CODEWITHSUNDEEP
graphite
shinyhappydan
shinyhappydan

Reputation: 728

Split count metric by delta across time interval in Graphite

I have a counter metric, which I will call a.metric.count

When graphed normally this would be an ever-increasing upwards line, which is not that useful. What I would like is to show the number of events which occurred every hour.

Upvotes: 3

Views: 3278

Answers (3)

Sandwich
Sandwich

Reputation: 61

Also using diffSeries, but results in a little bit of a nicer graph:

diffSeries(
    sumSeries(a.metric.count),
    sumSeries(timeShift(a.metric.count, "1h"))
)

If you want to avoid the ending spike you can use a double offset:

diffSeries(
    sumSeries(timeShift(a.metric.count, "1h")),
    sumSeries(timeShift(a.metric.count, "2h"))
)

Keep in mind that with this method you will be behind by the smaller offset.

Upvotes: 2

fourk
fourk

Reputation: 2272

Use the derivative function, as documented here: http://graphite.readthedocs.org/en/1.0/functions.html#graphite.render.functions.derivative

That will, at each data point, show how much the value increased from the previous timeseries datapoint.

If you want to view it on a per-hour basis, you can use summarize, as documented here: http://graphite.readthedocs.org/en/1.0/functions.html#graphite.render.functions.summarize

So, to view the hourly increase in .count, you could do:

summarize(derivative(a.metric.count), "1hour")

Upvotes: 8

shinyhappydan
shinyhappydan

Reputation: 728

First of all we need two different graphs:

  • a.metric.count
  • timeShift(a.metric.count, "1h")

Next we need to summarise these by hour, using the last value recorded for each hour:

  • summarize(a.metric.count, "1h", "last")
  • summarize(timeShift(a.metric.count, "1h"), "1h", "last")

Now we just need to show the difference between the two graphs:

diffSeries(summarize(timeShift(a.metric.count, "1h"), "1h", "last"), summarize(a.metric.count, "1h", "last"))

The only trouble with this approach is that the data for the latest hour is wrong because there is no "last" value on the non-timeshifted graph, which means that the graph will spike

Upvotes: 0

Related Questions