eBay · Primly Community

eBay data engineer interview: pipelines and SQL were the whole game

de_derek · 4 replies

Went through the eBay data engineer loop earlier this year for a senior DE role at the San Jose office. Four rounds total: one phone screen with a recruiter, one technical screen, then two onsite rounds (one data/SQL, one system design + behavioral). Remote options were available but I did it in person.

The technical screen was a SQL round on Coderpad. They gave me a messy schema that looked like an actual ecommerce orders/items/sellers setup, which honestly felt very close to real eBay data. Questions were: window functions, a multi-step aggregation problem, and one where you had to reason about data quality (null handling, deduplication). No LeetCode, no algorithms, just SQL. Medium-hard SQL, not trivial.

The onsite SQL round was similar but went deeper. Expect to talk about query optimization, indexing strategies, and at one point they asked how I'd redesign a pipeline that was getting slow at 10B rows. I walked through partitioning, incremental loading, and materialized views. They nodded but didn't tip their hand.

System design round was pipeline architecture. I got: design a real-time search indexing pipeline for eBay listings. So your seller posts an item and within 60 seconds it's searchable. I talked through Kafka for event streaming, a Flink job for transformation/enrichment, writing to Elasticsearch, and handling schema evolution. The interviewer was genuinely engaged here, asked good follow-up questions about backfill strategy and handling deletes.

Behavioral was standard: tell me about a time you dealt with bad data in prod, a time you influenced without authority. Two questions, maybe 30 minutes total.

Tech stack they mentioned internally: Spark, Airflow, Kafka, Hive, some internal tooling I didn't recognize. They seem to be in the middle of a Hadoop-to-cloud migration (AWS), so expect questions about that transition.

Total time from recruiter reach-out to offer: about 6 weeks. Offer was for L5 equivalent. If you're prepping, lock in your SQL window functions cold and have a real pipeline design story ready.

4 replies

analyst_ana

This is super helpful. Did the SQL coderpad round have a time limit? Like did you feel rushed or was there space to think out loud?

de_derek

No hard timer on screen but it was a 45 min slot and the interviewer kept things moving. I'd say roughly 15 min per question. Thinking out loud helped, they'd sometimes nudge if I was going in the wrong direction.

ds_dmitri

The Kafka/Flink combo for real-time indexing is pretty standard but it's good to know they ask about it at this depth. Did they want you to code any of the pipeline or just whiteboard/discuss?

infra_ines

The 'design a real-time search indexing pipeline' prompt is almost identical to what I got at a different large ecommerce company. Think they share interview playbooks? Or just a naturally obvious problem to ask at that scale.