When would you choose partitioning over bucketing, or vice versa?

Partitioning divides data by column values (e.g., date, region); enables partition pruning for range filters. Bucketing hashes keys into fixed buckets; optimizes joins and GROUP BY on the bucket key. When to use partitioning: Filter by partition column (time-series, region). When to use bucketing: Join/group by high-cardinality key; need even distribution when partitioning alone causes skew. Combine: Partition by date, bucket by user_id—prune on date, co-locate users for joins. Scalability trade-offs: Partition by high-cardinality = explosion; bucket count too high = small files. Cost implications: Partition pruning cuts scan cost; bucketing reduces shuffle. Best practice: Partition for time-series; bucket for join keys when partitioning causes skew.