This is a POC that implements two algorithms for automatically inferring relationships for a set of namespaces using some pretty straightforward heuristics as described in the architecture design doc for the data modeling project

• First algorithm is from foreign to local (I'm mostly including it for posterity and will probably drop before this lands in main): we start from _id field in a collection and then search through all the fields with matching schemas in other collections. This one is pretty greedy and might be especially expensive if _id is of some very generic type like a string or an int. There is also a bigger chance that an _id field from foreign collection is missing from the local one (for example you have users and comments, and user haven't left any comments yet).
• Second algorithm is from local to foreign: we look at indexes and any field that is objectid and assuming that if index exists or objectid was used, this is a foreign key in a local collection. This algorightm might return less results if indexes don't exist or _id field is of a custom type, but it has a benefit of being less expensive in terms of execution as the find operation always runs against _id. Also because we're starting with foreign keys from the local collection, there is a way higher chance we will find them in a foreign collection compared to the first algorithm

WIP, but that's already working and I'll continue adding some unit test for smaller functions that are used interchangeably across both methods while we're scoping the work.

For the purposes of testing this, I added some views and indexes to sample_airbnb and sample_mflix to our test cluster (Cluster 0) that you can try this out on