Joins let you connect different views so that you can explore data from more than one view at the same time and see how different parts of your data relate to each other.

For example, your database might include the tables order_items, orders, and users. You can use joins to explore data from all tables at the same time. This page explains joins in LookML, including specific join parameters and joining patterns.

Joins start with an Explore

Joins are defined in the model file to establish the relationship between an Explore and a view. Joins connect one or more views to a single Explore, either directly, or through another joined view.

Let’s consider two database tables: order_items and orders. After you generate views for both tables, declare one or more of them under the explore parameter in the model file:

When you run a query from the order_items Explore, order_items appears in the FROM clause of the generated SQL:

You can join additional information to our order_items Explore. For example, to add data about the order that the order_item is a part of, you can do something like this:

The LookML above accomplishes two things. First, you can see fields from both orders and order_items in the UI:

Second, the LookML describes how to join orders and order_items together. That LookML would translate to the following SQL:

These LookML parameters are described in greater detail in the following sections. See the join parameter reference page to learn more about how this LookML is translated into SQL.

Chat Team Tip: Users ask most about the validation error, “Unknown or inaccessible field,” which can be caused by a missing join. See the Help Center article about this error for more information.

Join parameters

Four main parameters are used to join: joins, join, type, relationship, and sql_on.

Step 1: Starting the Explore

First, create the order_items Explore:

Step 2: join

To join a table, you must first declare it in a view. In this example, orders is an existing view in our model.

Then, use the join parameter to declare that you want to join the orders view to order_items:

Step 3: type

Consider which type of join to perform. Looker supports LEFT JOIN, INNER JOIN, FULL OUTER JOIN, and CROSS JOIN. These correspond to the type parameter values of left_outer, inner, full_outer, and cross.

The default value of type is left_outer, and generally the popular join type.

Step 4: relationship

Define a join relationship between order_items and orders. Properly declaring the relationship of a join is important for Looker to calculate accurate measures. The relationship is defined from the order_items Explore to the orders view. The possible options are one_to_one, many_to_one, one_to_many, and many_to_many.

In this example, there can be many order_items for a single order. The relationship from order_items to orders is many_to_one:

If you do not include a relationship in your join, Looker defaults to many_to_one.

Step 5: sql_on

Declare how to join these two tables together with either the sql_on or the foreign_key parameter. We usually suggest sql_on since it can do everything foreign_key can do, but is typically easier to understand.

sql_on is equivalent to the ON clause in the generated SQL for a query. With this parameter, we can declare which fields should be matched up to perform the join:

You can also write more complex joins. For example, you may want to join only orders with id greater than 1000:

Check out substitution operators to learn more about the ${ ... } syntax in these examples.

Step 6: Testing

Test that this join is functioning as expected by going to the Order Items Explore. You should see fields from both order_items and orders.

See Model Development to learn more about testing LookML changes.

Joining through another view

You can join a view to an Explore through another view. In the example above, you joined orders to order_items via the order_id field. We might also want to join the data from a view called users to the order_items Explore, even though they don’t share a common field. This can be done by joining through the orders view.

Use sql_on or foreign_key to join users to orders instead of order_items. Do this by correctly scoping the field from orders as orders.user_id.

Here is an example using sql_on:

Joining a view more than once

A users view contains data for both buyers and sellers. To join data from this view into order_items, but do so separately for buyers and sellers, you can join users twice, with different names, using the from parameter.

The from parameter lets you specify which view to use in a join, while giving the join a unique name. For example:

In this case, only buyer data is joined as buyers, while only seller data is joined in as sellers.

Note: The users view must now be referred to by its aliased names buyers and sellers in the join.

Limiting fields from a join

The fields parameter lets you specify which fields are brought from a join into an Explore. By default, all fields from a view are brought in when joined. However, you might want to bring through only a subset of fields.

For example, when orders is joined to order_items, you may want to bring only the shipping and tax fields through the join:

You can also reference a set of fields, such as [set_a*]. Each set is defined within a view using the set parameter. Suppose you have the following set defined in the orders view:

You can choose to bring only these three fields through when you join orders to order_items:

Symmetric aggregates

Looker uses a feature called “symmetric aggregates” to calculate aggregations (like sums and averages) correctly, even when joins result in a fanout. Symmetric aggregates are described in more detail in A simple explanation of symmetric aggregates Help Center article, and the fanout problem they solve is explained in The problem of SQL fanouts article.

Primary keys required

To have measures (aggregations) come through joins, you must define primary keys in all views that are involved in the join.

Do this by adding the primary_key parameter to the primary key field definition in each view:

To correctly handle joined measures, Looker relies on you specifying a primary key where the values are completely unique, non-NULL values. If your data does not contain a primary key, consider whether the concatenation of several fields would result in a primary key of completely unique, non-NULL values. If your primary key is not unique or contains NULL values and your query includes data that reveal those issues, then Looker returns an error as described in this Help Center article.

Supported SQL dialects

For Looker to support symmetric aggregates in your Looker project, your database dialect must also support them. The following table shows which dialects support symmetric aggregates in Looker 22.10:

If your dialect does not support symmetric aggregates, be careful when executing joins in Looker, as some types of joins can result in inaccurate aggregations (like sums and averages). This problem and the workarounds for it are described in great detail in The problem of SQL fanouts Help Center article.

Learn more about joins

To learn more about join parameters in LookML, see the Join reference documentation.