Description of problem:
Qdisk does monitoring of external resources for a node in a cluster. Of course,
so does the watchdog daemon.
The request is to allow qdisk to use the watchdog's internal status as the sole
heuristic for fitness for participation in the cluster - thereby reducing the
load as well as duplicate effort on the system.
Qdisk currently uses blind heuristics based on external means, which gives it
great flexibility. The watchdog daemon uses internally-coded ones, which gives
it better operational capability in high-memory pressure or high-load situations.
This will require changes to the watchdog daemon.
Generally, what I'm after here is a way for other apps to read from the watchdog
daemon's internal state.
This functionality may already exist, however, the specifics are that it should
be a low-overhead method - like shared memory or something - so that we can be
as memory-neutral as possible (compared to a socket which allocates buffers, for
Moving this to 5.3 unless Marek has something ready.
This could be stabilizing to the watchdog daemon.