indygreg / collectd-carbon Goto Github PK

DeriveCounters doesn't really derive the values, but merely takes the difference between adjacent samples. It would be a handy feature for collectd-carbon to actually dispatch the time-derived values: (V2-V1)/dt.

Issue #8 has more context.

The ipmi collectd plugin creates metric names containing spaces, open parens, and close parens: [ ()].

For example: 1320877928.collectd.HOSTNAME.ipmi.temperature.Temp processor (3-2).value

Graphite discards these metrics.

I suspect only the spaces are causing trouble--I will update this issue after additional research/investigation.

Related
http://collectd.org/wiki/index.php/Plugin:IPMI

On a whim I rearranged my network and figured this out.

I have a an "ops" server that runs carbon and collectd, and many clients running collectd. I have recently added graphite/carbon/whisper recently and was simply reading the collectd RRD files. The inefficiencies of updating all the RRD files encouraged me to change to using the carbon writer plugin, as a bonus I my graphs don't lag by 5 minutes. I changed the collectd ops server which was receiving all other client's metrics to write using carbon writer instead of the rrd plugin, which worked, except DifferentiateCounters didn't seem to take effect. Eventually I was able to discover that metrics received over the network were not subject to differentiation, but metrics generated on the local operations machine were; this was not expected. I changed a few clients to write directly to carbon, instead of through the ops server, and the feature works as expected.

I'm not certain if this can be "fixed", as I don't know if the plugin has enough information about metrics received over the network to handle rewriting.

To get expected behavior I have modified all my clients to write to carbon, and skip the "ops" collectd step.

First let me say thank you for the plugin you've written !
I'm using collectd 4.10.1, the default in natty 11.04 x64.
I've started a cpu stress test to see what data gets sent to graphite.
I'm graphing cpu system and cpu user values and I see them jumping over the 100% value.
I'm seeing for example user cpu value staying at 3000 while top show it at 98.3%
Have you seen this type of behaviour ?

Hi,

Do you know if its possible to have more than one LineReceiverHosts specified in the Collectd carbon_writer conf file.

If so how do you specify this.

Cheers

definition:
<LoadPlugin "python">
Globals true

<Plugin "python">
ModulePath "/opt/collectd-plugins/"
Import "carbon_writer"
<Module "carbon_writer">
LineReceiverHost "graphite.testing.com"
LineReceiverPort 2003
DifferentiateCountersOverTime true
LowercaseMetricNames true
MetricPrefix "test"
MetricSeparator "_"
TypesDB "/usr/share/collectd/types.db"

The graph tree i get:
http://d.pr/i/r9d5

Collectd 5.0.0 was released recently. Need to verify that plugin works there.