Re: health monitoring scripts

[Roland McGrath <roland at redhat dot com>]

> Besides needing more information sources, we also need to think about
> what makes a system "unhealthy".  For instance, in the case of
> context_switches, the health monitoring code could check for too many
> context switches within a certain time interval.  Of course the hard
> part is knowing what is "too many" (or at least how to make it
> configurable).

That's probably a later fancy feature.  The basics of health monitoring is
tracking all the right "vital signs".  Then you can start with feeding that
into visualizers so you can eyeball "this graph looks real spikey".  Then
from there you can start coming up with the rules for turning those gauges
into idiot lights.  Having all the most useful gauges on the dashboard is
the biggest step.  After that, it's all gravy and each expert sysadmin or
management application writer has a book full of recipe ideas for gravy.


Thanks,
Roland