Historically, Thinking Sphinx has supported SQL-backed indices. However, in recent releases (since 3.0.4), real-time indices are supported as well. Real-time indices can be updated on a per-record basis, which removes the need for delta indices. The trade-off is that they’re usually a bit slower to populate (as each record must be iterated through, instead of bulk inserts via SQL queries).
The documentation here is now pretty good at distinguishing any different behaviours between the two indexing approaches, and there is a blog post from when the feature was first released which covers the basics quite well.
However, if you have used SQL-backed indices before and want to change, this page is the best place to understand what’s different.
If you’re changing from SQL-backed indices to real-time indices, make sure you stop Sphinx, and then delete your index and binlog files (which are stored by default in
tmp/binlog respectively). Otherwise, Sphinx gets confused between the old (SQL-backed) index data and the new (real-time) index definitions.
With real-time indices, you should never use
ts:rebuild. Their functional equivalents are
ts:regenerate (as covered on the rake tasks page). You only need to run
ts:regenerate when you’ve added an index, removed an index, or edited the structure of an index (e.g. added/removed/modified fields or attributes).
Indexes, Fields & Attributes
Real-time index definitions operate in a slightly different context: your model, rather than your model’s database table. The implications of this are as follows:
Methods and values
Field and attributes refer to methods rather than columns and associations. You can chain method calls, but each intermediate method must refer to a single object, and the final item in the chain must return the data in the right format. Thus, fields must return strings, attributes must return objects of their specified type, or arrays of such items, if they’re a multi-value array.
Because the database isn’t available as a reference, and Ruby is dynamically typed, attributes must have their type explicitly set, and if they’re a multi-value attribute, that is required as well.
If you were using SQL snippets to modify column data, or are aggregating values in some way, the real-time approach is to instead define a method in your model that does the same:
Because Thinking Sphinx is loading your records via ActiveRecord, you can define a custom scope to use - although this is only in play for bulk inserts (via
This allows eager loading of associations, or even filtering out specific values. However, keep in mind the default callbacks don’t use this scope, so a record that does not get included in this scope but is then altered will be added to your Sphinx data.
To ensure changes to your model instances are reflected in Sphinx, you’ll need to add a callback:
If you want changes to associated data to fire Sphinx updates for a related model, you can specify a method chain for the callback.
The rest of the callbacks documentation covers more advanced usage.
Because real-time indices are updated through your Ruby app, rather than by interacting with the database, you can use transactional fixtures in your tests. However, you’ll probably want to disable Sphinx and the callbacks in unit tests. Here’s an example for how I set things up with RSpec (where callbacks and Sphinx are only enabled for request specs):