reloading tire/elasticsearch mappings for a model that already has data stored

Question

I am using Tire and elasticsearch to provide search functionality on a MongoMapper model, which is part of a Rails App. I just stumbled across a problem where the mappings for this model were not being updated when I redeployed to an environment that uses the following configuration (in config/environments/env_name.rb):

config.cache_classes = true

reloading the class alone didn't seem to fix the issue (perhaps understandably, the new mappings might not be incompatible with existing data I guess?). instead I had to do the following:

MyModel.index.delete
<restart the app or reload the class>
MyModel.index.import MyModel.all

I just wondered if there's a better way of a). ensuring the latest mappings defined in my model code are being used by elasticsearch after each deployment but b). avoiding unnecessary repopulating the index with the complete dataset?

We normally deploy using Chef, so I could automate the three steps I used successfully without too much trouble. But I'm new to elasticsearch and tire so I thought it's highly likely I'm misusing both or making things unnecessarily difficult.

Answer 1

Couple of points here:

• Tire tries to create the index with correct mapping when the class loads
• but Tire does not attempt to create the index for the model when it already exists

So, your question is really more about the proper workflow? When you deploy a new version of the application, you shouldn't re-populate the index, in the same way you don't re-populate the database from some kind of backup.

Automatically checking for index mappings conforming to current definition in the model is certainly possible (compare the MyModel.tire.index.mapping with MyModel.tire.mapping, re-populate if different, etc), it's something I'd be wary to do.

The developer usually knows when she changed the mapping and should re-index the data. Dropping the index, and re-populating also means search downtime, and isn't even feasible for large applications.

A nicer solution is to use a specific index name such as my-index-2012-12 when importing the data, and point a my-index alias to this index. Then you can freely re-populate the index, and flip the alias when you're done, without downtime. Tire tries hard to support you in this kind of workflow (the Rake import task, etc).