Full text search in Odoo Ecommerce Website

Helping customers find the right products quickly is crucial to running a successful ecommerce website. Therefore, product searching deserves attention to its performance. Odoo, a batteries included solution to managing enterpries, comes with its own off-the-shelf search functionality. It works out of the box with reasonable speed for small inventory. For larger product collection, it leaves something to be desired. In this post, we are looking at how Odoo is implementing its search function and possible methods to speed it up.

1. How odoo is doing search currently

When an user enters a query into Odoo’s ecommerce search box, Odoo is using Sql’s ilike operator to find relevant products. ilike does case-insensitive substring matching. Products that contain the search query in theirs name or description are returned. To increase search recall, i.e returns as many related results as possible, Odoo tokenizes a search query into words and do ilike match for each word separately. That way, positions of matched words in resulted products’ names can be flexible. A query of “Apple keyboard” will be able to match with a product named “Apple wireless keyboard”.

In order to find the result, Postgres would need to run pretty exhaustive search. It would look through every product in store, and for each loop through name and description fields to find matched substrings. Needless to say, this approach will be more costly as database grows.

Let say an user searches for office lamp, the equivalent sql query is roughly as below. (for ease of reasoning, this version of the sql query is simplified)

select id from product_template where name ilike '%office%' and name ilike '%lamp%'

Running the same query through Postgres query planner, surprisingly, a sequential scan is done on product_template table, even though there exists an name_index on name column.

Sequential scan is required to search for products

Turns out, the current index is of BTree type, which is only made use of if the search operator like and query is matched at the beginning of the text. This is understandable, given how BTree is implemented. A record would be added to BTree index by comparing it lexicographically with existing nodes. Lexicographic order starts from the beginning. A little modification to the above query would allow Postgres to use the existing index:

select id from product_template where name like 'office%' and name like 'lamp%'

BTree index is used to search for products

Though using BTree index significantly improves search time, only matching queries at the beginning of a product’s name is not helpful. Let see if we can do better.

2. Speed up using index

Though BTree does not work for our need, PostgreSQL comes with other types of indices out of the box. Among them is GIN (generalized inverted index), which, as its name implies, provides inverted index for full-text search. By default, GIN tokenizes word by whitespaces, thus allows only queries that match exact words. With the help of pg_trgm trigram extension, we can solve the problem of partial text match. This blog post is pretty helpful in explaining Postgres GIN index.

create extension if not exists pg_trgm;
create index name_search_index on product_template using gin (name gin_trgm_ops)

If you don’t want to get your hands dirty, we have already pre-packaged a full-text search module for product search in Odoo. Feel free to try it and let us know if you have any suggestions.