Rollup

Overview

Rollup pre-aggregations summarize raw data into aggregated tables, grouped by specified dimensions. They're the most effective way to improve query performance.

Example

pre_aggregations:
  - name: orders_daily
    type: rollup
    measures:
      - count
      - total_revenue
    dimensions:
      - status
      - category
    time_dimension: created_at
    granularity: day
    partition_granularity: month

Time-Based Rollups

Basic Time Rollup

- name: daily_orders
  measures:
    - count
  time_dimension: created_at
  granularity: day

Partitioned Rollup

Partition large datasets for efficient refreshes:

- name: orders_monthly_partitioned
  measures:
    - count
    - total_revenue
  time_dimension: created_at
  granularity: day
  partition_granularity: month

Granularity Options

Granularity	Description
`second`	Per-second aggregation
`minute`	Per-minute aggregation
`hour`	Hourly aggregation
`day`	Daily aggregation
`week`	Weekly aggregation
`month`	Monthly aggregation
`quarter`	Quarterly aggregation
`year`	Yearly aggregation

Refresh Strategies

Time-Based Refresh

- name: orders_hourly
  measures:
    - count
  time_dimension: created_at
  granularity: hour
  refresh_key:
    every: 1 hour

Incremental Refresh

Only refresh recent partitions:

- name: orders_incremental
  measures:
    - count
    - total_revenue
  time_dimension: created_at
  granularity: day
  partition_granularity: month
  refresh_key:
    every: 1 day
    incremental: true
    update_window: 7 day

SQL-Based Refresh

Refresh when data changes:

- name: orders_on_change
  measures:
    - count
  refresh_key:
    sql: SELECT MAX(updated_at) FROM orders

Indexes

Add indexes for better query performance:

- name: orders_by_status
  measures:
    - count
  dimensions:
    - status
    - category
  indexes:
    - name: status_idx
      columns:
        - status

Query Matching

A query uses a rollup if:

All requested measures are in the rollup
All requested dimensions are in the rollup
Time dimension granularity is compatible
Filters match the rollup's data

# This rollup...
- name: orders_daily
  measures:
    - count
    - total_revenue
  dimensions:
    - status
  time_dimension: created_at
  granularity: day

# ...matches queries like:
# - orders.count by day
# - orders.total_revenue by status by week (day rolls up to week)
# - orders.count where status = 'completed' by month

Best Practices

Match your queries — design rollups around common access patterns
Use incremental refresh — for large time-series data
Partition wisely — monthly partitions work well for most cases
Add indexes — for high-cardinality dimension filters