went through the Meta data engineering loop last month for a DE role on the infrastructure side (Hive/Spark/Presto stack). sharing specifics because most posts online conflate DE and DS interviews and they're genuinely different.
what Meta DE interviews test:
SQL is the first thing. but for DE specifically, they push harder on performance. not just "write the query" but "this query runs for 3 hours on a 50TB table, why, and what do you do?" answers they want to hear: partition pruning, avoiding cartesian joins, columnar storage (why Parquet), understanding shuffle cost in Spark.
pipeline design round: this is the system design equivalent for DE. they gave me: "design a pipeline that ingests clickstream data from mobile apps, enriches it with user metadata, and produces a daily user-level aggregation table. SLA is 4 hours after midnight."
things that mattered: idempotency (if the job reruns at 2am what happens?), late-arriving data handling, alerting strategy, testing strategy. I drew a DAG on the whiteboard and walked through each node. they cared that I thought about failure modes, not just the happy path.
coding round: yes there's still coding for DE. mine was not leetcode-hard. one question was string parsing to extract a log format, one was more of a data transformation problem. they care that you can write clean Python or Scala, not that you can do DP on a grid.
tooling they asked about: "how does Spark handle skew and what do you do about it?" "when would you choose Presto over Hive?" "what does watermarking mean in streaming and when do you use it?"
you don't need to have used all of these but knowing the conceptual tradeoffs is table stakes at senior level.
what surprised me: they asked almost no ML infrastructure questions. this was a pure data engineering loop. if you're coming from an MLE background expecting to talk about feature stores and training pipelines, that's a different role.
overall vibe: more rigorous on fundamentals than I expected. they really do care that you understand what's happening under the hood.