unrecoverable kernel prevents the user from SSHing back in [the sessions get purged locally], which eventually leads to an inactive VM.

[carlesoctav]

nvm upgrading the cli solve the issue, problem: some command doesnt refresh the 1-hour token, and the cli think it was a dead session.