Schemas vs. Schemaless structures and The PostgreSQL Type Farm   

There has been a lot of talk lately about schemaless models touted by NoSQL groups and how PostgreSQL fits into this New world order. Is PostgreSQL Object-Relational? Is it Multi-Model. We tend to think of PostgreSQL as type liberal and it's liberalness gets more liberal with each new release. PostgreSQL is fundamentally relational, but has little bias about what data types define each column of related tables. One of PostgreSQL great strengths is the ease with which different types can coexist in the same table and the flexible index plumbing and plan optimizer it provides that allows each type, regardless of how wild, to take full advantage of various index strategies and custom index bindings. Our 3 favorite custom non-built-in types we use in our workflow are PostGIS (of course), LTree (Hierarchical Type), and HStore (Key-Value type). In some cases, we may use all 3 in the same database and sometimes the same table - where we use PostGIS for spatial location, LTree for logical location, and Hstore just to keep track of random facts about an object that are easier to access than having a separate related table and are too random to warrant devoting a separate column for each. Sometimes we are guilty of using xml as well when we haven't figured out what schema model best fits a piece of data and hstore is too flat of a type to work. The advent of JSON in PostgreSQL 9.2 does provide for a nested schema-less model similar to what the XML type offers, but more JavaScript friendly. I personally see JSON as more of a useful transport type than one I'd build my business around or a type you'd use when you haven't figured out what if any structure is most suitable for your data. When you have no clue what structure a piece of data should be stored, you should let the data tell you what structure it wants to be stored in and only then will you discover by storing it in a somewhat liberal fashion how best to retrofit in a more structural self-descriptive manner. Schemas are great because they are self-describing, but they are not great when your data does not want to sit in a self-described bucket. You may find in the end that some data is just wild and refuses to stay between the lines and then by all means stuff it in xml or json or create a whole new type suit it feels comfortable in.