customershidden dataset
customer_idbigint
emailstring
source_idint
updated_attimestamp
50,000,000 rowsskewed key
Real schemas, big data
Hidden datasets up to 50M rows, so a naive answer times out instead of passing.
Browse problems →Free forever for core practice
Crack the data engineering interview on real Spark problems.
SQL, PySpark and Databricks questions asked at Amazon, Adobe and Deloitte, graded on hidden datasets with runtime and shuffle metrics.
No card. 20 graded submissions every month on Free.
6,433+ engineers already learn with GeekCoders
Deduplicate customers (PySpark)Accepted 7/7
from pyspark.sql import DataFrame, functions as Ffrom pyspark.sql.window import Window def solve(customers: DataFrame) -> DataFrame: w = (Window.partitionBy("customer_id") .orderBy(F.desc("updated_at"), F.desc("source_id"))) return (customers .withColumn("rn", F.row_number().over(w)) .filter(F.col("rn") == 1) .drop("rn"))Runtime1.42 s
Memory612 MB
Shuffles2
Rows1.28M
Built by Sagar PrajapatiDatabricks practitioner, founder of GeekCoders
6,433+GeekCoders learnersCourses and video lessons
51Graded questionsPySpark, Python, SQL +1 more13Company prep setsAmazon, Adobe, Deloitte +106Certification examsVerifiable, add to LinkedInSound familiar?
Most practice sites train you for a round data teams stopped running. Here is what changes on CodeArena.
You solve it at your desk, then freeze when they ask why you joined that way.
Every accepted answer shows how it ran, and Pro adds optimisation feedback on what it would do at scale: the “why” interviewers ask about.
Optimisation feedback →Practice sites give you three rows of data. Interviews are about skew, nulls and 50 million rows.
Hidden test datasets go up to 50M rows, with skewed keys and nulls. A naive answer times out instead of quietly passing.
Hidden datasets →“Accepted” doesn’t tell you whether your answer shuffled twice and collected to the driver.
The verdict reports runtime, peak memory, shuffle stages and partitions on every submission.
Run metrics →The posting asks for Databricks and window functions. You’re grinding tree traversal.
GeekMatch AI reads a real posting against your resume, names the gaps, and points you at the questions that close them.
GeekMatch AI →What makes this different
The same tools and data shapes your next team uses, not toy inputs.
Accepted7 / 7 tests · beats 87%
Runtime1.42 s
Peak memory612 MB
Shuffle stages2
Partitions8
Graded on how it ran
Runtime, memory, shuffle stages and partitions on every submission.
Try a problem →
21 questionsAsked at real companies
Prep sets for Amazon, Adobe, Deloitte and more, by experience band.
All company sets →Proof you can share
Timed exams that issue a credential anyone can verify, with one-click LinkedIn.
View a sample credential →6 certifications you can earn
Timed, graded on the server, and verifiable by anyone you send the link to.
Apache IcebergAssociate
PySpark DeveloperAssociate
Python for Data EngineeringAssociate
Snowflake for Data EngineersAssociateIssued by GeekCoders. Not vendor certifications.
Who built this
Sagar Prajapati
6,433+GeekCoders learners
50k+ LinkedIn followers
CodeArena turns the GeekCoders course questions into problems you can run and get graded on.
Core practice is free, and stays free.
Upgrade only when you want unlimited grading and the exams.
Free
₹0
- Every published question, filters and company pages
- The full editor, hidden test cases and verdicts
- A public profile you can share
- 20 graded submissions a month
- 5 GeekMatch AI runs a month
- Every free question, solved as often as you like
- 1 exam a month
ProRecommended
₹499/ month
- Everything in Free
- Optimization feedback on accepted solutions
- Company prep sets and experience tracks
- Timed mock interview rounds
- All certification exams and credentials
- Every Pro question
- 2 exams a month
- 15 GeekMatch AI runs a month
Pro Max
₹999/ month
- Everything in Pro
- Certification exams with no monthly cap
- Every course, with its quizzes
- Databricks advanced labs
- 25 GeekMatch AI runs a month
Questions before you start
Do I need to be a data engineer already?
No. Questions are tagged by experience band. The lowest starts at joins, grouping and reading a file in Python; the highest is distributed systems, cost and governance. You pick the band, not us.
How is this different from LeetCode?
It grades SQL, Python, PySpark, Spark, Databricks and data modelling against hidden datasets in a real sandbox, and reports runtime, memory and shuffle stages alongside the verdict.
What do I get without paying?
Every published question with its schema, worked example and public tests, the editor, and a verdict. Free includes 20 graded submissions a month against the hidden datasets. Nothing is deleted when you run out — grading pauses until the month turns over.
Is the certification a vendor certification?
No. It is our exam and our credential, issued by GeekCoders and verifiable on a public page. It is not affiliated with or endorsed by Databricks, Snowflake, Microsoft or the Apache Software Foundation.
Can I just paste an answer from a model?
You can, and it will pass. It will also report the same runtime and shuffle count to you that it would to an interviewer, and the round you are preparing for ends with somebody asking you to explain it.
Your next interview is a data engineering interview.
Start practising freeJoin 6,433+ engineers learning with GeekCoders