I’m doing a bit of research into what may or may not be an issue with a specific database in our BigCouch cluster, but regardless of the outcome of that side of things I thought I’d share how I used Python and couchdb-python to dig into the problem.

In our six-server BigCouch cluster we noticed that on the database for one of our most heavily trafficked applications the document counts displayed in Futon for each of the cluster members don’t match. As I said above this may or may not be a problem (I’m waiting on further information on that particular point), but I was curious which documents were missing from the cluster member that has the lowest document count. (The interesting thing is the missing documents aren’t truly inaccessible from the server with the lower document count, but we’ll get to that in a moment.)

BigCouch is based on Apache CouchDB but adds true clustering as well as some other very cool features, but for those of you not familiar with CouchDB, you communicate with CouchDB through a RESTful HTTP interface and all the data coming and going is JSON. The point here is it’s very simple to interact with CouchDB with any tool that talks HTTP.

Dealing with raw HTTP and JSON may not be difficult but isn’t terribly Pythonic either, which is where couchdb-python comes in. couchdb-python lets you interact with CouchDB via simple Python objects and handles the marshaling of data between JSON and native Python datatypes for you. It’s very slick, very fast, and makes using CouchDB from Python a joy.

In order to get to the bottom of my problem, I wanted to connect to two different BigCouch cluster members, get a list of all the document IDs in a specific database on each server, and then generate a list of the document IDs that don’t exist on the server with the lower total document count.

Here’s what I came up with:

>>> import couchdb
>>> couch1 = couchdb.Server(‘http://couch1:5984/’)
>>> couch2 = couchdb.Server(‘http://couch2:5984/’)
>>> db1 = couch1[‘dbname’]
>>> db2 = couch2[‘dbname’]
>>> ids1 = []
>>> ids2 = []
>>> for id in db1:
…     ids1.append(id)
… 
>>> for id in db2:
…     ids2.append(id)
… 
>>> missing_ids = list(set(ids1) – set(ids2))

What that gives me, thanks to the awesomeness of Python and its ability to subtract one set from another (note that you can also use the difference() method on the set object to achieve the same result), is a list of the document IDs that are in the first list that aren’t in the second list.

The interesting part came when I took one of the supposedly missing IDs and tried to pull up that document from the database in which it supposedly doesn’t exist:

>>> doc = db2[‘supposedly_missing_id_here’]

I was surprised to see that it returned the document just fine, meaning it must be getting it from another member of the cluster, but I’m still digging into what the expected behavior is on all of this. (It’s entirely possible I’m obsessing over consistent document counts when I don’t need to be.)

So what did I learn through all of this?

• The more I use Python the more I love it. Between little tasks like this and the fantastic experience I’m having working on our first full-blown Django project, I’m in geek heaven.
• couchdb-python is awesome, and I’m looking forward to using it on a real project soon.
• Even though we’ve been using CouchDB and BigCouch with great success for a couple of years now, I’m still learning what’s going on under the hood, which for me is a big part of the fun.