And it's getting a little old.
No solution works for everything all the time. Can we just stipulate that?
I agree that relational databases are often applied to problems where other systems would be better, and we should work on studying those other systems, improving their implementations, and educating engineers about their use. But we should be "pro DHT" or "pro document store" or "pro column store", not "anti SQL."
You gotta kinda wrap your mind around SQL though and if you don't have a support group helping you do that, you're probably going to struggle with it. So I can understand the desire to find a simpler alternative to solve information based systems, something between file based Perl and SQL.
I also remember when I first started with sql that I could read and understand (most of) it quite easily. Writing it was of course another story, especially if optimization mattered, but it wasn't hard to read and not too hard to learn at least the basics. At this point the market is filled with books like "SQL for Dummies" and "The Manga Guide to Databases" that provide very gentle introductions for complete beginners as well as numerous classes at practical trade schools.
Mastering SQL and RDBMS is definitely a work for years and creates dedicated specialists just like any deep field, but picking up the basics is relatively easy compared to many other tasks in this field.
Map/reduce, though, that will do your head in! It's like changing from procedural to functional programming. To me, it's much harder and far less intuitive. So I can't imagine how anyone could think using M/R is somehow easier than learning SQL. Anyone who thinks that is in for a nasty surprise. It's like claiming Erlang is easier to learn than Ruby.
The problem you are going to run into, I know a company they wanted a particular database, so they got a designer turned programmer who built a system for them in Perl using flat files. His architecture was to put one data field per line in a file and then put the one-to-many records, comments about the master record in the file starting at line x. Well, now they have fields for the master item all the way to line x minus 2 and they know they want like 5 more fields.
If the developer who built the system had started on an SQL database, then the solution to the problem of needing more fields would be really simple.
Now this problem is just one issue. Add to this the problem of needing to query the master records on particular fields. The programmer has to open up every single file and go to the particular line where the field exists and see what the value is. Of course as the file based database grows, it's really starting to impact performance.
The big problem here is that the original programmer didn't know SQL or how to manage a SQL database so he downloaded some Perl code (note, I absolutely LOVE perl, so don't be hating) that managed a dataset in perl and just went to town.
This is not the way to build a robust information based system.
SQL databases are a tool and files are a tool and CouchDB is a tool and a lot of people are going to read this biased stuff about noSQL and think, "hey, let's use couchdb, et al" and they only have 5 users on the system and, you know, maybe a million records and Access, MySQL, SQL Server Express, or any number of other free sql based systems would handle a problem like that just fine.
Instead, they're going to go write a ton of code to fit a key/value store to the problem and it's going to be a nightmare to support as the system grows and they need to implement security and manage more fields.
But the pro-noSQL people don't talk about all those issues that are going to arise later and focus solely on the scalability issue when in the vast majority of cases scalability is really the last thing you need to think about.