[tech] [committee] Temperature Monitoring in Server Room [repost]

[Michael Deegan]

Afternoon,

On Tue, Mar 19, 2019 at 03:55:24PM +0800, Andrew Williams <andrew at ucc.gu.uwa.edu.au> wrote:
> > As for what UCC does, are you saying that the club should choose externally written code over the code written by a club member?
> > 
> > I've had a look at Icigna2, and it seems really good, to be fair. But it would not give me the experience of interacting more directly with the hardware and writing some interesting code.
> > 
> > And it would not encourage others to make the type of home-grown solutions that encourage learning.
> 
> It depends what people want to learn - having experience with a 
> widely-used server monitoring tool would be good experience for anyone 
> wanting to get a sysadmin job in a large organisation. Working with 
> hardware and rolling your own code is good experience for anyone wanting 
> to be a developer.

Agreed; if the aim is learning (or you simply have time to burn) then
definitely try the DIY path.  Though most people eventually reach the point
where they prioritise functioning systems over the satisfaction of having
done everything themselves, and if someone else has already written a tool
that does the job adequately, then why not?  :P Also, do try to avoid "Not
Invented Here" Syndrome, which is a Thing.

> > One thought - running such a thing on a VM is a bad idea given it should run on every single machine to report its own internal state, not all machines can run VMs, and that will waste a hell of a lot of CPU and resources.
> 
> You _can_ run an instance of icinga on every machine, and set them up as 
> a cluster, but you wouldn't run each instance in a VM - that way it 
> could only report the state of its own VM, not the host machine.
> 
> In practice, I found it easier to run one instance of the icinga process 
> and web server (which we have a VM for, but you don't need one). That 
> process then uses SSH to run the plugins on each machine - essentially, 
> to monitor disk space, for example, it's doing the equivalent of 'ssh 
> foo at bar "df"' once every hour, or day, or whatever interval you specify. 
> You need to set up a monitoring account with the icinga public SSH key 
> on each monitored machine, to accept the SSH connections without a 
> password.

If you're looking for performance graphing, in addition to alerting, then I
recommend taking a look at Munin (which is quite easy to set up, and
requires a small agent installed on each monitored host), Cacti (which
generally collects its data via SNMP), or MRTG.

HTH,

-MD

-- 
-----------------------------------------------------------------------------
Michael Deegan               Hugaholic               http://www.deegan.id.au/
------------------------  Jung, zr jbeel?  ----------------------------------