Google · Primly Community

Google data engineer interview, pipelines and SQL: what they actually focus on

de_derek · 4 replies

Just finished my Google data engineer interview loop for a L5 role out of the NYC office. Sharing the breakdown because I couldn't find a good current post when I was prepping in early 2026.

The loop was 5 rounds plus a hiring manager chat. Here's what each one covered:

Coding (x2): Felt like LeetCode medium-hard, but both were explicitly data-flavored. One problem was about processing a stream of events and computing a rolling window aggregation. The other was a graph traversal problem in the context of dependency resolution for a DAG pipeline. Don't skip graph problems even if you're a pure SQL person.

SQL/Analytics: This round surprised me. They gave a schema with 4 tables (users, sessions, events, conversions) and asked me to write progressively complex queries. Started with a basic join, then windowing functions, then an efficiency question about how I'd rewrite a certain query to avoid a full scan. I'd say this was the round I felt most confident in, but they go deep. Know PARTITION BY cold.

System Design (data): Designing a pipeline that ingests real-time clickstream data, lands it in a warehouse, and serves aggregated dashboards with low latency. They care about: fault tolerance, exactly-once semantics, partitioning strategy, and backfill strategy. I talked through Pub/Sub to Dataflow to BigQuery and they asked good follow-ups about late-arriving data.

Behavioral: Standard Google Googleyness rubric. Leadership, ambiguity, cross-functional conflict. Have 5-6 STAR stories ready and be ready to go deeper on each one.

Overall the loop was tougher than I expected for a DE role. They treat it more like an SWE loop with data specialization layered on top, not a pure analytics or warehousing interview. LeetCode hard problems showed up in my specific loop.

Offer came back at L5 in the $320-340k TC range for NYC. Took about 3.5 weeks from recruiter screen to offer call. The debrief period was genuinely nerve-wracking because I knew I'd had a shaky coding round.

Happy to answer follow-ups.

4 replies

ds_dmitri

The SQL round description matches what a friend went through for Google's DS track too. The windowing function depth is real. Did they care about BigQuery-specific syntax or were they looking for standard SQL and then asking about BQ specifics separately?

de_derek

They let me write standard SQL first, then asked how I'd optimize it in BigQuery specifically. Partitioning on a date column came up, and they asked about clustering vs partitioning tradeoffs. Knowing BQ isn't required but it helps a lot for the follow-ups.

analyst_ana

This is really helpful. I'm targeting L4 DE roles. Do you think the SQL depth is similar at L4 or does it ease off a bit at the lower level?

staff_steph

The 'SWE loop with data specialization layered on' framing is accurate for basically all of big tech DE roles at this point. They stopped treating DE as a separate hire category a few years ago. If you can't grind leetcode you're going to struggle regardless of how good your pipeline architecture knowledge is.