Window functions in Snowflake - Commands & Configuration

Introduction

Sometimes you want to analyze data by looking at rows related to the current row without grouping everything together. Window functions let you do this by calculating values across a set of rows related to the current one, like running totals or rankings.

When you want to calculate a running total of sales per customer without losing individual sale details.

When you need to rank employees by salary within each department.

When you want to find the difference between each row's value and the previous row's value in a time series.

When you want to calculate averages over a sliding window of rows, like the last 7 days of data.

When you want to add row numbers to your query results for pagination or ordering.

Commands

This command calculates a running total of sales amounts for each customer ordered by the date of the order. It uses a window function SUM() OVER() to keep the individual rows while adding the running total.

Terminal

SELECT customer_id, order_date, amount, SUM(amount) OVER (PARTITION BY customer_id ORDER BY order_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS running_total FROM sales ORDER BY customer_id, order_date;

Expected Output

CUSTOMER_ID | ORDER_DATE | AMOUNT | RUNNING_TOTAL

1           | 2024-01-01 | 100    | 100

1           | 2024-01-05 | 150    | 250

2           | 2024-01-02 | 200    | 200

2           | 2024-01-06 | 100    | 300

This command ranks employees by salary within each department. The RANK() window function assigns ranks without collapsing rows, so you see each employee's rank in their department.

Terminal

SELECT employee_id, department, salary, RANK() OVER (PARTITION BY department ORDER BY salary DESC) AS salary_rank FROM employees ORDER BY department, salary_rank;

Expected Output

EMPLOYEE_ID | DEPARTMENT | SALARY | SALARY_RANK

101         | Sales      | 90000  | 1

102         | Sales      | 85000  | 2

201         | HR         | 70000  | 1

202         | HR         | 65000  | 2

This command shows each row's value and the previous row's value ordered by date. The LAG() function looks back one row to compare values.

Terminal

SELECT date, value, LAG(value, 1) OVER (ORDER BY date) AS previous_value FROM metrics ORDER BY date;

Expected Output

DATE       | VALUE | PREVIOUS_VALUE

2024-01-01 | 10    | NULL

2024-01-02 | 15    | 10

2024-01-03 | 20    | 15

Key Concept

If you remember nothing else from this pattern, remember: window functions let you calculate values across related rows without grouping and losing individual row details.

Common Mistakes

Using GROUP BY instead of window functions to calculate running totals.

GROUP BY collapses rows and loses individual row details, so you can't see running totals per row.

Use SUM() OVER() with PARTITION BY and ORDER BY to keep rows and calculate running totals.

Not specifying ORDER BY inside the OVER() clause for functions like SUM() or RANK().

Without ORDER BY, the window function does not know the order of rows, so results like running totals or ranks will be incorrect or meaningless.

Always include ORDER BY inside OVER() when order matters for the calculation.

Confusing window functions with aggregate functions without OVER(), expecting row-level results.

Aggregate functions without OVER() collapse rows into one result, losing row-level data.

Add OVER() clause to aggregate functions to turn them into window functions that keep row-level data.

Summary

Use window functions with OVER() to calculate values across related rows without grouping.

Include PARTITION BY to group rows logically and ORDER BY to define row order inside the window.

Common window functions include SUM(), RANK(), and LAG() for running totals, rankings, and comparisons.

Practice

1. What does a window function in Snowflake do?

easy

A. Calculates values across rows related to the current row without grouping them into fewer rows

B. Groups rows and reduces the number of rows returned

C. Deletes duplicate rows from the result set

D. Creates a new table from existing data

Solution

1. Step 1: Understand window function purpose

   Window functions perform calculations across a set of rows related to the current row but do not reduce the number of rows returned.

2. Step 2: Compare with grouping

   Unlike GROUP BY, window functions keep all rows visible while calculating values like running totals or ranks.

3. Final Answer:

   Calculates values across rows related to the current row without grouping them into fewer rows -> Option A

4. Quick Check:

   Window functions analyze rows without grouping = A [OK]

Hint: Window functions keep all rows, unlike GROUP BY [OK]

Common Mistakes:

• Confusing window functions with GROUP BY aggregation
• Thinking window functions reduce row count
• Assuming window functions delete duplicates

2. Which of the following is the correct syntax to calculate a running total of sales using a window function in Snowflake?

easy

A. SELECT SUM(sales) GROUP BY region ORDER BY date FROM sales_data;

B. SELECT sales + PREVIOUS(sales) FROM sales_data;

C. SELECT RUNNING_TOTAL(sales) FROM sales_data;

D. SELECT SUM(sales) OVER (PARTITION BY region ORDER BY date) FROM sales_data;

Solution

1. Step 1: Identify correct window function syntax

   SUM(sales) OVER (PARTITION BY region ORDER BY date) correctly calculates a running total partitioned by region and ordered by date.

2. Step 2: Eliminate incorrect options

   SELECT SUM(sales) GROUP BY region ORDER BY date FROM sales_data; uses GROUP BY which reduces rows, not a window function. Options C and D use invalid functions or syntax.

3. Final Answer:

   SELECT SUM(sales) OVER (PARTITION BY region ORDER BY date) FROM sales_data; -> Option D

4. Quick Check:

   SUM() OVER with PARTITION BY and ORDER BY = B [OK]

Hint: Look for SUM() OVER with PARTITION BY and ORDER BY [OK]

Common Mistakes:

• Using GROUP BY instead of OVER clause
• Using non-existent functions like RUNNING_TOTAL
• Omitting ORDER BY in window function

3. Given the table sales with columns region, date, and amount, what is the output of this query?

SELECT region, date, amount, RANK() OVER (PARTITION BY region ORDER BY amount DESC) AS rank FROM sales;

medium

A. Ranks sales amounts within each region from highest to lowest

B. Ranks sales amounts across all regions ignoring region groups

C. Calculates cumulative sum of amounts per region

D. Returns the total number of sales per region

Solution

1. Step 1: Understand RANK() with PARTITION BY and ORDER BY

   RANK() assigns ranks starting at 1 within each partition (region), ordering by amount descending.

2. Step 2: Interpret the query output

   The query shows each sale with its rank in its region based on amount, highest amount ranked 1.

3. Final Answer:

   Ranks sales amounts within each region from highest to lowest -> Option A

4. Quick Check:

   RANK() OVER PARTITION BY region ORDER BY amount DESC = A [OK]

Hint: RANK() with PARTITION BY ranks within groups [OK]

Common Mistakes:

• Thinking RANK() ignores PARTITION BY
• Confusing RANK() with cumulative sum
• Assuming ranks are across all rows without grouping

4. Identify the error in this Snowflake query:

SELECT employee_id, salary, ROW_NUMBER() OVER (ORDER BY salary) PARTITION BY department FROM employees;

medium

A. ORDER BY cannot be used in window functions

B. ROW_NUMBER() cannot be used with ORDER BY

C. PARTITION BY must come before ORDER BY inside OVER()

D. Missing GROUP BY clause for department

Solution

1. Step 1: Check window function clause order

   In Snowflake, PARTITION BY must appear before ORDER BY inside the OVER() clause.

2. Step 2: Identify syntax error

   The query places PARTITION BY after ORDER BY, which is invalid syntax.

3. Final Answer:

   PARTITION BY must come before ORDER BY inside OVER() -> Option C

4. Quick Check:

   PARTITION BY before ORDER BY in OVER() = D [OK]

Hint: PARTITION BY always before ORDER BY in OVER() [OK]

Common Mistakes:

• Placing PARTITION BY after ORDER BY
• Thinking ROW_NUMBER() disallows ORDER BY
• Adding unnecessary GROUP BY for window functions

5. You want to calculate the average sales per region and also show each sale's rank by amount within its region. Which query correctly combines these using window functions?

hard

A. SELECT region, amount, AVG(amount) PARTITION BY region, RANK() ORDER BY amount DESC FROM sales;

B. SELECT region, amount, AVG(amount) OVER (PARTITION BY region) AS avg_region, RANK() OVER (PARTITION BY region ORDER BY amount DESC) AS rank FROM sales;

C. SELECT region, amount, AVG(amount), RANK() FROM sales GROUP BY region ORDER BY amount DESC;

D. SELECT region, amount, AVG(amount) OVER (), RANK() OVER (ORDER BY amount) FROM sales;

Solution

1. Step 1: Use AVG() as window function partitioned by region

   AVG(amount) OVER (PARTITION BY region) calculates average sales per region without grouping rows.

2. Step 2: Use RANK() partitioned by region ordered by amount descending

   RANK() OVER (PARTITION BY region ORDER BY amount DESC) ranks sales within each region.

3. Step 3: Verify query correctness

   SELECT region, amount, AVG(amount) OVER (PARTITION BY region) AS avg_region, RANK() OVER (PARTITION BY region ORDER BY amount DESC) AS rank FROM sales; correctly uses window functions with proper syntax and clauses.

4. Final Answer:

   SELECT region, amount, AVG(amount) OVER (PARTITION BY region) AS avg_region, RANK() OVER (PARTITION BY region ORDER BY amount DESC) AS rank FROM sales; -> Option B

5. Quick Check:

   AVG() and RANK() with PARTITION BY region = C [OK]

Hint: Use OVER(PARTITION BY region) for both AVG and RANK [OK]

Common Mistakes:

• Using GROUP BY instead of window functions
• Incorrect syntax for window functions
• Omitting PARTITION BY for per-region calculations