[Linux-HA] Shortest monitor interval practical?
beekhof at gmail.com
Thu Sep 20 01:11:12 MDT 2007
On 9/19/07, Doug Knight <dknight at wsi.com> wrote:
> What is the shortest monitoring interval that's practical? I know it
> relates to system performance, but assuming a properly sized hardware
> platform, and a reasonable loading, what are people's experiences with
> short monitor intervals? Is it possible that a short interval could
> cause monitor requests to get queued up, or does heartbeat wait for a
> response from the monitor request before issuing the next one? How does
> timeout value factor in (i.e. is there any rule of thumb for its
> relation to monitor interval, shorter longer, doesn't matter)? In our
> case, we are considering monitoring a resource every 15 seconds, with a
> timeout of 10 seconds, but may need to ensure a restart in a shorter
> period of time.
i believe we previously established that regardless of the interval,
requests will never get queued (since the interval starts at the point
the previous operation finished, not started).
so any interval should be acceptable (well perhaps if the monitor op
was causing a significant load you might want to rethink it).
timeouts are a different story, if an operation hits the timeout, then
it (and therefor the resource) is considered to have failed and will
be restarted (and possibly moved).
timeouts should really be set for worst case scenarios... ie.
monitoring under high (cpu|disk|network) load
hope that helps
More information about the Linux-HA