Derivative aggregator should take the difference between the last point of the current bucket and the last point from the previous bucket [EDIT] This issue will address queries with no group by time and group by time + fill. Issue #1030 has been created to fix group by without fill.

[jvshahid]

Currently derivative won't work with one point per bucket

[prune998]

Currently derivative won't work with whatever point per bucket if data is not changing often because the change happening between two buckets is lost

[ragzilla]

Interested in this (use case, storing 32/64 bit counters in influxdb). Current workaround of doing a group by(2*sampleperiod) throws away good data. I'm somewhat familiar with Go but have no idea where to start on this issue.

[michail-nikolaev]

same thing for "difference" aggregator too

[tcolgate]

This makes using counters with influxdb entirely impractival especially if you deduplicate the data before pushing (I'm using tcollector to send data). I shouldn't need to know the sample period in order to take a derivative (what happens if I change my sample period for operational reasons? I can no longer reliably query the data).
Calculating rates on counters is an important technique in monitoring, and really needs first class support (TSDB is actually quite good at this, but I'd much rather be using Influx :) )

[johnarnold]

I'd love to see an update on this as well...

[damm]

I believe this is the largest blockers from going to production with InfluxDB.

You can either use graphite-api to talk to influxdb; or modify how you collect so you don't ever have to use this function.

:(

Updated comment for clarification thanks @prune998

[prune998]

rephrasing that : until Influxdb have a working derivative (and others), you can't get rid of graphite... and can't go on production and grow Influxdb...

This should seriously be on your top priorities !

[duane23]

Agree with the above comments - this is a show-stopper for my org with the current behaviour.

[cnf]

I am running into this problem as well. As this is about half my use case, I must agree with the above assessment.

[cboggs]

+1

[chausat]

+1

I think calculating rate between each point for derivative if no group by time is given is very important and very very urgent.

[chausat]

"0.8.4 RC1" doesn't work!

when using "select op_num from test where cmd ='set' ", the result is:

when using "select DERIVATIVE(op_num) from test where cmd ='set' ", the result is:

ONLY ONE RECORD!!!

[cnf]

@chausat derivative can't work without a group by statement.

[moorereason]

@cnf Except for the fact that this issue explicitly references derivative "with no group by time."

If "group by" is required for derivative, the parser should throw an error.

[dgnorton]

If there's no group by it should calculate the rate of change between each point. There is a test in the DataTestSuite for this case that passes but I am able to reproduce what @chausat is seeing with different data.

[dgnorton]

@cnf @moorereason, My previous comment was wrong. Since there is no grouping, all points are put in the same bucket and it calculates the rate of change between the first and last value in the bucket. It always yields one point.