Lag over

Overview

The "lag over (order by )" query pattern accesses the field value of the previous row as determined by some ordering.

For "lag over (order by)" queries whose ordering can be represented by some equality condition (such as when ordering by a field that increases at a regular interval), Materialize provides an idiomatic SQL as an alternative to the window function.

{{< callout >}}

Materialize and window functions

{{< idiomatic-sql/materialize-window-functions >}}

{{</ callout >}}

Idiomatic Materialize SQL

{{< important >}}

Do not use if the "lag over (order by)" ordering cannot be represented by an equality match.

{{</ important >}}

Exclude the first row in results

Idiomatic Materialize SQL: To access the lag (previous row's field value) ordered by some field that increases in a regular pattern, use a self join that specifies an equality condition on the order by field (e.g., WHERE t1.order_field = t2.order_field + 1, WHERE t1.order_field = t2.order_field * 2, etc.). The query excludes the first row since it does not have a previous row.

Use a self join that specifies an equality match on the lag's order by field (e.g., fieldA). The order by field must increment in a regular pattern in order to be represented by an equality condition (e.g., WHERE t1.fieldA = t2.fieldA + ...). The query excludes the first row in the results since it does not have a previous row.

{{< important >}}

The idiomatic Materialize SQL applies only to those "lag over" queries whose ordering can be represented by some equality condition.

{{</ important >}}

-- Excludes the first row in the results --
SELECT t1.fieldA, t2.fieldB as previous_row_value
FROM tableA t1, tableA t2
WHERE t1.fieldA = t2.fieldA + ... -- or some other operand
ORDER BY fieldA;

Avoid the use of LAG(fieldZ) OVER (ORDER BY ...) window function when the order by field increases in a regular pattern.

-- Anti-pattern. Avoid. --
SELECT fieldA, ...
    LAG(fieldZ) OVER (ORDER BY fieldA) as previous_row_value
FROM tableA;

Include the first row in results

Idiomatic Materialize SQL: To access the lag (previous row's field value) ordered by some field that increases in a regular pattern, use a self LEFT JOIN/LEFT OUTER JOIN that specifies an equality condition on the order by field (e.g., ON t1.order_field = t2.order_field + 1, ON t1.order_field = t2.order_field * 2, etc.). The LEFT JOIN/LEFT OUTER JOIN query includes the first row, returning null as its lag value.

Use a self LEFT JOIN/LEFT OUTER JOIN (e.g., FROM tableA t1 LEFT JOIN tableA t2) that specifies an equality match on the lag's order by field (e.g., fieldA). The order by field must increment in a regular pattern in order to be represented by an equality condition (e.g., ON t1.fieldA = t2.fieldA + ...). The query includes the first row, returning null as its lag value.

{{< important >}}

The idiomatic Materialize SQL applies only to those "lag over" queries whose ordering can be represented by some equality condition.

{{</ important >}}

-- Includes the first row in the results --
SELECT t1.fieldA, t2.fieldB as previous_row_value
FROM tableA t1
LEFT JOIN tableA t2
ON t1.fieldA = t2.fieldA + ... -- or some other operand
ORDER BY fieldA;

Avoid the use of LAG(fieldZ) OVER (ORDER BY ...) window function when the order by field increases in a regular pattern.

SELECT fieldA, ...
    LAG(fieldZ) OVER (ORDER BY fieldA) as previous_row_value
FROM tableA;

Examples

{{< note >}}

The example data can be found in the Appendix.

{{</ note >}}

Find previous row's value (exclude the first row in results)

Using idiomatic Materialize SQL, the following example finds the previous day's order total. That is, the example uses a self join on orders_daily_totals. The row ordering on the order_date field is represented by an equality condition using an interval of 1 DAY. The query excludes the first row in the results since the first row does not have a previous row.

-- Excludes the first row in results --
SELECT o1.order_date, o1.daily_total,
    o2.daily_total as previous_daily_total
FROM orders_daily_totals o1, orders_daily_totals o2
WHERE o1.order_date = o2.order_date + INTERVAL '1' DAY
ORDER BY order_date;

{{< important >}}

The idiomatic Materialize SQL applies only to those "lag over" queries whose ordering can be represented by some equality condition.

{{</ important >}}

<red>Avoid the use of LAG() OVER (ORDER BY ...) window function to access previous row's value if the order by field increases in a regular pattern.</red>

-- Anti-pattern. Includes the first row's value. --
SELECT order_date, daily_total,
    LAG(daily_total) OVER (ORDER BY order_date) as previous_daily_total
FROM orders_daily_totals;

Find previous row's value (include the first row in results)

Using idiomatic Materialize SQL, the following example finds the previous day's order total. The example uses a self LEFT JOIN/LEFT OUTER JOIN on orders_daily_totals. The row ordering on the order_date field is represented by an equality condition using an interval of 1 DAY. The query includes the first row in the results, using null as the previous value.

-- Include the first row in results --
SELECT o1.order_date, o1.daily_total,
    o2.daily_total as previous_daily_total
FROM orders_daily_totals o1
LEFT JOIN orders_daily_totals o2
ON o1.order_date = o2.order_date + INTERVAL '1' DAY
ORDER BY order_date;

{{< important >}}

The idiomatic Materialize SQL applies only to those "lag over" queries whose ordering can be represented by some equality condition.

{{</ important >}}

<red>Avoid the use of LAG() OVER (ORDER BY ...) window function to access previous row's value if the order by field increases in a regular pattern.</red>

-- Anti-pattern. Includes the first row's value. --
SELECT order_date, daily_total,
    LAG(daily_total) OVER (ORDER BY order_date) as previous_daily_total
FROM orders_daily_totals;

See also

• Lead over
• INTERVAL
• LEFT JOIN/LEFT OUTER JOIN
• LAG()
• Window functions