Saturday, 17 March 2012

TDD - All the time

Test Driven Development (TDD) is one of the best practices that extreme programming (Xp) suggests. As defined by Wikipedia, TDD is a software development process, in which a developer writes a failing test case, that defines a new improvement, or new functionality, then produce the code to pass that test and finally refactors it to acceptable standards.

Having said that, the series of questions I hear from people who are introduced to TDD are, "Is it really important to write tests in first place? I know the functionality. Why can't I just sit down and start coding. What do I gain by writing tests first? The whole idea of TDD is completely counter intuitive. Of course I understand the importance of tests. But I can add them later." Very valid questions. There are two simple reasons why I think TDD helps in development.

First thing - TDD is a very intuitive process. Whenever we write a piece of code everyone of us know, what will be the output of the code given an input. We keep simulating that in our brain. Nobody starts coding without a problem statement in mind. Everybody knows what is expected out of the code at every point I'm time. And we validate it with sample input every time. This, exactly, is the idea behind TDD. Set the expectation for the code you write. Implement it. Validate it. Isn't TDD a intuitive process?

I believe in the philosophy of 'Self Documenting Code', code that explains itself without the need for extra documents. The testcases are a very useful add-on to this philosophy. In addition to validating the code, it also captures the intent behind writing a particular code, which comes handy later in the life-cycle of a project, either while debugging or while revisiting a functionality.

Friday, 16 March 2012

Missing data during Tire indexing

Am working on a RoR project, which uses Postgres for data storage and ElasticSearch (ES) for providing the search functionality. We were using Tire to talk to the ES server. Ran into an issue recently, where in the search results didn't return a particular document.  For the purpose of this post lets consider a model like the one defined below.

class Document < ActiveRecord::Base
# t.integer id
# etc., etc.,

The particular record, that i was looking for, was available in the DB, but was not available in the ES index. So, the first and foremost thing that i did is to re-build the index. Tire provides you with a rake task to import the indices. So, I can do

rake tire:import CLASS=Document FORCE=true

to rebuild my indices. Did that and tire reported that it has indexed all the records from the database to ES. But the document that i was looking for is still not returned in the search results.

Suspected ES to be the culprit and
1) Checked the number of documents indexed and found that the number of records, that tire reported to have indexed, and the actual number of documents indexed did not match.
2) Grep-ed the ES data folder for the unique identifier of the missing document and it was nowhere to be found in those folders.
Read some more in the Tire documentation and found that every record in the database can be indexed individually. Enter rails console.

Document.find('The missing document').update_index

I was expecting this to fail and throw some exception to give me some clue in this debugging process. But to my surprise it was successful and the document was returned in the search results. Digging the tire documentation for some more time tells you that you can also do

Document.index.import Document.all

to index all the documents from DB to ES. Again, I was expecting this to give me the same results as the rake task, because ideally they are doing the same thing. I was in for a surprise again. The document still came up in the search results. This is when I started suspecting the rake task and looked up the tire source code to see what it actually does. Ignoring the code for progress bar and other things, only the three steps are of interest to me.

index =
index.create :mappings => klass.tire.mapping_to_hash, :settings => klass.tire.settings
index.import(klass, 'paginate' {}) do |documents|

So, I tried doing the same thing in the rails console, by just replacing the 'klass' with 'Document'. And the document went missing again. But this time it gave some visibility into what is happening internally. When we ask tire to rebuild the indices, it doesn't get all the documents in one go and build the index. It fetches the documents page by page (with a max of 1000 documents per page) by calling the 'paginate' method, and the paginated results had duplicate records across the pages.

We were using 'kaminari' in our project to take care of the pagination and hence we had to inject this method for the models that needs indexing and it looked like:

module Paginatable
module KlassMethods
def paginate(options)

def self.included(klass)
klass.extend KlassMethods

I just had to change it to include some default ordering, to make sure that tire gets all the records (without any duplicates) for indexing.

def paginate(options)