Claculation of Transactions per minute / Request per minute


how calculates the kibana apm ui the metrics for transaction or request per minute.

For example.

I have custom span test with the java agent. It writes one transaction per minute. The "transaction per minute" inidcator varies depending on the choosed time slice.

Time slice / last x minutes Transaction per minute
24 hours 1
60 min 2
15min 6
10min 12
5min 60
1min 60

That's little confusing

Which version of Kibana are you using? With the latest version I don't see that behavior.

Apm 1..6.1
Kibana and Es 7.0

I have tracked my jsf mapge with a single request. There is the same.

Last one hour

Last 30 min

Last 15 Minutes

Last 5 Minutes

That are all the same sampling.

That looks like a bug in the UI to me.
/cc @sqren

The average rpm underneath the chart looks more consistent, do you agree?

It’s expected that the rpm will differ greatly when “zooming out” if you don’t have continuous data.

Eg. if you make a request every second then there will be 60 requests made after minutes. Doing this for 5 minutes will create 300 requests. That is 60rpm for both the 1min interval and the 5 min interval. However, if you look at it over a 24hour interval it would be 0.2rpm (300 requests / 24 hours).

At first, I was confused by the screenshots of your graphs which show 120 rpm for last one hour and 2000 rpm for last minute. But that’s for the same reason - The UI does not show the max rpm for each time bucket but the average. Especially if looking at narrow spikes, the rpm is high when looking at it at high granularity like 5m. But when the granularity is lower like 1h or 24h, one data point represents a larger time span and the spikes even out.

Yes sure. But That's realy confusing if the count of transaction per minutes increases if you make your time slice lower.

How do you will find outliners with that approach ?

I write 2^n transactions per minute until 8192.

That's expected behaviour

And that's the confusing one.

If it's so, then the name of the metric is choosed wrong I think. Isn't that the throughput istead of Transaction or request per minute ?

I agree that this is misleading and we're discussing alternatives. I'll keep you up to date.

Could you explain what the difference between throughput and requests per minute is from your point of view?

Throughput is IMHO exactly that indicator what Kibana APM is showing.

The per minute caclualation should calculate buckets of per minute over the given time range. So that you can see the real otuliners.

We've created an issue here:

1 Like

This topic was automatically closed 20 days after the last reply. New replies are no longer allowed.