If a plugin is initially created with COUNTER and no .max, there may be spikes in the back-end RRD database when the counter goes back to zero. There's a simple way to fix this.

First, the plugin must be updated to present a '.max' value.

Once that has been accomplished, and the munin-server has polled the plugin at least once, the munin-server datafile will have the .max value stored for the plugin.

Then the following perl 1-liner can be used to update all matching RRDs. Replace DOMAIN with the name of the Munin domain being polled, and replace PLUGIN with the name of the plugin RRD databases that need updating.

cd /var/lib/munin/DOMAIN
perl -ne 'next unless /:PLUGIN/; if (/.*;(\S+):(\S+)\.(\S+)\.max\s+(\d+)/) {foreach (glob "$1-$2-$3-?.rrd") {print qq{File: $_\tMax: $4\n};qx{rrdtool tune $_ -a 42:$4};qx{rrdtool dump $_ > /tmp/rrdtool-xml};qx{mv $_ $_.bak};qx{rrdtool restore -r /tmp/rrdtool-xml $_};qx{chown munin:munin $_}}}' ../datafile

An example, updating the MAX value in all the xvm_ plugins that have been updated to report a .max value:

cd /var/lib/munin/
perl -ne 'next unless /:xvm_/; if (/.*;(\S+):(\S+)\.(\S+)\.max\s+(\d+)/) {foreach (glob "$1-$2-$3-?.rrd") {print qq{File: $_\tMax: $4\n};qx{rrdtool tune $_ -a 42:$4};qx{rrdtool dump $_ > /tmp/rrdtool-xml};qx{mv $_ $_.bak};qx{rrdtool restore -r /tmp/rrdtool-xml $_};qx{chown munin:munin $_}}}' ../datafile
File:    Max: 34359738352
File:    Max: 34359738352
File:   Max: 25769803764

This is a distillation of the following process, outlined by Greg Connor in a post to munin-users:

Here is a quick recipe for getting rid of spikes.

In this example "dnscache" is the name of the plugin and "" is one of the affected hosts. The desired MAX value is 1000.

Verify that the plugin now sets the right .max parameter. Fix the plugin if needed.

# cd /var/lib/munin
# cat datafile | grep dnscache | grep max

View the rrd file with "rrdtool info"

# rrdtool info DNS/

to look for

ds[42].max = NaN

This indicates that the plugin didn't have a .max defined at the time the rrd file was started. If you don't mind losing the data, you can delete the RRD files at this point and the new ones will be created with the right max.

If you want to save the data, first modify each .rrd file so that it has a max for datasource 42 (not sure why it is always 42, probably a tribute to d.adams) (Best to do this right after an update like at :06 :11 or so)

# bash
# for j in `find /var/lib/munin -name "*dnscache*-c.rrd"`; do \
rrdtool tune $j -a 42:1000;

*NOTE:* Replace 1000 with the desired MAX value.

Finally, dump each rrd file to xml and restore with -r flag. There will be some output to let you know which data points were dropped and replaced with NaN.

# bash
# for j in `find /var/lib/munin -name "*dnscache*-c.rrd"`; do
rrdtool dump $j > /tmp/xml ;
mv $j $j~ ;
rrdtool restore -r /tmp/xml $j;
chown munin:munin $j ;

If you are impatient, rebuild one host's graphs and look at it, or just wait 5 min and check.

# su munin -c "/usr/share/munin/munin-graph --nolazy --host DNSOverview "

[ Based on a post by Greg Connor, 2004-05-07 post to munin-users ]

Last modified at 2013-04-22T09:49:08+02:00 Last modified on 2013-04-22T09:49:08+02:00