MongoDB and Python

Nothing. It's a matter of scale.

Start everything with PG.

It's less a matter of scale to me and more a matter of facilities. Given the ability to scale PostgreSQL in places like Amazon RDS into utterly enormous system proportions, way less about performance to me. Is your data heterogeneous at a record-level? Do you store hierarchical data more than relational? Mongo might be a better choice with a schema-tracking system on top of it.

Saying that Mongo is the choice just for scaling is silly.

That said, Postgres is my go-to database as well, almost always, and I've never much needed anything different. Yet. Waiting to find something that it doesn't fit.

Not only are people trashing Mongo, but this article is a ten minute walk through the barest of API surface for a library... There doesn't seem to be anything too novel in the API itself, and the pymongo project is a few years old...

> what are the things that Mongo excels at these days that are missing from pg?

I wouldn't say Mongo excels at these, but basic distributed stuff, like failover, sharding, rebalancing, resyncing, operations in general, etc. PostgreSQL can never get to MongoDB's level there, not even theoretically.

For me, the beauty of Mongo is when it’s paired with C#. There is no object relational mismatch and the Mongo Linq driver is much less obtuse than Entity Framework.

When you work with Mongo in C#, you’re dealing with strongly type collections where the compiler enforces consistency in your schema and your strongly typed linq queries are translated to MongoQuery.

I would not use a schemaless data store with a weakly typed language.

Why on Earth are people downvoting you? This article is absolutely blogspam. It's a paraphrasing of the PyMongo documentation with absolutely nothing else added, not even any editorializing.