Question 1

Describe a scenario where partitioning and bucketing would improve query performance.

Accepted Answer

Situation: An events table with billions of rows serving time-range and user-level analytics. Task: Achieve sub-second query latency while controlling storage and compute costs. Why Partitioning: Partition pruning at read time eliminates entire data scans—a query filtering by date_range only touches relevant partition dirs. This reduces I/O by orders of magnitude (e.g., 365 partitions → scan 1 vs all)....

Question 2

How do you handle late-arriving data in Spark Structured Streaming?

Accepted Answer

Watermark: Defines max lateness (e.g., 10 min). Events older than (max_event_time - watermark) are dropped. State: Kept for aggregations within watermark; beyond that, state is purged to avoid unbounded growth. Output modes: Append (only finalized results), Update (changed rows), Complete (full result). Why: Trade-off between latency and correctness—tighter watermark = less state, more dropped late data. Looser = more state, more complete results....

Question 3

What is the small-file problem in Spark, and how do you solve it?

Accepted Answer

Problem: Thousands of tiny files (KB–MB) cause metadata overhead, slow S3/HDFS listing, many small tasks, I/O thrashing. Root causes: High parallelism, over-partitioning, streaming micro-batches. Why it hurts: Each file = task; 10K files = 10K tasks = scheduling overhead. S3 LIST is rate-limited; listing 100K files can take minutes. Solutions: 1) coalesce/repartition before write to target 128MB–1GB per file. 2) Delta Lake/Spark auto-compaction. 3) Larger partition sizes....

Question 4

How do you optimize Spark jobs for better performance? Mention at least 5 techniques.

Accepted Answer

1) Broadcast joins for small tables—avoid shuffle. 2) Predicate pushdown—filter at source (Parquet/ORC) to reduce scan. 3) Partition tuning—spark.sql.shuffle.partitions ~2–4× cores; match partition columns to filter/join keys. 4) Cache only when reused; unpersist when done to free memory. 5) Prefer Spark SQL over UDFs—Catalyst optimization. 6) Skew handling—salted keys, AQE skew join. 7) Kryo serialization for RDD; avoid Java default. 8) Coalesce before write to avoid small files....

Question 5

What are decorators in Python, and how do they work?

Accepted Answer

Decorator: Higher-order function that wraps another function; @decorator syntax. Implementation: def timer(f): def wrapper(*args, **kwargs): start=time(); r=f(*args,**kwargs); print(time()-start); return r; return wrapper. Why: Cross-cutting concerns—logging, retries, auth, caching—without cluttering business logic. In data pipelines: @retry, @logged, @rate_limited on API fetchers. Scalability: Decorators add indirection; overuse can make stack traces deep....

Question 6

Explain the difference between args and kwargs in Python.

Accepted Answer

*args: Variable positional args, received as tuple. **kwargs: Variable keyword args, received as dict. Why: Flexible signatures—wrappers, decorators, config-driven functions. Example: def run_pipeline(*sources, **config): ... allows run_pipeline('s3://a','s3://b', parallelism=4). Order: (regular, *args, keyword-only, **kwargs). Pitfall: Passing through—func(*args, **kwargs)—preserves interface. In data pipelines: **kwargs for optional connector params (region, timeout, retries)....

Question 7

Explain the trade-offs between batch and real-time data processing. Provide examples of when each is appropriate.

Accepted Answer

**Why the distinction matters**: Systems are optimized for different access patterns. Batch assumes bounded, complete datasets; streaming assumes unbounded, infinite data. The choice cascades into storage layout, compute model, operational complexity, and cost structure. **Batch**: Designed for high-throughput, bounded processing. Lower cost per record due to amortized compute (e.g., spot instances). Easier to reason about correctness (all-or-nothing). Trade-off: End-to-end latency is O(hours)....

Question 8

Retrieve the most recent sale_timestamp for each product (Latest Transaction).

Accepted Answer

**SQL (full row)**: `SELECT * FROM (SELECT *, ROW_NUMBER() OVER (PARTITION BY product_id ORDER BY sale_timestamp DESC) AS rn FROM sales) t WHERE rn = 1` — Returns the full record. **SQL (timestamp only)**: `SELECT product_id, MAX(sale_timestamp) FROM sales GROUP BY product_id` — Use when you only need the timestamp. **PySpark**: `df.withColumn('rn', F.row_number().over(Window.partitionBy('product_id').orderBy(F.desc('sale_timestamp')))).filter(F.col('rn') == 1).drop('rn')` — Full row....

Swiggy Data Engineer Interview Questions

+14 More Questions with Expert Answers

Other Companies

Swiggy Data Engineer Interview Questions

+14 More Questions with Expert Answers

Other Companies