[collectd] Confusion regarding metric names

[Lindsay Holmwood]

Hi Denis,

On 3 June 2015 at 01:50, Dennis Jacobfeuerborn <dennisml at conversis.de> wrote:
> Hi,
> I'm trying to figure out how I can get collectd supplied data into
> influxdb and it's tag based format.
> My issue right now is that I'm not 100% sure I understand the metric
> format used by collectd. Take this metric for example:
>
> <host>.cpu-3.cpu-idle 99.600929 1433256464
>
> This seems to conform to the format
> "<host>.<plugin>-<plugin-instance>.<type>-<type-instance>". This value
> however:
>
> <host>.vmem.vmpage_io-swap.in 0.000000 1433256464
>
> seems to have a plugin "vmem", no plugin-instance, a type of "vmpage_io"
> and a type instance of "swap". What does the ".in" represent though?

The additional part you're seeing in the metric name is the data
source (DS) name.

The "<host>.<plugin>-<plugin-instance>.<type>-<type-instance>" format
you mentioned is known as the "identifier" for the measurement.

Some plugins (e.g. vmem, interface, load) will pack multiple values
into a measurement. Every value has a name that's determined by
types.db. If the measurement has multiple values, each will have a
unique name.

For example, the load plugin packs 3 values (shortterm, midterm,
longterm) into the measurement, so the corresponding metric names
being passed to InfluxDB would be:

 - example-host.load.load.shortterm
 - example-host.load.load.midterm
 - example-host.load.load.longterm

You can see which plugins emit multiple values by checking out
types.db, which specifies the names for each of the data sources in a
measurement.

https://github.com/collectd/collectd/blob/master/src/types.db
https://collectd.org/documentation/manpages/types.db.5.shtml

You may be wondering why other plugins (like the cpu plugin) don't
have value names? The vast majority of collectd plugins will only emit
a single value in their measurements and simply use "value" as the
data source name.

How the DS name gets stored can depend on:

 - The behaviour of the write plugin in collectd. For example, by
default the write_graphite plugin will only pass include DS name if
there is more than one DS in a measurement (although this behaviour is
configurable in the plugin). Alternatively, the rrdtool plugin will
store multiple DS in a single file that matches the metric identifier.
 - The behaviour of the collectd network protocol listener. If you're
pushing collectd packets directly to InfluxDB's collectd input plugin,
it has its own logic for how that data gets serialised out into
InfluxDB.

Check out these pages on the collectd wiki for more background on
collectd's internals:

https://collectd.org/wiki/index.php/Naming_schema
https://collectd.org/wiki/index.php/Value_list
https://collectd.org/wiki/index.php/Data_source

Cheers,
Lindsay

-- 
w: http://fractio.nl/
t: @auxesis