Dealing with long-running sessions?

General Discussion
1

At our site, a few users have good reasons to keep their XFCE Desktops in OOD running for a few days (workloads that are both interactive and long-running at the same time).

In some situations, however, they just leave the sessions on out of convenience, which is obviously not ideal. Rather than policing this behavior, I would like to provide tools to make it unnecessary and I have thought to running XFCE as a VM, so users could checkpoint and restart at any time exactly where they were (as a side benefit this might help for the long-running jobs too in case we need to do emergency maintenance).

I was wondering if anybody has experimented with such a thing and if so with what results. I am also curious if anybody thought about this and decided against if for some reason.

Last, but not least, has anybody addressed this issue in other ways (other than trying to police the users)?

2

Hi!

This is nothing less than the holy grail of interactive desktop computing :slight_smile:

if anybody thought about this

yes, a bit, the tools to build that are available..

  • openstack and nova suspend / resume to manage compute desktop VM
  • cgroups to monitor cpu cycles in time to detect idle sessions
  • abstraction in form of slurm disposable cloud nodes

The challenge is switching roles of bare metal servers between (dual boot?) openstack host and normal slurm compute, and scripting some state manager orchestrating reliably all that.

Indeed defining static openstack pool of hardware makes this quite doable, but wastes unused CPU time, so not solving the original problem definition.