Datadog · Primly Community

Datadog data engineer interview: pipelines, SQL, and a curveball I didn't see coming

de_derek · 4 replies

Finished the Datadog data engineer loop two weeks ago (NYC role, hybrid). Sharing notes while they're fresh because I couldn't find much specific to the DE track when I was prepping.

Five rounds total:

Recruiter screen - 30 min, pretty standard. They asked about my pipeline experience, specifically Kafka and Spark. No gotcha questions, just making sure I wasn't a resume ghost.

Technical phone screen - 45 min with a senior DE. Split roughly 50/50 between SQL and a system design-ish discussion around pipeline architecture. The SQL was not baby stuff. They gave me a multi-table scenario involving event data (very much in keeping with what Datadog actually does) and asked me to write a query to compute rolling averages and detect anomalies in a metrics stream. Window functions, lag/lead, the works. Know your CTEs.

Take-home - They sent a small dataset and asked me to build a simple ETL pipeline, document the schema decisions, and write SQL to answer three analytic questions. I used Python + duckdb locally and wrote it up in a notebook. Took me about 3 hours but I went deep on the write-up. I think that mattered.

On-site (virtual) round 1 - More SQL plus a deep dive into my take-home. They asked why I chose my partitioning strategy, what I'd do differently at 100x scale, and whether I'd considered streaming vs batch for one of the transforms. Have answers for these in advance.

On-site round 2 - Behavioral. Focused on cross-functional work and how I handle pipeline failures in prod. Standard STAR stuff. Nothing Datadog-specific but be ready for: "tell me about a time a data quality issue made it into a downstream dashboard and what you did."

The curveball: one interviewer asked me to walk through how I'd instrument a pipeline to emit custom metrics to a monitoring system. Basically asking if I actually understand observability tooling. Felt like a Datadog-culture question as much as a technical one. Worth thinking through.

Total timeline: applied via referral, recruiter reached out in 5 days, full loop over 3 weeks. Still waiting on decision as of today.

4 replies

analyst_ana

This is super helpful. Did they specify SQL dialect or was it platform-agnostic? I always get nervous when I write Postgres-flavored stuff and they're running BigQuery.

de_derek

They said "any dialect, just explain your choices." I went with standard SQL with a few Postgres-isms and called them out when I used them. Didn't seem to be an issue.

ds_dmitri

The rolling average + anomaly detection SQL question is a classic signal question for DE loops at observability companies. They're checking if you actually think in time series or if you just know SELECT. Good prep advice: practice window functions on event-log shaped data, not just transactional.

infra_ines

The observability instrumentation curveball makes total sense at Datadog. They want DEs who treat pipelines like prod services, not batch scripts that run and pray. If you're prepping, have an answer ready for how you'd add metrics/traces/logs to a pipeline you built.