sql_distinct_key is usually used when there is no primary key on the table you are aggregating over. In most cases, you will not need sql_distinct_key if you have a primary key.

Measures of type sum_distinct and type average_distinct can optionally take a new parameter: sql_distinct_key.

Normally, sum_distinct and average_distinct measures use the declared primary key of the view they are contained in as unique value associated with the number on which they are performing the operation. This parameter lets sum_distinct and average_distinct work in situations where there is no primary key or the view is denormalized (for example, a very wide table with subtables embedded).

Example

Suppose you have a table that looks like the following:

| order_time | order_id | order_shipping_amount | customer_id | item_id | item_name | item_amount |

In this particular table, there is no primary key and each row is repeated for every item in an order.

A single order may consist of multiple rows in this table.

Summing order_shipping_amount would be incorrect, because it would be repeated for each item.

You could use LookML's sum_distinct to perform this instead:

measure: total_shipping_amount {
  type: sum_distinct
  sql: ${TABLE}.order_shipping_amount ;;
  sql_distinct_key: ${TABLE}.order_id ;;
}