On Wed, 21 Jul 2004, ti chi wrote:
> have many web customers monitored with pcp ?
> everything put on one master server with a directory
> for each customer (logger, pmie, etc.) so the
> customers are silo?ed so to speak into almost 100
> directories.
>
> this layout scale things well since have almost 700
> servers monitored. the problem is we want to
> coordinate pmie alerts/events that occur in the
> customer environments with our environment ? basically
> a global pmie.
>
> don?t know good solution since: not practical to
> copy/merge all customers pmlogger files to our system
> and run pmie on logger file ? this too much data and
> takes too long. not practical to have our pmie or
> pmlogger query all customer pmda ? to much bandwidth
> and slams pmie.
>
> so silo?ed environment solved scale problem but
> created uncoordination of events - any thoughts on
> good way to coordinate pmie for whole environment?
> thanks for any insight!
I would think you'd want to use as many distributed pmies as makes sense
in terms of network bandwidth and management complexity to filter the
data close to the machines you are monitoring, and then use the pmia alarm
mechanism to forward just the alerts to a central event clearinghouse.
Look at the pmie examples ... disk.00, disk.20 or uag.20 for hints on
how the "shell" action could be used to forward events to your central
alert management system ... of course you'd need a mechanism for sending
events on every managed system, and some way of accumulating and dealing
with those events at the central point ... e-mail is a quick and dirty
way of doing this to prototype the idea ... other similar schemes can
be devised with not much effort.
|