Overview
Z-order sorting interleaves the bits of multiple columns to achieve multi-dimensional data locality. This dramatically improves query performance when filters span multiple columns.
When to Use Z-Order
Z-order is ideal when your queries frequently filter on 2 or more columns simultaneously:
-- Queries that benefit from z-order on (region, date):
SELECT * FROM events WHERE region = 'US' AND date > '2026-01-01';
SELECT * FROM orders WHERE customer_id = 123 AND status = 'pending';
How It Works
- Normalize: Each column value is converted to an 8-byte comparable representation
- Interleave: Bits from all columns are woven together
- Sort: Records are sorted by the interleaved value
- Write: Parquet files preserve the z-order, creating natural clustering
Configuration
In the Managed Lakehouse destination node:
- Set Sort Strategy to “Z-Order”
- Enter columns:
region, date, user_id
Or via pipeline JSON:
{
"managedLakehouseSettings": {
"sortStrategy": "z_order",
"sortColumns": ["region", "date", "user_id"]
}
}
Supported Column Types
| Type | Normalization |
|---|
int, int32, int64 | Sign-bit flip for unsigned comparison |
float32, float64 | IEEE 754 total-order encoding |
timestamp, date | Microseconds since epoch |
string | First 8 bytes for locality |
boolean | 0 or 1 |
Tips
Don’t z-order by columns that are already partition keys — they’re already isolated into separate directories.
- Choose 2–4 columns that are most frequently used together in WHERE clauses
- Timestamp columns are excellent z-order candidates
- Z-order is applied per batch; compaction further improves clustering over time
- For single-column queries, standard sort order is sufficient — z-order helps with multi-column patterns
