PARTITION BY for grouping windows in PostgreSQL - Step-by-Step Execution

Concept Flow - PARTITION BY for grouping windows

Start with full table

↓

Define window with PARTITION BY

↓

Split rows into groups by PARTITION BY column

↓

Apply window function within each group

↓

Combine results preserving original row order

↓

Output result with window function values per row

The PARTITION BY clause splits the table into groups. Then the window function runs separately on each group, keeping the original row order.

Execution Sample

PostgreSQL

SELECT dept, emp, salary,
       RANK() OVER (PARTITION BY dept ORDER BY salary DESC) AS rank
FROM employees;

This query ranks employees by salary within each department separately.

Execution Table

Step	Row (dept, emp, salary)	Partition Group	Order in Group	RANK() Value
1	(Sales, Alice, 7000)	Sales	1	1
2	(Sales, Bob, 6000)	Sales	2	2
3	(Sales, Carol, 6000)	Sales	3	2
4	(HR, Dave, 8000)	HR	1	1
5	(HR, Eve, 7500)	HR	2	2
6	(HR, Frank, 7000)	HR	3	3
7	(IT, Grace, 9000)	IT	1	1
8	(IT, Heidi, 8500)	IT	2	2
9	(IT, Ivan, 8500)	IT	3	2
Exit	All rows processed	-	-	Window function applied per partition

💡 All rows processed; RANK() calculated within each department partition.

Variable Tracker

Variable	Start	After 1	After 2	After 3	After 4	After 5	After 6	After 7	After 8	After 9	Final
Current Row	None	(Sales, Alice, 7000)	(Sales, Bob, 6000)	(Sales, Carol, 6000)	(HR, Dave, 8000)	(HR, Eve, 7500)	(HR, Frank, 7000)	(IT, Grace, 9000)	(IT, Heidi, 8500)	(IT, Ivan, 8500)	All rows processed
Partition	None	Sales	Sales	Sales	HR	HR	HR	IT	IT	IT	All partitions assigned
Rank	None	1	2	2	1	2	3	1	2	2	Ranks assigned per partition

Key Moments - 3 Insights

Why do some employees have the same rank number in the same department?

Because RANK() assigns the same rank to rows with the same salary within the partition, as shown in rows 2 and 3 for Sales and rows 8 and 9 for IT in the execution_table.

Does PARTITION BY change the order of rows in the final output?

No, PARTITION BY groups rows for the window function but the original row order is preserved in the output, as seen in the execution_table where rows appear in original order.

What happens if PARTITION BY is omitted?

The window function runs over the entire table as one group, so ranks would be assigned globally, not per department.

Visual Quiz - 3 Questions

Look at the execution_table, what is the RANK() value for (HR, Eve, 7500)?

A 3

B 1

C 2

D None

💡 Hint

Check the row with (HR, Eve, 7500) in the execution_table under the RANK() Value column.

At which step does the partition change from Sales to HR?

A Step 4

B Step 5

C Step 3

D Step 6

💡 Hint

Look at the Partition Group column in the execution_table to see when it switches from Sales to HR.

If we remove PARTITION BY dept, how would the ranks change?

A Ranks would reset for each employee

B Ranks would be assigned across all employees regardless of department

C Ranks would be all 1

D Ranks would be assigned randomly

💡 Hint

Recall that PARTITION BY groups rows; removing it means one big group for ranking.

Concept Snapshot

PARTITION BY splits rows into groups for window functions.
Window functions run separately in each group.
Original row order stays the same.
Example: RANK() OVER (PARTITION BY col ORDER BY col2).
Same values get same rank within partition.
Without PARTITION BY, function runs on whole table.

Full Transcript

The PARTITION BY clause in SQL divides the table into groups based on the specified column(s). Then, window functions like RANK() operate within each group independently. This means each partition gets its own ranking or calculation. The original order of rows is preserved in the output. For example, ranking employees by salary within each department uses PARTITION BY department. Rows with the same salary get the same rank. If PARTITION BY is omitted, the window function treats the entire table as one group. This visual trace shows each row's partition, order within the partition, and the rank assigned step-by-step.

Practice

1. What does the PARTITION BY clause do in a window function in PostgreSQL?

easy

A. It filters rows before the window function is applied.

B. It divides the rows into groups to perform calculations within each group.

C. It sorts the entire table before applying the window function.

D. It removes duplicate rows from the result set.

Solution

1. Step 1: Understand the role of PARTITION BY

   PARTITION BY groups rows so window functions calculate results within each group separately.

2. Step 2: Differentiate from filtering and sorting

   Filtering removes rows, sorting orders rows, but PARTITION BY only groups rows without removing or sorting them globally.

3. Final Answer:

   It divides the rows into groups to perform calculations within each group. -> Option B

4. Quick Check:

   PARTITION BY = grouping rows for window functions [OK]

Hint: Remember: PARTITION BY groups rows, it does not filter or sort [OK]

Common Mistakes:

• Confusing PARTITION BY with WHERE filtering
• Thinking PARTITION BY sorts all rows
• Assuming PARTITION BY removes duplicates

2. Which of the following is the correct syntax to use PARTITION BY in a window function?

easy

A. SELECT col, SUM(value) OVER (PARTITION BY col ORDER BY date) FROM table;

B. SELECT col, SUM(value) PARTITION BY col FROM table;

C. SELECT col, SUM(value) OVER PARTITION BY col FROM table;

D. SELECT col, SUM(value) WHERE PARTITION BY col FROM table;

Solution

1. Step 1: Identify correct window function syntax

   The correct syntax uses OVER() with PARTITION BY inside parentheses.

2. Step 2: Check each option

   SELECT col, SUM(value) OVER (PARTITION BY col ORDER BY date) FROM table; correctly uses SUM(value) OVER (PARTITION BY col ORDER BY date). Others misuse PARTITION BY outside OVER or with WHERE.

3. Final Answer:

   SELECT col, SUM(value) OVER (PARTITION BY col ORDER BY date) FROM table; -> Option A

4. Quick Check:

   Window functions use OVER() with PARTITION BY inside [OK]

Hint: PARTITION BY must be inside OVER() parentheses [OK]

Common Mistakes:

• Placing PARTITION BY outside OVER()
• Using WHERE instead of PARTITION BY
• Omitting OVER() entirely

3. Given the table sales with columns region, month, and amount, what will this query return?

SELECT region, month, amount, SUM(amount) OVER (PARTITION BY region ORDER BY month) AS running_total FROM sales ORDER BY region, month;

medium

A. Monthly sales amount without any cumulative calculation.

B. Total sales amount for all regions combined.

C. An error because PARTITION BY cannot be combined with ORDER BY.

D. A running total of sales amount per region ordered by month.

Solution

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

   SUM() OVER (PARTITION BY region ORDER BY month) calculates a running total per region ordered by month.

2. Step 2: Analyze the output

   Each row shows the cumulative sum of amount for that region up to the current month.

3. Final Answer:

   A running total of sales amount per region ordered by month. -> Option D

4. Quick Check:

   PARTITION BY groups, ORDER BY defines running total order [OK]

Hint: Running totals use PARTITION BY with ORDER BY inside OVER() [OK]

Common Mistakes:

• Thinking it sums all regions together
• Ignoring ORDER BY for running total
• Assuming syntax error with ORDER BY in window

4. Identify the error in this query:

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

medium

A. Missing parentheses around PARTITION BY and ORDER BY clauses.

B. RANK() cannot be used with PARTITION BY.

C. ORDER BY cannot be used inside window functions.

D. employee_id must be included in PARTITION BY.

Solution

1. Step 1: Check window function syntax

   Window functions require OVER() with parentheses enclosing PARTITION BY and ORDER BY.

2. Step 2: Identify missing parentheses

   The query misses parentheses after OVER, causing syntax error.

3. Final Answer:

   Missing parentheses around PARTITION BY and ORDER BY clauses. -> Option A

4. Quick Check:

   OVER() must enclose PARTITION BY and ORDER BY [OK]

Hint: Always use parentheses after OVER() for window clauses [OK]

Common Mistakes:

• Omitting parentheses after OVER
• Thinking ORDER BY is not allowed in window functions
• Assuming all columns must be in PARTITION BY

5. You want to calculate the average sales per product category and also show each sale's rank within its category by sale date. Which query correctly uses PARTITION BY to achieve this?

hard

A. SELECT product, category, sale_date, amount, AVG(amount) OVER (PARTITION BY sale_date), RANK() OVER (PARTITION BY product ORDER BY category) FROM sales;

B. SELECT product, category, sale_date, amount, AVG(amount) PARTITION BY category, RANK() PARTITION BY category ORDER BY sale_date FROM sales;

C. SELECT product, category, sale_date, amount, AVG(amount) OVER (PARTITION BY category), RANK() OVER (PARTITION BY category ORDER BY sale_date) FROM sales;

D. SELECT product, category, sale_date, amount, AVG(amount) OVER (ORDER BY category), RANK() OVER (ORDER BY sale_date) FROM sales;

Solution

1. Step 1: Use AVG() with PARTITION BY category

   AVG(amount) OVER (PARTITION BY category) calculates average sales per category.

2. Step 2: Use RANK() with PARTITION BY category ordered by sale_date

   RANK() OVER (PARTITION BY category ORDER BY sale_date) ranks sales within each category by date.

3. Step 3: Verify syntax correctness

   SELECT product, category, sale_date, amount, AVG(amount) OVER (PARTITION BY category), RANK() OVER (PARTITION BY category ORDER BY sale_date) FROM sales; uses correct OVER() syntax with PARTITION BY and ORDER BY clauses for both functions.

4. Final Answer:

   SELECT product, category, sale_date, amount, AVG(amount) OVER (PARTITION BY category), RANK() OVER (PARTITION BY category ORDER BY sale_date) FROM sales; -> Option C

5. Quick Check:

   Use PARTITION BY category inside OVER() for grouping [OK]

Hint: Use OVER(PARTITION BY category) for grouping in window functions [OK]

Common Mistakes:

• Omitting OVER() or PARTITION BY
• Mixing PARTITION BY columns incorrectly
• Using PARTITION BY without parentheses