KIbana Visualization support order by percentile metric

[nkhatwani-unicommerce]

Describe the feature:
If we do term aggregation on one axis and 99 percentile on another axis, currently we cannot sort results by 99 percentile values. For more details refer to this discuss item

Describe a specific use case for the feature:
Sorting results by 99 percentile could be used to show Top N slowest performing API's, considering es storing log records.

[elasticmachine]

Pinging @elastic/kibana-app

[monfera]

From @timroes:

That unfortunately isn't supported right now. The reason for that is, that the percentile aggregation is a so called multi metrics aggregation and can return multiple values (e.g. for different percentiles you specified). Since there potentially could be multiple values, Kibana currently doesn't allow sorting by them. Please feel free to open a feature request on GitHub for that feature.

[s-bauer]

This is really an important feature in my eyes. I always use percentiles instead of a plain average, to sort out strangely high values that only occur once in a while. It sucks that I cannot sort the charts after the percentile value to e.g. show the top 5 slowest queries. What I do instead, is to use the average for sorting, but this can be very different to the 95 or 99 percentiles.

[hydrojumbo]

This is a critical feature for using Kibana for DevOps: every quantitative metric about service instance health (response time, cpu/memory utilization, network i/o, etc) is reported with some service instance ID (aka "Term"), and like @s-bauer points out (and Werner Vogels writes about), looking at different percentiles is the key for really understanding the experience your service is providing to its clients. Sorting by average is not the same for the same reason we do not manage services based on average performance!

Note: I believe this is also behind why you cannot sort by Median either.

[Ethan777100]

i agree with the importance of such a sorting avenue too.

i have a use case on my end whr im aggregating 100+ data buckets, reflecting 3 bars per bucket: 2, 50, 98th percentiles.

because no such sorting is available, i dont even know what's logical and sensible to sort my data buckets by. Im currently sorting by the Avg of each bucket - but that doesn't hold much water at all. Ppl dont want Avg all the time. Sometimes they might want to see the "TOP20 best performing / TOP20 worst performing" wrt a percentile

PS: i have also tried sorting "custom metric"= "percentile rank" & sorting by Alphabetical order. The results dont look correct.

hope sorting by percentiles can be implemented.

[GregoryLamoureux]

+1

This is a useful feature for my users as well. If you have many terms on the x axis then it's simply not possible to show all of them in one visualization. Due to inability to sort for percentile, the visualization becomes misleading because the visible data is an arbitrary subset.

Sorting using "custom metric" as "percentile rank" would be a workaround but it produces an error.

Thanks for your consideration! So far using ELK has been a blast and we are grateful for your support.

[dej611]

This feature is already supported by Lens Top values function. Do you think we can close this issue @stratoula ?

[flash1293]

Closing this issue because this is supported in Lens visualizations. In Lens individual percentiles are individual dimensions and each of them can be picked to rank the terms/top values funcion.

Since this would require a large development effort to implement in Visualize/TSVB and our efforts are focused on extending the functionality of Lens in easy-to-use and consistent way. We aim to provide the best possible visualization editing experience in Lens and are working to improve that tool and help users migrate.