CS // 29 Dec 2010 // Comments Off on Debian System Load monitoring

One or two of our servers have been a little bit overloaded recently.

They’re going to be replaced with beefier machines, but due to a number of issues I haven’t been able to replace them yet.

Issue #1 – Pre expo, we weren’t allowed to replace anything.
Issue #2 – Post Expo, I’m no longer allowed in the data center!

We’re working on sorting issue #2 out, but in the interim I need to keep the older machines running.

I was previously using Monit to monitor system load.

Monit would be a good solution – it has a web ui, it can stop services if system load goes too high, and generally works when everything else is failed. This is great when things go poopy, but it has one fatal issue.

It doesn’t know how to restart stuff if load is back to normal.
This typically means that something will put the server load into unusability for a sustained period of time (due to lots of visitors), monit will go ooh, apache has gone awol, and stop it.
Unfortunately if its back to normal, monit doesn’t have a way to start it up again, so I need to manually go to the monit page, and start the service. I do get emailed on things like this, but it leads to complaints from the 2 clients that appear to monitor their particular websites more than monit does.

So, I’ve been looking at other solutions.

One such solution is sysfence

While sysfence is severly underdocumented, the examples provided don’t even work!, and appears to be abandoned, it does do the job.
Sysfence is a no bells and no whistles precursor to monit, but it has that killer feature that monit is missing.

So, how do we use sysfence?

apt-get install sysfence

Will install it, but unfortunately no config is installed.

So, start off by creating a /etc/sysfence folder

mkdir /etc/sysfence
cd /etc/sysfence

We’ll need to create a config file for it, so

pico sysfence.conf

My sample sysfence script is below (explanation underneath script)


rule "ApacheStop" {
la1 >= 10.00 or la5 >= 6.0
}
run '/etc/init.d/apache2 stop;'


rule "ApacheStart" {
la1 <=2 } run once '/etc/init.d/apache2 start;'

rule "warning" { la1 >= 8.00 } run once 'echo "Load High: BACKUP" | mail lawrence@computersolutions.cn'


I'm having issues with apache causing load to rocket, so I've setup some rules as follows:

If load average for 1 minute > 10 (ie server is going bonkers), and load average for the last 5 minutes > 6 then stop apache.
if load average for 1 minute > 8 send me an email.
if load average for 1 minute < 2 then start apache. This will only run one time if load is below 2. The documentation http://sysfence.sourceforge.net/ goes over how to write a rule. Note that the examples are broken;

eg
if {
la1 >= 8.00
} run once 'echo "SHOW FULL PROCESSLIST" | mysql | mail my@email.com'


Issue? All rules need to have a "rule name" specified.

So a corrected working version would be:

if "some rule" {
la1 >= 8.00
} run once 'echo "SHOW FULL PROCESSLIST" | mysql | mail my@email.com'


Back to our setup..

Now we've setup a ruleset, we need to run it. Calling sysfence /etc/sysfence.conf

Will run it as a daemon.

ps -ef shows our rulesets running:

root 7260 1 0 05:51 ? 00:00:01 sffetch
root 7261 7260 0 05:51 ? 00:00:00 sfwatch 'warning'
root 7262 7260 0 05:51 ? 00:00:00 sfwatch 'ApacheStop'
root 7263 7260 0 05:51 ? 00:00:00 sfwatch 'ApacheStart'

sffetch is the daemon, and sfwatch are the rules it runs.

As sysfence is quite rudimentary, you'll need to kill it if you change rules.

You'll also need to add it to your startup scripts or create one. I'll be lazy and not go over that right now. If people are interested, add a comment, and I'll put something up.

Sysfence can be downloaded here - http://sysfence.sourceforge.net/ (or via apt-get if on a Debian based OS)

---

Man page for sysfence below (note examples require adding "rulename" after if... { or rule ... {):

NAME

SYNOPSIS

DESCRIPTION

USAGE

EXAMPLES

AUTHOR