His points are fair, but it should be titled What CouchDB Sucks at Doing, because it is far from sucking, and he concedes that.

I'll agree with that :) I did chose an overly controversial title. My apologies.

Well your first line is CouchDB really sucks at doing some things, so it's not so bad :)

Where you (found that) CouchDB did fit, how did you find it?

For everything that I've used it for, it's been fantastic.

I was part of a team that built a social web crawler to look for XFN links. It worked fantastically for that.

I used it to create the tumblog on my website, which is pretty much an ideal use case for it. The source code for that is available so you can check it out here: http://github.com/ericflo/django-couch-lifestream/tree/maste...

At my day job I've found Lucene to outperform CouchDB in query performance by orders of magnitude. I deal with a huge volume of manufacturing data, some of which fits into a "document model". The Solr frontend to Lucene is a poor substitute for CouchDB's web APIs though, too much configuration required -- it was easier to roll my own REST-ish front end with JSPs.

Well you can't bone a chicken drumstick very easily with a bread knife, so why would one try to use CouchDB as a relational database. I really fail to see the point of this article, particularly given the title

Unfortunately, I think too many people are using Fad-Driven Development where you use the newest and coolest tools even if they're totally inappropriate. The parade of articles about how Relational Databases Are Dead (even if your data is relational?) doesn't help, so I see this post as a nice counterpoint. The title is indeed trollicious.

Fad-Driven Development, I love it :) Sounds like a new O'Reilly book in progress.

Well certainly the changing landscape of the Web has caused many to revisit the use of relational databases. There's considerable overlap with what app servers do, schemas are static beasts in general and often create application lock-in, bla, bla.. There is a fundamental tension in O-R mapping that I've concluded is best solved by dropping the R and using alternative persistence models.

That said, relational databases are phenomenal and aren't going away anytime soon.

I think CouchDB hits a sweet spot in a number of areas and is worth a very close look. It's still early and needs a lot of work but I think it's a keeper, especially given it's use of Erlang/OTP. Erlang is very mature and though it wasn't built with multi-cores in mind it's concurrency model is one of the few games around.

Your point is valid, and I wish I could give you an extra vote for coining(?) the word "Fad-driven development".

This should be called "Where CouchDB Sucks" he highlights some points but also concedes it's great in many places. Frankly a lot of his criticisms are openly pointed out by the CouchDB docs.

When do you use Couchdb then?