Month over month calculation in visualizations

[AlonaNadler]

Often users want to compare in percentage between different time ranges, e.g. this month and previous month M/m or D/d to understand hows the change in the value of a specific metric as a percentage of the previous month’s value. (This_month - Last_month)/Last_month
This is currently possible in TSVB but not very straightforward using serial difference and bucket script aggregations

Provide an easy way in Kibana to visualize M/m in percentage where the users can define what is the interval which they want to compare (day, month, years, dynamic) and based on which metric (sum(revenue), average (bytes)

Below how can this be done in TSVB

[elasticmachine]

Pinging @elastic/kibana-app

[polyfractal]

The MovingFunction pipeline agg could do this a bit easier than serial-diff/bucket_script.

• window: 2 to get two values
• shift: 1 to shift the window so that it includes "current" and "current - 1" bucket values
• script to perform the month-over-month calculation

[AlonaNadler]

do this a bit easier than serial-diff/bucket_script.

is it easier because it uses only one function? or is it better from a performance perspective? or maybe something else :)

[polyfractal]

Easier because it's only one function, and essentially what moving_function was designed for (generating a metric based on a sliding window of buckets). Also keeps the response JSON tidier since you'll have the metric of interest directly and on "intermediate" steps.

Probably not relevant for this feature, but the other advantage is that moving_function allows different windows of time, not just the last month.

Pipeline aggs are essentially free (relatively speaking) since they are all "post-processing" events after the data has been aggregated, so performance between the two is negligible for the most part.

[wylieconlon]

I've looked into this a bit, and think this is part of a broader group of "time-shifted" comparisons. It's actually really similar to another discussion we've had about how to support cumulative sum with a starting value, and I built support for that into my recent spacetime project. The way it works is by letting users select a "relative time range" for each layer, and then having a way to compare the two layers:

The main benefit of this approach is that you can look at numbers over two completely separate time periods, for example "hourly metrics today vs hourly metrics one week ago today". Also, it scales to other types of queries.

[wylieconlon]

This implementation depends on how Lens will implement time offsets in general- there's a proposal for the general feature in that issue.