DP-750 · 3.1.5 · Prepare and process data
Partitioning strategy
Design and implement a data partitioning scheme
Partition only when scale and access patterns justify it; prefer liquid clustering for most modern Databricks tables.
Partition only when pruning pays off
- 1Query filtersIdentify repeated selective predicates
- 2CardinalityAvoid too many values
- 3Table sizeConfirm the table is large enough
- 4Layout choiceCompare partitioning with clustering
A good partition key lets queries skip large amounts of data. A bad one creates many small directories and files.
Watch cardinality
Good candidate
- Values
- Low to moderate
- Filters
- Common
- Result
- Useful pruning
Poor candidate
- Values
- Very high
- Filters
- Rare
- Result
- Tiny partitions
Do not partition by habit. For most new tables, evaluate liquid clustering before committing to a static partition scheme.
Revisit access patterns
Physical layout should follow observed query behavior, not the business importance of a column.
Key points
- Partitioning: Physically separates files by partition values
- Cardinality: Too many values create many small partitions
- Pruning: Useful when filters align with partition columns
- Modern default: Prefer liquid clustering for most managed tables
Exam trap
Over-partitioning can make performance worse by creating many tiny directories and files.
Sources · 1 Microsoft Learn page Show ▾
Checked on October 4, 2026. Report a problem
Go deeper on Microsoft Learn
Checked against Microsoft Learn on October 4, 2026. Report a problem with this card