#influxdb

Hello

we're going to use influxdb in our production enviroment

with collectd as data source

we have about 200 servers, and we're thinking of 60s interval between updates

but we are afraid that using 1 serie for each type of measurment will gonna kill influxdb performance

ie. when we'll try to query avg load from only 1 server in grafana

do we need to make separate series for each server and each type of measurment?

matei: what do you use to send this data?

Norrland: collectd on each server

it's pretty light and it can be installed without hassle from official debian repo

matei: ok. Doesn't that send hostname/tag along with the metrics?

Norrland: yes, it does. But writing data from all hosts to 1 serie won't affect performance?

matei: Not sure, but can't see why it should.

huge amount of data gonna be processed when we'll try to make query in grafana

If you would have one series for each server 'foo1_load1' 'foo2_load1' 'fooN_load1' it would be quite a hassle to handle that in grafana

Norrland: tbh at the moment we're using influxdb 0.8 with graphite data source named like "collectd.comatv_i_red-sky_pl.memcached.percent-hitratio"

and everything works fine ;)

but now we're trying to move to influx0.9/0.10

mkay. I'm still very new to influxdb. Running 0.9.6, and telegraf-statsd on my clients to collect statsd data.

Norrland: I'm currently running influxdb as a system to make graphs out of our monitoring with 3 databases now

one is being filled by Icinga, which has like ~200 servers I think

then I'm polling vcenter with a self-written python script for the io stats

I'm using 2 measurements, and everything is working properly

for example, all my read statistics, go in a measurement called "read" where I tag them with the hostname of the server the stats belong to, the kind of statistic (iops, bytes, latency) and the datastore

it works very well, and that's with only 2 vCPU's and 3GB of RAM

oh and icinga is running on the same machine (yes I know I should split them up already :P)

also, I'd recommend you to read this page: https://docs.influxdata.com/influxdb/v0.9/intro...

specifically this: Conceptually you can think of a measurement as an SQL table, with rows where the primary index is always time. tags and fields are effectively columns in the table. tags are indexed, fields are not. The difference is that with InfluxDB you can have millions of measurements, you don’t have to define schemas up front, and null values aren’t stored.

so the only thing you need to watch out for I think (performance-wise) is not tagging your data with constantly changing values. Ie. don't tag your data with its value

B1nny: ah okey, will have a look.

hi, has anyone had any success with clustering using 0.10-beta2?

hello

I'm running telegraf 0.10 and trying to collect Jolokia data, but I don't value store

is the syntax changed // https://github.com/influxdata/telegraf/tree/0.1...

at least "pass" is not supported anymore

i am having these two types of resources for Influx DB