**Techniques**: (1) **dropDuplicates** on keys. (2) **Window**: row_number() over (partition by key order by ts desc) — keep latest. (3) **Delta MERGE** with dedup in merge key. (4) **Hash** (e.g., md5 of cols) for idempotency. (5) **Bloom filter** for approximate.
**When**: Event streams (at-least-once), CDC, replays.
**Why**: Duplicates corrupt analytics, billing, inventory.
**Scalability Trade-offs**: Window on 1B rows = shuffle. Partition by key....
The complete answer continues with detailed implementation patterns, architectural trade-offs, and production-grade considerations. It covers performance optimization strategies, common pitfalls to avoid, and real-world examples from companies like Virtusa. The answer also includes follow-up discussion points that interviewers commonly explore.
Continue Reading the Full Answer
Unlock the complete expert answer with code examples, trade-offs, and pro tips - plus 1,863+ more.
Or upgrade to Platform Pro - $39
Engineers who used these answers got offers at
AmazonDatabricksSnowflakeGoogleMeta
According to DataEngPrep.tech, this is one of the most frequently asked Spark/Big Data interview questions, reported at 1 company. DataEngPrep.tech maintains a curated database of 1,863+ real data engineering interview questions across 7 categories, verified by industry professionals.