NMIS relies primarily on SNMP and ping for collecting measurements of your infrastructure's health, but there are situations where this is insufficiently flexible: not every interesting device out there is SNMP-capable, and there are concepts and services that do not directly fit into the SNMP universe.
In the Managing Servers and Services with NMIS8 document we describe what mechanisms NMIS itself provides for managing non-SNMP services: e.g. checking ports, looking at process status, checking the DNS and checking textual protocols with send/expect scripts - and the recently added capability to run external programs for getting a service's status.
This page describes a more generic approach to this kind of problem which doesn't rely on custom features programmed into NMIS: instead we show how to extend the standard Net-SNMP snmpd with a script or program of your choice, to make an arbitrarily non-standard 'thing' accessible via SNMP (and thus available to NMIS).
Use-cases for this infrastructure include collecting statistics from services that don't offer SNMP (e.g. the bind DNS server), capturing the status of multi-component services (e.g. email end-to-end) and so on.
snmpd and pass_persist programs
snmpd's manual page describes a number of extensibility mechanisms, one of them called "pass_persist programs": snmpd starts that program and delegates an OID subtree to it. Whenever it is queried for variables in that subtree it forwards the request to the pass_persist program which provides an answer. As the communication is very simple (write to the program's STDIN, read from its STDOUT) it's a very flexible way of capturing custom things; It's also very efficient because the pass_persist program is running permanently and there is no repeated startup overhead, and the program can do whatever it needs to do, whenever and however it wants to.
There are a few caveats:
- The documentation for this snmpd-to-program communication isn't complete - a blank 'command' is meant to tell your program that it should terminate.
- You'll need to pick an unused OID subtree to attach your script at; the Net-SNMP documentation recommends you use .1.3.6.1.4.1.8072.2.255 or .1.3.6.1.4.1.2021.255.
- The snmpd will block until the program has responded. This means your program needs to perform its operations in a non-blocking fashion, or your snmp infrastructure will suffer badly.
An example pass_persist program
Here is an example program in perl, which reads /proc/loadavg every 42 seconds and makes this information available at .1.3.6.1.4.1.2021.255.1 to .3, without blocking your snmpd.
Download: passpersist-example.pl
You would use it by adding the following line to your snmpd.conf:
pass_persist .1.3.6.1.4.1.2021.255 /wherever/you/put/your/passpersist-example.pl
It's quite simple, commented and less than 200 lines of code. Feel free to use it as a template for your own extensions.
Where to go from here
Once you have captured your custom measurements and you've tested the snmpd-pass_persist interaction with snmpwalk or snmpget, the next step would be to extend the most appropriate model with your new measurements. This part is a straightforward modelling exercise, and you will find ample documentation in the NMIS section of this site, and lots of examples in the models-install/ directory of your NMIS installation.