... in Pacemaker Explained. Note that container in this case is a generic term and does not necessarily refer to Linux containers in the usual sense. A resource that has `container` set to another resource is considered a filler resource of that container resource.

Document the restart-container value for on-fail, and that filler resource actions should use (and default to) it.

Various things that may be worth documenting:

• If a filler with on-fail=restart-container fails, its stop is ignored (considered done) and its container is stopped
• Filler fail-counts are propagated to container
• A filler is ordered/colocated relative its container
• Fillers are not probed
• This is used for guest nodes' implicit remote connections

Investigate how starts are handled (relies on mechanism outside cluster in container start?). Code implies that containers may not be nested, but not sure if anything prevents it.