Sum of non_negative_derivative with group by

1

Hi,

I planned to use InfluxDB to monitor our HBase cluster with region-level metrics, where my data looks like:
> select * from “regionserver.Regions.readRequestCount” limit 5
name: regionserver.Regions.readRequestCount
time cluster host namespace region table value
---- ------- ---- --------- ------ ----- -----
1492152641000000000 sh host1 default 00461c76de8093dd93c31ba95bce8f7f table1 4.4121848e+07
1492152641000000000 sh host2 default 00e563b860ecc360f4dd686e4629b9f7 table1 1.3915744e+07
1492152641000000000 sh host3 default 03d503204f5a4307bc0c8a12926fd2bb table3 0
1492152641000000000 sh host4 default 08b463b37012bb4aae44bac78eda7aa2 table2 3.189501e+06
1492152641000000000 sh host5 default 118dbc704b724f0668f4cd2bf825afac table3 0

I would like to find out how many reads a table has during the last week.
So I should probably have something like
select sum(non_negative_derivative(“regionserver.Regions.readRequestCount”) from “regionserver.Regions.readRequestCount” group by table, time(1m) where “table”=‘table1’

But there are two problems:

  1. sum can only apply to field, instead of functions
  2. If i write like derivative(sum(“regionserver.Regions.readRequestCount”)), then it will first sum and then calculate derivative, which will have bad results when some point is lost, interpolation seems not working here.

Can anybody help with this? Thanks.

1 Like
2

Your query is close…do something like this:

select sum(non_negative_derivative) from (select last(<fieldname>) from regionserver.Regions.readRequestCount where time > now() - <sometime> and table = 'table1' group by *,time(1m)) where time > now() - <sometime> group by time(1m),table

You have to be careful when using these aggregate queries, as you can’t just do a select * operation on the whole data and expect a rate to be calculated.

This is a really, really good reference point: Data Exploration | InfluxData Documentation Archive

1 Like