METRIC_VIEW_WINDOW_MEASURE_REFERENCES_WINDOW_MEASURE - fixing metric views after the June 2026 Databricks validation change

#What changed

From the Databricks Runtime maintenance updates (29 June 2026, 17.3 LTS, 18 LTS and 18.2):

• Metric views now reject window measures that reference other window measures.
• The error is METRIC_VIEW_WINDOW_MEASURE_REFERENCES_WINDOW_MEASURE.
• The check can be turned off with spark.databricks.sql.metricView.rejectWindowOnWindowMeasure.enabled = false.

Earlier in 2026 (18 May), Databricks also fixed window measures returning wrong results with GROUP BY, IN/BETWEEN filters and mixed predicates on the order column. Taken together: window measures on top of window measures were an area where results could be wrong, and the platform now stops you building them.

#Example of what's now rejected

version: 0.1
source: main.sales.orders
dimensions:
  - name: order_date
    expr: order_date
measures:
  - name: revenue
    expr: SUM(amount)
  - name: revenue_running_total          # window measure
    expr: MEASURE(revenue)
    window:
      - order: order_date
        range: cumulative
        semiadditive: last
  - name: revenue_running_total_7d_avg   # window measure on a window measure -> rejected
    expr: MEASURE(revenue_running_total)
    window:
      - order: order_date
        range: trailing 7 day
        semiadditive: last

(The YAML shape follows the metric view specification; field names can differ slightly between spec versions. Use the version your workspace documents.)

#1. Find affected views

Views fail when they're created or altered, and queries against existing views may fail on the newer runtimes. List metric views and check their definitions:

SELECT table_catalog, table_schema, table_name
FROM system.information_schema.tables
WHERE table_type = 'METRIC_VIEW';

For each, SHOW CREATE TABLE (or open it in Catalog Explorer) and look for window measures whose expr uses MEASURE() of another window measure.

#2. Rewrite

Make each window measure reference base (non-window) measures only. For the example above, compute the 7-day trailing average directly from the base measure:

  - name: revenue_7d_avg
    expr: MEASURE(revenue) / 7
    window:
      - order: order_date
        range: trailing 7 day
        semiadditive: last

If the business really needs "a window over a window" (for example, the average of a running total), do it in the query that reads the metric view, or in a downstream table where the logic is explicit and testable:

WITH daily AS (
  SELECT order_date, MEASURE(revenue_running_total) AS running_total
  FROM main.sales.orders_metrics
  GROUP BY order_date
)
SELECT order_date,
       AVG(running_total) OVER (ORDER BY order_date ROWS BETWEEN 6 PRECEDING AND CURRENT ROW) AS running_total_7d_avg
FROM daily;

#3. Check numbers against a reference

Because results of these measures could be wrong before the May fix, compare the rewritten measures with an independent SQL calculation for a few dates and filter combinations before publishing them to dashboards and Genie spaces.

#4. The opt-out flag

SET spark.databricks.sql.metricView.rejectWindowOnWindowMeasure.enabled = false;

Use it only to keep a critical dashboard running for a few days while you rewrite. It turns off a check that exists because these results weren't reliable.

#Checklist

• Metric views with window-on-window measures identified
• Window measures rewritten to reference base measures only
• Window-over-window logic moved to queries or downstream tables
• Results compared with independent SQL
• Opt-out flag removed (if used)

#Sources

• Databricks Runtime maintenance updates
• Databricks Runtime 18 LTS release notes