On Feb 5, 2013, at 11:13 AM, Andreas Ericsson wrote:
> On 02/05/2013 06:04 PM, Daniel Wittenberg wrote:
>>=20
>> So it almost seems like you could have a remote worker installed on
>> each host, and just have it check that one host=85other than you'd have
>> a ton of remote workers, what would be the downside of doing that?
>=20
> None, apart from the fact that in that case you'd probably rather have
> each worker containing its own scheduler and submitting its results
> once every minute or so. Preferrably via udp to avoid having to
> maintain a bazillion network sockets. We'll hit the open filelimit
> really quickly otherwise, and the various input polling interfaces
> tend to fall apart or grind to a halt when handling more than X
> sockets (where X is usually a number between 1024 and 32768).
>=20
>> Removes the need for nrpe or other options for remote checking=85
>>=20
>=20
> It does indeed. It also removes the need for a powerful monitoring
> server. With "onsite" workers and a special-purpose snmp worker
> (or two. Or three), you could monitor hundreds of thousands of
> hosts from a laptop.
Yeah, that was my thought, and also make the config a little simpler and ea=
sily dynamic for cloud solutions. What would be really cool is if you coul=
d just have a worker report in for what he wanted to monitor and if core do=
esn't have it just add a new host based on a template
So when new cloud=systems come online they could automatically get registered=85back to the =
dynamic adding I guess=85
I would also imagine anyone with a larger scale setup has already raised th=
e ulimit settings, that's usually one of the first things I do, but might s=
till hit some system limits depending on the size, but would be interesting=
to see how far you could scale that type of setup.
We'd have to make sure we could also handle having a worker check in for a =
node that core might not know about just yet, so in case the server stand-u=
p happens faster than the config updates. Just making sure it doesn't blow=
up or something when that worker checks in.
Dan=
This post was automatically imported from historical nagios-devel mailing list archives
Original poster: daniel.wittenberg.r0ko@statefarm.com