System jobs, Stable and auto revert

Can someone confirm that system jobs support auto-revert? The docs hint towards rolling updates being supported, but in my experimentation with the two job types (one defined as service, another as system), a system job version never gets marked as Stable (while service does) and as such when an updated deployment of a system jobs fails, there is no auto rollback to the previous stable version (as there never was a ‘stable’ version)

sample jobs attached (that spin up a python http server). Nomad 0.10.4

[vagrant@server1 ~]$ nomad job history http-svc
Version     = 0
Stable      = true
Submit Date = 2020-03-25T02:58:38Z

[vagrant@server1 ~]$ nomad job history http-sys
Version     = 0
Stable      = false
Submit Date = 2020-03-25T02:59:04Z

To test failure is as simple as updating the raw_exec config args to add a typo in the python module name, for example:

      config {
        command = "/usr/bin/python"
        args = [ "-m", "SimpleHTTPServer-oops", "8000" ]
      }

http-sys.hcl.txt (427 Bytes)
http-svc.hcl.txt (443 Bytes)

http-sys.status.txt (1.7 KB)
http-svc.status.txt (2.0 KB)

I stumbled on this thread today after reporting similar behavior here System Jobs never marked as stable and redeploys occur · Issue #9804 · hashicorp/nomad · GitHub

Interesting…