Getting usage of tables in queries

projection_usage would show you projections used, and you could easily infer what table that was from the projection name.

But the better way, IMHO, is to retain query_requests, or query_profiles, because generally these tables only have a few days' or weeks worth of data in them. You can increase the retention on these tables, but it's better to just run a cron process to just copy the data from these tables into a permanent table that you can manage yourself.

Then slap a text index on the query field. Then you can run queries like how many times a given table was referenced in a query, or when the last time a table was used, etc. And because of the text index, you're not parsing a giant query string. The text index will make it a sub-second query, versus one that might take about 10-20 seconds otherwise. Powerful stuff.

BTW, that'll need to be a table you own (not query_profiles) because a text index requires a unique primary key, and you can't create that on query_profiles.