Create percentile aggregates from pre-aggregated data

[davibowinnow]

Hello there!

Context

We are using Timescale as a metric/observability backend for OpenTelemetry metrics.

One of the types of information that we'd like to collect is latency histograms. From the source over OpenTelemetry these get represented as an already aggregated data type; so currently our database tables where these are collected have the following columns:

• min
• max
• count
• sum
• histogram, which is a JSON type with 2x arrays named: bucketCounts, and explicitBounds

What our aim is

We'd like to have a way of turning this information into an easily aggregatable structure, that supports percentile approximation and similar functionality.

Timescale Toolkit already has this capability - uddsketch and tdigest, but these structures can only be created from raw datapoints, not the already pre-aggregated form that we have.

Feature request

We'd like to ask for the ability to either

• turn our pre-aggregated data structure into uddsketch or tdigest
• OR have a similar construct introduced that works with OpenTelemetry data types

Alternatively, if you can suggest a different approach to how to solve this problem, we're open to considering that.

Many thanks,
David