Hi Everyone, We have a use case where we want to write a Java Map object onto a pinot table. We have tried both writing it as JSON and flattening the map, creating a separate column for each map key. Using tables with a few million rows, when testing query performance, we noticed that using separate columns is much more performant than using a JSON column for the map, so we are proceeding with flatting out the map. As you would expect, since the map key maps to a specific table column which is of a specific data type, all values for the same key in different records have to be the same data type. However, we do have situation where the same key can be one of several data types. Was wondering if anyone else had a similar use case, and if they found a solution that works. One solution that comes to mind is to create columns based on both the key and the data type, so that if they key appeared as both a string and a long, there would be 2 columns, s_key and l_key. Seems pretty straight forward, but it complicates queries, in that we need to know what columns we have created and query against all of them (could be more than just these 2).

JSON should have performed similar, because both use inverted index underneath. What are the data types?

one of string, long, double, boolean we also have to retrieve/aggregate on the values, so we had to use extract_scalar, but I recall that failed if data types were not consistent. I'll have to touch base with the engineer who was doing this.