Joins, window functions, deduplication
Out loud, in order: clarify, build it, check it, explain it.
Practise SQLData engineering interview · Big Tech & Enterprise Software
SQL + Python pairing and a strong system-design round with explicit prompts (10TB-in-5-minutes, GDPR-deletion data lake).
01
And what its data engineers work with.
02
Stages in the order candidates meet them.
6 stages. 3 rounds have reported questions, described below.
03
The topics it covers and the questions candidates reported.
3 reported questions. See them below
2 reported questions. See them below
3 reported questions. See them below
04
By round. Questions marked Expected are written from the company's standards.
SQLMedium confidence
Python / CodingMedium confidence
System DesignMedium confidence
Nobody reported these exact questions at Salesforce. They are the questions a loop for each of its top topics is known to ask, so prepare for them as likely, not as confirmed.
05
Ranked by how much of the evidence points at each topic.
06
Do the first group before the second.
SQL (window, dedup, cohort, incremental, perf)
Do first (3)System design (batch vs stream, storage choice, idempotency/exactly-once/DR)
Python pipeline transformations
Spark/Hadoop/Hive
ThenSalesforce values behavioral
07
Why candidates were turned down, as reported.
Shallow system design (no idempotency/exactly-once/DR/storage trade-offs)
Design is named in 27 of 50 company accounts.Weak SQL cohort/incremental
SQL is named in 37 of 50 company accounts.Thin Spark
Spark is named in 25 of 50 company accounts.08
3 reported, 2 composite. Each opens and closes. Composites are labelled.
This candidate reached the fifth round. Round three was a SQL test of five questions on HackerRank, rated above average to hard. Then came Python pair coding with a lead engineer and a behavioral round with the hiring manager. They were rejected.
What to take from it The SQL test is a real filter. Practise hard SQL on a timer.
An online test with coding and SQL, then technical rounds on project background and DevOps setup, two medium coding problems, SQL, system design and Python.
What to take from it Be ready to talk about how you deploy and set up infrastructure, not only code.
A short screen, then three technical rounds of about 50 minutes each. Mostly SQL and Python, with design questions built on scenarios.
What to take from it Expect scenario questions, so practise explaining your thinking.
Composite walkthroughs. Built from the company's documented loop, not from one person's account.
This is a composite, not one person's account. It follows the usual order: a recruiter call, a technical screen with SQL and an algorithm, then an onsite with SQL (removing duplicates, cohorts), Python pair coding, system design (moving 10 TB in 5 minutes, a GDPR data lake) and behavioral.
What to take from it Use it as a map of the loop.
This is a composite, not one person's account. It is built from the loop we have on record for Salesforce. Questions reported for this round: Joins, window functions, deduplication; Incremental transformations, cohort-style queries; Performance awareness.
What to take from it Practise these questions before you reach this round.
DE-specific verbatim first-hand was thin; the system-design prompts are aggregator-reported examples.
09
Their reported questions, one at a time, against a stopwatch.
Joins, window functions, deduplication
Out loud, in order: clarify, build it, check it, explain it.
Practise SQL10
Pick your level and the weeks you have.
Built from what Salesforce asks, strongest first. For the wider basics at your level, follow the Mid level plan alongside it.
Pipeline and system design
Read: Data pipeline chaptersDrill: Pipeline practiceSay it aloud: Data pipeline questionsFull plan: Ace the pipeline round
Python and coding
Read: Python chaptersDrill: Python practiceSay it aloud: Python questionsFull plan: Ace the Python round
Behavioral and communication
Read: Behavioral storytellingSay it aloud: Behavioral questions
Rehearse the Salesforce loop
Run the question deckThen repeat your weakest topic once more.