Dynamic Configuration for Golang Data Catalog + Parquet vs. Custom Provider

[max-romagnoli]

My organisation is looking to introduce pygeoapi in our stack to serve Features (and possibly in the future raster, tiles, etc...).
In our set up we have a Data Catalog client exposing metadata from all our datasets; and different clients, among them a tabular Feature client, exposing the actual data.

1: Custom vs. Parquet Provider for Features

In our Feature client SDK we already implement some of the querying and filtering specified in the OGC Features standard, although it is not a 1:1 match. The Feature SDK serves Parquet files so we could either:

1. Create our own custom provider plugin that uses our SDK to serve Features through pygeoapi;
2. Point directly to our Parquet files using your ParquetProvider, letting pygeoapi do all the filtering and querying.

One issue with the second option is that ParquetProvider requires an ID and not all our datasets have one, also not all our datasets have a geometry field.

Would you have any insights on which option to go for?

2: Golang Data Catalog Configuration

The Data Catalog client is in Golang and returns the metadata that we would need to configure pygeoapi. In your documentation, there are examples of static YML files where you can specify resources and data providers. Our thought was to generate this dynamically by writing a collection provider that fetches our datasets metadata from Go at container start-up and populates the pygeoapi configuration accordingly. Here we would also specify which provider is used (Parquet for features, or in the future other providers for raster) and the path to files (or to the custom plugin depending on which option we go for above).

Would you have any insights on best practices to dynamically configure pygeoapi using the clients in our current setup?