[c-nsp] ME3600x SNMP items for power supplies disappear when PSU is pulled

[Richard Hartmann]

(This is mainly for the benefit of others; ytti and me discussed this
on IRC as well)

On Sat, Sep 14, 2013 at 8:37 AM, Saku Ytti <saku at ytti.fi> wrote:

> This sounds like you're polling.

Correct.


> I pains me how little used SNMP traps are, which to me is the first check
> list item (before even ICMP ping) for functional NMS.
> [...]
> It's cheaper (computationally) and faster than polling.


Traps can get lost.

As we are not using traps, we would need to start listening for them.
And then design monitoring to ensure that we can receive traps.

Plus, even with SNMP inform, in a worst case scenario, the device will
retry sending the trap, but not be able to deliver. If that's because
of misconfiguration, it will be permanently silent with no way of
catching it.
So now we have to maintain a list of all trap-sending devices and make
them send a trap every 24 hours so we _know_ what's up.


Polling will either return a result you can trigger upon or not return
a result, which a second level trigger can then act upon.
While more expensive, it's an inherently safer design.



As per IRC, we are working with Zabbix to have their solution trigger
on certain SNMP polling items going into/being in "not supported"
internally.

In the meantime, it looks as if I will parse our "show foo" backup
database and look for missing PSUs that way. As those are
well-maintained and monitored aggressively we _know_ that the data in
there is both current and good.


-- 
Richard