A while ago I remember that I had to explicitly create my collections and define a shard key in order to use CosmosDB with MongoDB API with database-level throughput.
I see today that this does not seem necessary anymore. I turn on database-level throughput and run my code and unit tests without defining a shard key or anything and it all, so far, "just works". Searching for documentation on this I can't find anything. Are they just defining a shard key for me under the hood?
Not having to explicitly create collections with a shard key and not having to painstakingly modify all my queries to include a shard field that I don't really even need (there's nothing of any great size or need to scale in my scenario) would be wonderful. I always thought it was a big bummer to be forced to do that JUST to change the cost model on the database. Adding to that needing to change functionality (you can't create a unique index with a hashed shard key, for example). Ugh.
Yes, we made this change when we launched 3.6 support. Agree it's not really documented but we mention it in our announcement blog post here.
Thanks.