OSCON 2008, Session 1: CouchDB

RDBMS

• CouchDB is not a relational database.
• Usually you design a schema up front

What is CouchDB

• Stores Documents (individual data records)
• No Schema!
• Columns containing NULL don’t make sense
  • "My Business card doesn’t have ‘Fax Number:’ and then NULL"
• Natural data behavior

Documents

• Store your document data in a JSON string.
• Talked about how Ruby is basically JSON compatible with no translations
• XML Sucks.

Short example:

  {
    "_id":"223BDCD",
    "_rev":"834BC",
    "age":54,
    "name":"Darth Vader",
  ...
  }

• Revision allows you to fetch the document, write a new copy, and save it as the latest revision of the document. This allows you to turn back time per database row!

How do I talk to it?

HTTP REST API

• Create: HTTP PUT /db/docid
• Read: HTTP GET /db/docid
• Update: HTTP POST /db/docid
• Delete: HTTP DELE /db/docid
• The ID does not have to be generated by the user. Just don’t provide one. If you provide one, it has to be a string.
• JSON doesn’t deal with binary data, you have to BASE64 encoding. There apparently is some other way to handle it.
• If you don’t have well-formed JSON, all calls will result in an error. They don’t have a way to specify a way to enforce writes.
• Type integrity checking: They don’t care.
• Is there a way to get a document using something other than an id? Yes.

There are 2 more features that make Couch really cool:

Views

• Filter, Collate, Aggregate
• Powered by map/reduce! (They improved it a little bit!)
• Views are built incrementally and on demand. Reduction is optional.
• Sends diffs around to sync db data. VERY FAST!
• No write penalty with views.
• The view is simply the result of a map/reduce function stored in a btree.

Example: Tag Cloud

• We have a db full of tagged documents
• We must know how often each tag appears
• Use map/reduce!
• Works well since it’s in Erlang, which can be massively parallel

Replication

• CouchDB was originally designed for an offline replication of your database.
• Replication works a lot like rsync
• They don’t use auto_increment
• Full new revisions of documents, not partial changes.

Built for the future

• Written in Erlang.
• Non-locking MVCC and ACID compliant data store
  • No locking of the data store ever
• Damien Katz invented it. Self-funded fulltime development for 2 years.
• Now it’s backed by IBM.