I could use some advice! I have been trying to build a reporting engine on top of TS, and it seems, unfortunately, it’s not quite up to our specific use case. We are trying to group by a facet, and then return bucketed counts of facet values that with that value in common. For example, if I want to see two years+ of data, it would be >730 searches in a multi_search (not even sure that’s doable..). On top of that, each search has to be filtered by a list of facet values, potentially in the 100s of thousands of values… This is clearly not going to work, but was a fun exercise to see if it could be done 🙂 The kind of data we have looks like this:

{
  "activityKind_facet": "activity.platform/reaction",
  "channelId_facet": "d1541dde-47b6-444c-b2ce-52668dd8d3cd",
  "channelKind_facet": "chat",
  "communityId_facet": "8e914cd4-a155-40cb-84e7-f476d3429d5c",
  "date_facet": 1656028800,
  "personaId_facet": "62f13402-1393-4963-a0c0-89d10e1237f3",
  "primaryLabel": "Added an emoji: :airplane:",
  "timestamp": 1656087039
}

..and what I’d like to be able to do is group by

date_facet

, and have counts of other facets by that group, filtered by (potentially) a large number of

personaId_facet

, as I mentioned above. I can see how we can manually build rollup tables of all this data, but it feels annoying that we have to define what we can get out ahead of time - any advice on tech that can tackle this kind of problem without too much hassle, from flat documents, with consistent performance?

I will recommend using Clickhouse for this. In addition to supporting a full fledged SQL engine, it will allow you to quickly create aggregated materialized views (https://altinity.com/blog/clickhouse-materialized-views-illuminated-part-1) that are kept upto date as new data is ingested. These aggregated views will be much faster to query so you can run real-time reports.