My view is that you should do this through node_exporter on the
Prometheus server, even if you can get similar metrics from Prometheus.
The node_exporter metrics are directly exposing what the filesystem
says, and it is what the filesystem says that matters the most; if
Prometheus thinks it's not out of space but the filesystem thinks it is,
the filesystem wins and you have a problem.
(The direct filesystem metrics will also include usage from other things,
not just Prometheus. Perhaps in theory there shouldn't be any other usage,
but in practice sometimes things happen, including surprise gigantic core
dumps.)
- cks