Depends on what you're doing, I suppose. CouchDB's multimaster concurrency features are still effectively best in class, but there are so many managed DB services now, that it doesn't seem like a headline feature.

The RESTful interface is pretty cool. And Couch's in-built security features are near enough a best kept secret.

CouchDB always lost to Mongo in discussions about speed unfortunately. Data safety and security notwithstanding. You can speed it up if you're willing to use Erlang as the primary query language, which wasn't something many were keen to do.

Another big hamstring is data modelling in a document db...

I'm not a data scientist, but anytime I hear complaints about Mongo, and have access to the codebase that's the subject of the complaint, I invariably find whopping great multi-property entities confined to a single document, without anything close to an attempt to normalise the data (or worse, a concerted intellectual effort to stuff everything into a single document...).

There's more I can say...

I think CouchDB is woefully underrated.

It's on version 3 now...

Another CouchDB fan here. CouchDb's mango queries are implemented as Erlang map functions under the hood. Users now can get the speed of Erlang and the usability of Mongo-like query syntax.

Thanks for the CouchDB feedback guys! I am curious to try it out. I should have been more clear, I was actually referring to Couchbase (https://www.couchbase.com/) not CouchDB :) Very unfortunate naming conflict considering they are both JSON document stores. Is there something about the word 'couch' that I'm missing?

From my experience with Couchbase though, I would recommend it to others. They support SDKs in many languages, have decent documentation, and the server has performed well for us. The query language 'N1QL', tries to emulate SQL syntax, and I found it nice to work with.

Couchbase was a commercial CouchDB spinoff product, co-created by the programmer who created CouchDB. Much of the API is (was?) the same, but where possible, Erlang was exchanged for C++.

It also has a built-in, in-memory cache layer, which CouchDB doesn't have.

Oh wow, thanks. I didn’t know the history of it!

I used CouchDB for a project back in 2012 when all of these things were fairly new. The reason was to embed the DB on an Android device. This turned out to not work as well as advertised. I have no idea what's happened with Couch since, but started got pretty familiar with Mongo in the last 12 months. It's OK, but it's expensive to make it performant. As other commenters here have mentioned, the primary use case for these Object/Document storage solutions seems to be to remove knowledge of Relational DB as a requirement for developers.

I think it's probably excellent for prototyping, but I wouldn't want it running a heavily used product in production.