#influxdb

mre__: you can write to any node of the cluster, it will automatically right to the correct location

is there any way to seed up 'list series' and 'select' queries in influx?

Civil: list series should be really fast in 0.8

Dieterbe: for graphite with 5 lines and 500k series it tooks me around 300ms for each

what's a line?

it does several list series /^metric.name/ queries

each takes 1100ms right now to fetch

because I've got 638688 series in influx

hmm, for me right now it takes 829.029148ms

yup, and that's a bit too slow :)

each graph takes seconds and seconds to draw :)

yeah, i thought it was much faster, weird

maybe cause all my lists are cached

Dieterbe: also when graphite gets list of metrics, I think you should use this cache to look through for metrics in the leafs.

and also for particular queries

Dieterbe: also as for me, list series is slower then select *

around twice slower :)

'select *' returns in 0.5 ms for me

Dieterbe: btw, can you have a look at a patch for graphite-api, before I send PR?

http://paste.ubuntu.com/8177710/ - it should fetch all the data in one query even for multiple targets

with it

i don't have time for this

ok...

Civil: what did you mean with looking through metrics in the leafs

Dieterbe: it's better to run 'list series', save it for some significant amount of time and then when you get "assureSeries" call - look through cache

It'll have worse performance for first call, but better performance until cache is expired

at least right now

and do the regex matching in python?

yup

that's how it used to work, actually

and it was slower then call 'list series' ?

huh? that approach used 'list series'!

er... I mean now I see that graphite_influxdb executes 'list series /^/' for each metric.

I mean maybe it'll be better to have full 'list series' in cache and match it in python?

in my use case, the change i did made things faster

here's an overview of perf measurements

but the measurements there for list series /regex/ i had around 150~300ms

now (see above) it's 830ms.. so that's a big difference. not sure where that's coming from

maybe there was some performance degradation

somewhere

maybe because my influxdb is more loaded now

Dieterbe: I'm getting 1sec on almost fully idle influx

weird

you query on the same machine that runs influx right?

it receives only one select from graphite

Dieterbe: yes

so you're not counting network traffic time

yup

in terms of inserts - I've got something like "30 metrics every 30 seconds"

oh wow

only this machine itself is monitored right now

anyway i'm sure 'list series' and 'list series /regex/' speeds can be improved in influxdb

Dieterbe: yeah, that's true. But right now for me to draw graph it takes around 7 seconds.

and for dashboard it's very-very slow...

that's why I'm curious is there a way to improve list performance

afk

ahh

i wonder if it's because they introduced sorting recently

Civil: https://github.com/influxdb/influxdb/issues/884

Dieterbe: maybe, that can be true

thx

Dieterbe: and why they decided that series should be in sorted order?

maybe I should revert this patch for my current installation...