[collectd] Configure reporting interval on network plugin

Yves Mettier ymettier at free.fr
Tue Feb 16 19:04:45 CET 2016


Hello,

Isn't this a InfluxDB feature (or issue) ?
I have already used collectd->collectd : no problem. And 
Collectd->influxdb (using the write_tsdb plugin patched with pull 
request #1107) and I have seen the same latency as you.
Not sure that this is a network plugin issue.

You may try to dump the network traffic (tcpdump, wireshark) and check 
if this is a collectd or a InfluxDB issue.

You may also try to send the metrics to both InfluxDB and another 
Collectd (with network plugin as the input plugin and something like csv 
output plugin) and see if the csv output is as slow as InfluxDB.

Regards,
Yves

Le 2016-02-16 18:00, Joel Serrano a écrit :
> Hi Marc,
> 
> We are collecting data every 10 seconds, and its being sent every 2-3
> minutes (that must be how long it takes to fill the buffer).
> 
> We are not losing metrics as in influx we can see the correct timestamp 
> of
> each event with a 10 second gap between them. We only wanted to report 
> the
> metrics faster, so per your answer the way would be to decrease the 
> buffer
> so the metrcis are sent every more often.
> 
> I'm going to update the github issue with this info.
> 
> Thanks for your help!
> 
> Now everything is clear.
> 
> Best regards,
> Joel.
> 
> On Tue, Feb 16, 2016 at 8:39 AM, Marc Fournier 
> <marc.fournier at camptocamp.com
>> wrote:
> 
>> Excerpts from Joel Serrano's message of 2016-02-15 20:41:58 -0800:
>> >
>> > Has anyone successfully configured the network plugin in collectd to send
>> > metrics every X seconds? (like 10 seconds).
>> 
>> You can't control how frequently collected data is flushed to the 
>> write
>> target(s). Only how frequently read targets are probed.
>> 
>> In the case of the network plugin, the data gets flushed once its 
>> buffer
>> is full, which size defaults to 1.5kB. Are you collecting very little
>> data with a relatively high read interval ? If not you might want to
>> fire tcpdump to check you aren't loosing packets (remember the 
>> protocol
>> is UDP-based).
>> 
>> Note that the values dispatched include the collection time timestamp.
>> So even if there is some delay before they reach their final
>> destination, they should be stored with the correct time reference.
>> 
>> Cheers,
>> Marc
>> 
> 
> _______________________________________________
> collectd mailing list
> collectd at verplant.org
> http://mailman.verplant.org/listinfo/collectd



More information about the collectd mailing list