SQL Practice Problems
1,500 SQL practice problems with worked solutions, graded against a real Postgres warehouse. Sourced from verified data engineering interview reports. 16 topic catalogs, 4 difficulty bands.
1,500 SQL practice problems scored against a real Postgres 16 instance with randomized seed data on every submission. Every problem traces back to a verified data engineering interview report. Pick a topic below or open a random problem to start.
Know the patterns before the interviewer asks them.
How often each topic shows up in a real SQL round
Bars show how often each topic appears across verified interview reports. The first four cover most of every round. The bottom topics are senior-signal territory: rare enough that volunteering them lands well, common enough at L5 and above that you can't skip them.
- 01GROUP BY and aggregation142 problemsEasy → Medium
- 02Joins (INNER, LEFT, anti, self)168 problemsEasy → Hard
- 03Window functions124 problemsMedium → Hard
- 04CTEs and recursive queries96 problemsMedium → Hard
- 05Date and time58 problemsMedium
- 06NULL handling (COALESCE, NULLIF)52 problemsEasy → Medium
- 07Subqueries and EXISTS88 problemsMedium → Hard
- 08CASE WHEN and conditional logic64 problemsEasy → Medium
- 09Deduplication patterns32 problemsMedium
- 10RANK, DENSE_RANK, ROW_NUMBER48 problemsMedium
- 11Self joins and inequality joins42 problemsMedium → Hard
- 12PARTITION BY patterns56 problemsMedium → Hard
- 13Gaps and islands38 problemsHard
- 14Pivot and unpivot36 problemsMedium → Hard
- 15String and regex28 problemsEasy → Medium
- 16Recursive CTEs22 problemsHard
Why a real database changes how you practice
LeetCode and most online SQL trainers check your output by string-matching against a single reference dataset. You can pass by reading the expected output and writing SQL that happens to produce those rows. That's not how a real interview assesses you, and it's not how this grader works.
Every submission spins up a fresh Postgres database with randomized seed data inside the problem's schema. A query that hard-codes a customer_id, depends on insertion order, or uses LIMIT without an ORDER BY tiebreaker passes one or two seeds and fails the rest. The feedback names which seed broke and what changed between it and the seeds that passed.
NULL handling is the other thing this catches. A naive COUNT(col) versus COUNT(*), an INNER JOIN that should have been LEFT, a comparison against NULL that silently returns UNKNOWN. The seeds carry NULLs in the columns where interviewers know to test.
What a scored problem looks like
Medium difficulty, window function with dedup. Roughly one in three real SQL rounds opens with a problem of this shape.
Prompt: Find users whose third login fell within 30 days of signup. The grader runs your query against ten random seed datasets; rows must match for every one. ORDER BY is checked. NULLs in login_at trip the input validator before scoring begins.
-- Find users whose third login was within 30 days of signup.
-- Schema: users(user_id, signup_at), logins(user_id, login_at)
WITH ranked AS (
SELECT
user_id,
login_at,
ROW_NUMBER() OVER (PARTITION BY user_id ORDER BY login_at) AS n
FROM logins
)
SELECT u.user_id
FROM users u
JOIN ranked r ON r.user_id = u.user_id AND r.n = 3
WHERE r.login_at <= u.signup_at + INTERVAL '30 days';The common wrong answer uses LIMIT 1 OFFSET 2 inside a per-user subquery. It passes one fixture and fails the other nine because LIMIT without ORDER BY is non-deterministic in Postgres.
How the adaptive engine picks the next problem
After five or six problems, the engine knows which topic-difficulty cells you've cleared. The next problem is sampled from the edge: the topic you most recently got wrong, at the difficulty band one rung below where you stalled.
There's no streak counter, no XP, no daily quest. The mechanic borrows from the same spaced-repetition research that runs Anki: a problem you missed three weeks ago surfaces again before you've forgotten the fix, and a topic you've nailed three times in a row drops out of rotation for a while.
If you want to override, set the topic and difficulty filters explicitly. The engine logs that choice and treats it as signal too.
Top Active Senders per Channel
Top three messages per channel by replies.
Pulled from debriefs where SQL was the gate.
How the grader actually runs your query
3 numbers worth knowing before you submit anything.
Dialect notes
The grader runs Postgres 16. About 85 percent of the bank uses ANSI-portable syntax that runs the same in Snowflake, BigQuery, Redshift, and MySQL. Postgres-specific features (FILTER on aggregates, RANGE BETWEEN window frames, lateral joins, DISTINCT ON) are tagged on the problem so you can skip them or use them as a chance to learn.
If you're targeting Spark SQL or BigQuery and want to stay dialect-pure, the 'dialect-clean' toggle at the top of /problems narrows the bank to portable syntax only.
How to mix your practice time
What works for candidates who pass: 80 to 150 problems total, distributed by difficulty rather than topic. The bars show the recommended share of your time, not the share of problems.
- Easy30 problems
Build fluency with the grader and the schema panel. Get GROUP BY and basic JOIN to reflex.
- Medium60 problems
Distributed across joins, window functions, GROUP BY with HAVING, CTEs, conditional aggregation. Most of your time goes here.
- Hard20 problems
Weighted toward whichever topic the company you're targeting tends to ask. Saved for the final week.
- Mock interviews10 sessions
Timed, vague prompts, AI follow-ups. The practice that closes the gap between knowing the answer and saying it out loud.
Common questions
Are these SQL practice problems really free?+
How is this different from LeetCode SQL?+
Can I practice SQL online for free without signup?+
What SQL dialect should I expect?+
What are the best SQL practice problems for interview prep?+
How many SQL problems should I solve before a loop?+
Are problems sourced from real interviews?+
Pick the topic you've been avoiding
- 01
Active recall beats re-reading by 50%
Cognitive-science meta-reviews (Dunlosky et al., 2013) rank practice testing as a top-tier study technique, while re-reading and highlighting rank near the bottom
- 02
76% of hiring managers reject on the coding task, not the resume
From HackerRank's 2024 Developer Skills Report. Candidates who look strong on paper still fail the live screen if they haven't done timed, executable practice
- 03
Five problem shapes cover 80% of data engineer loops
Dedup, sessionization, top-N-per-group, slowly-changing dimensions, partition tricks. Writing the shapes by hand turns the unfamiliar into pattern recognition