feat: semi-additive measures

[betodealmeida]

Summary

This PR adds metric reaggregation support for semi-additive measures, aligned with the proposal #2245. Metrics can declare protected dimensions, and the aggregation behavior changes when the query grain drops the dimension.

For example, suppose we have this metric:

name: v3.daily_balance
type: metric
query: SELECT SUM(line_total) FROM v3.order_details
reaggregate:
  rules:
    - dimension: v3.date.date_id[order]
      fn: last_value

Meaning:

daily_balance is additive within a day, but not additive across order date. If date disappears from the query grain, collapse with the last date’s value.

If we get a request that includes the protected dimension:

metrics: [v3.daily_balance]
dimensions: [v3.date.date_id[order]]

We generate this SQL:

WITH order_details_0 AS (
  SELECT
    order_date AS date_id_order,
    SUM(line_total) AS line_total_sum
  FROM v3.order_details
  GROUP BY order_date
)
SELECT
  date_id_order,
  SUM(line_total_sum) AS daily_balance
FROM order_details_0
GROUP BY date_id_order

Because date_id_order is still in the output grain, there is no semi-additive collapse. Normal aggregation is safe.

On the other hand, if we don't request the protected dimension we shouldn't aggregate over it:

metrics: [v3.daily_balance]
dimensions: [v3.product.category]

The generated SQL:

WITH order_details_0 AS (
  SELECT
    product_category AS category,
    order_date AS date_id_order,
    SUM(line_total) AS line_total_sum
  FROM v3.order_details
  GROUP BY product_category, order_date
)
SELECT
  category,
  MAX_BY(line_total_sum, date_id_order) AS daily_balance
FROM order_details_0
GROUP BY category

Here, the user asked for category only, but DJ keeps date_id_order as a private inner grain. Then it collapses each category’s daily balances with MAX_BY(value, date), meaning "take the value from the latest date."

Without this, DJ would generate something like:

SELECT
  product_category AS category,
  SUM(line_total) AS daily_balance
FROM v3.order_details
GROUP BY product_category

That would incorrectly sum balances across dates, which is wrong for snapshots or balances, for example.

I've updated the UI to show and allow setting the semi-additive dimension:

When editing:

Valid options:

Test Plan

• PR has an associated issue: #
• make check passes
• make test shows 100% unit test coverage

Deployment Plan

[netlify]

✅ Deploy Preview for thriving-cassata-78ae72 canceled.

Name	Link
🔨 Latest commit	49d5866
🔍 Latest deploy log	https://app.netlify.com/projects/thriving-cassata-78ae72/deploys/6ab3fdd77932e200088b84f8

[shangyian]

LGTM!