Pre-aggregations

Pre-aggregations are materialized query results persisted as tables. Cube has an ability to analyze queries against a defined set of pre-aggregation rules in order to choose the optimal one that will be used to create pre-aggregation table.

If Cube finds a suitable pre-aggregation rule, database querying becomes a multi-stage process:

1. Cube checks if an up-to-date copy of the pre-aggregation exists.

2. Cube will execute a query against the pre-aggregated tables instead of the raw data.

Pre-aggregations can be defined in the pre_aggregations available on each cube. A more in-depth diagram detailing how pre-aggregations are selected can be found here.

Pre-aggregations must have, at minimum, a name and a type. Pre-aggregations must include all dimensions, measures, and filters you will query with.

Parameters

name

The name parameter serves as the identifier of a pre-aggregation. It must be unique among all pre-aggregations within a cube and follow the naming conventions.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: orders_by_status
        dimensions:
          - CUBE.status
        measures:
          - CUBE.count

This name, along with the name of the cube, will be used as a prefix for pre-aggregation tables created in the database.

type

Cube supports the following types of pre-aggregations:

• rollup
• original_sql
• rollup_join
• rollup_lambda

The default type is rollup.

rollup

Rollup pre-aggregations are the most effective way to boost performance of any analytical application. The blazing fast performance of tools like Google Analytics or Mixpanel are backed by a similar concept. The theory behind it lies in multi-dimensional analysis, and a rollup pre-aggregation is the result of a roll-up operation on an OLAP cube (opens in a new tab). A rollup pre-aggregation is essentially the summarized data of the original cube grouped by any selected dimensions of interest.

The most performant kind of rollup pre-aggregation is an additive rollup: all measures of which are based on decomposable aggregate functions (opens in a new tab). Additive measure types are: count, sum, min, max or count_distinct_approx. The performance boost in this case is based on two main properties of additive rollup pre-aggregations:

1. A rollup pre-aggregation table usually contains many fewer rows than its' corresponding original fact table. The fewer dimensions that are selected for roll-up means fewer rows in the materialized result. A smaller number of rows therefore means less time to query rollup pre-aggregation tables.

2. If your query is a subset of dimensions and measures of an additive rollup, then it can be used to calculate a query without accessing the raw data. The more dimensions and measures are selected for roll-up, the more queries can use this particular rollup.

Rollup definitions can contain members from a single cube as well as from multiple cubes. In case of multiple cubes being involved, the join query will be built according to the standard rules of cubes joining.

Rollups are selected for querying based on properties found in queries made to the Cube REST API. A thorough explanation can be found under Getting Started with Pre-Aggregations.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: orders_by_company
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status

original_sql

As the name suggests, it persists the results of the sql property of the cube. Pre-aggregations of type original_sql should only be used when the cube's sql is a complex query (i.e. nested, window functions and/or multiple joins).

They often do not provide much in the way of performance directly, but there are two specific applications:

1. They can be used in tandem with the use_original_sql_pre_aggregations option in other rollup pre-aggregations.

2. Situations where it is not possible to use a rollup pre-aggregations, such as funnels.

For example, to pre-aggregate all completed orders, you could do the following:

cubes:
  - name: completed_orders
    sql: >
      SELECT * FROM orders WHERE completed = true
 
    pre_aggregations:
      - name: main
        type: original_sql

rollup_join

Cube is capable of performing joins between pre-aggregations from different data sources to avoid making excessive queries to them.

In the following example, we have a users cube with a users_rollup pre-aggregation, and an orders cube with an orders_rollup pre-aggregation, and an orders_with_users_rollup pre-aggregation. Note the following:

• Both cubes have different values for data_source.
• The type of orders_with_users_rollup is rollup_join.
• This pre-aggregation has a special property rollups which is an array containing references to both "source" rollups.

cubes:
  - name: users
    data_source: postgres
    sql_table: users
 
    pre_aggregations:
      - name: users_rollup
        dimensions:
          - CUBE.id
          - CUBE.name
 
    measures:
      - name: count
        type: count
 
    dimensions:
      - name: id
        sql: id
        type: number
        primary_key: true
 
      - name: name
        sql: "{CUBE}.first_name || {CUBE}.last_name"
        type: string
 
  - name: orders
    data_source: mssql
    sql_table: orders
 
    pre_aggregations:
      - name: orders_rollup
        measures:
          - CUBE.count
        dimensions:
          - CUBE.user_id
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: day
 
      - name: orders_with_users_rollup
        type: rollup_join
        measures:
          - CUBE.count
        dimensions:
          - users.name
        rollups:
          - users.users_rollup
          - CUBE.orders_rollup
 
    joins:
      - name: users
        relationship: many_to_one
        sql: "{CUBE.user_id} = {users.id}"
 
    measures:
      - name: count
        type: count
 
    dimensions:
      - name: id
        sql: id
        type: number
        primary_key: true
 
      - name: user_id
        sql: user_id
        type: number
 
      - name: status
        sql: status
        type: string
 
      - name: created_at
        sql: created_at
        type: time

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: orders_rollup
        measures:
          - CUBE.count
        dimensions:
          - users.name
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: day

rollup_lambda

A rollup_lambda pre-aggregation is a special type of pre-aggregation that can combine data from data sources and other rollups. It is extremely useful in scenarios where real-time data is required.

Lambda pre-aggregations can be used to combine data from a data source and a pre-aggregation, or even from multiple pre-aggregations across different cubes that share the same dimensions and measures.

measures

The measures property is an array of measures from the cube that should be included in the pre-aggregation:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: users_rollup
        measures:
          - CUBE.count
 
    measures:
      - name: count
        type: count

dimensions

The dimensions property is an array of dimensions from the cube that should be included in the pre-aggregation:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: users_rollup
        dimensions:
          - CUBE.status
 
    dimensions:
      - name: status
        type: string
        sql: status

time_dimension

The time_dimension property can be any dimension of type time. All other measures and dimensions in the data model are aggregated. This property is an extremely useful tool for improving performance with massive datasets.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: orders_by_status
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: day
 
    measures:
      - name: count
        type: count
 
    dimensions:
      - name: status
        type: string
        sql: status
 
      - name: created_at
        type: time
        sql: created_at

A granularity must also be included in the pre-aggregation definition.

granularity

The granularity property defines the granularity of data within the pre-aggregation. If set to week, for example, then Cube will pre-aggregate the data by week and persist it to Cube Store.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: users_rollup_by_week
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: week

The value can be one of second, minute, hour, day, week, month, quarter or year. This property is required when using time_dimension.

segments

The segments property is an array of segments from the cube that can target the pre-aggregation:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        segments:
          - CUBE.only_complete
 
    measures:
      - name: count
        type: count
 
    segments:
      - name: only_complete
        sql: "{CUBE}.status = 'completed'"

partition_granularity

The partition_granularity defines the granularity for each partition of the pre-aggregation:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: users_rollup
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: month
 
    # ...

The value can be one of hour, day, week, month, quarter, year. A time_dimension and granularity must also be included in the pre-aggregation definition. This property is required when using partitioned pre-aggregations.

Number of partitions to be built per cube is calculated as build_range divided by partition_granularity. Number of partitions to be built per cube is multiplied by the count of time zones and tenants in case different tenants have different pre-aggregation SQL.

refresh_key

Cube can also take care of keeping pre-aggregations up to date with the refresh_key property. By default, it is set to every: '1 hour', if neither of the cubes' pre-aggregation references don't override refresh_key.

sql

You can set up a custom refresh check strategy by using the sql property:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        refresh_key:
          sql: SELECT MAX(created_at) FROM orders
 
    # ...

In the above example, the refresh key SQL will be executed every 10 seconds, as every is not defined. If the results of the SQL refresh key differ from the last execution, then the pre-aggregation will be refreshed.

every

The refresh_key can define an every property which can be used to refresh pre-aggregations based on a time interval. By default, it is set to 1 hour unless the sql property is also defined in any of cubes pre-aggregation references, in which case it is set to 10 seconds. For example:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        refresh_key:
          every: 1 day
 
    # ...

For possible every parameter values please refer to refreshKey documentation.

You can also use every with sql:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        refresh_key:
          every: 1 hour
          sql: SELECT MAX(created_at) FROM orders

In the above example, the refresh key SQL will be executed every hour. If the results of the SQL refresh key differ from the last execution, then the pre-aggregation will be refreshed.

incremental

You can incrementally refresh partitioned rollups by setting incremental: true. This option defaults to false.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: day
        refresh_key:
          every: 1 day
          incremental: true
 
    # ...

update_window

The incremental: true flag generates a special refresh_key SQL query which triggers a refresh for partitions where the end date lies within the update_window from the current time.

In the example below, it will refresh today's and the last 7 days of partitions once a day. Partitions before the 7 day interval will not be refreshed once they are built unless the rollup SQL is changed.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: day
        refresh_key:
          every: 1 day
          incremental: true
          update_window: 7 day
 
    # ...

This property is required when using incremental refreshes.

allow_non_strict_date_range_match

The allow_non_strict_date_range_match property can be used to allow queries to match this pre-aggregation when using non-strict date ranges. This is useful when using a visualization tools such as Apache Superset that use loose date ranges by default.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measure:
          - CUBE.count
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: day
        allow_non_strict_date_range_match: true

use_original_sql_pre_aggregations

Cube supports multi-stage pre-aggregations by reusing original SQL pre-aggregations in rollups through the use_original_sql_pre_aggregations property. It is helpful in cases where you want to re-use a heavy SQL query calculation in multiple rollup pre-aggregations. Without use_original_sql_pre_aggregations enabled, Cube will always re-execute all underlying SQL calculations every time it builds new rollup tables.

cubes:
  - name: orders
    sql: >
      SELECT * FROM orders1 UNION ALL SELECT * FROM orders2 UNION ALL SELECT *
      FROM orders3
 
    pre_aggregations:
      - name: main
        type: original_sql
 
      - name: statuses
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status
        use_original_sql_pre_aggregations: true
 
      - name: completed_orders
        measures:
          - CUBE.count
        time_dimension: CUBE.completed_at
        granularity: day
        use_original_sql_pre_aggregations: true
 
    # ...

scheduled_refresh

To always keep pre-aggregations up-to-date, you can set scheduled_refresh: true. This option defaults to true. If set to false, pre-aggregations will always be built on-demand. The refresh_key is used to determine if there's a need to update specific pre-aggregations on each scheduled refresh run. For partitioned pre-aggregations, min and max dates for time_dimension are checked to determine range for the refresh.

Each time a scheduled refresh is run, it takes every pre-aggregation partition starting with most recent ones in time and checks if the refresh_key has changed. If a change was detected, then that partition will be refreshed.

In development mode, Cube runs the background refresh by default and will refresh all pre-aggregations which have scheduled_refresh: true.

Please consult Production Checklist for best practices on running background refresh in production environments.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: orders_by_status
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: month
 
    # ...

build_range_start and build_range_end

The build range defines what partitions should be built by a scheduled refresh. By default, the build range is defined as the minimum and maximum values possible for the time_dimension used in the rollup.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: month
        build_range_start:
          sql: SELECT NOW() - INTERVAL '300 day'
        build_range_end:
          sql: SELECT NOW()
 
    # ...

It can be used together with the pre-aggregation's refresh_key to define granular update settings. Set refresh_key.update_window to the interval in which your data can change and build_range_start to the earliest point of time when history should be available.

In the following example, refresh_key.update_window is 1 week and build_range_start is SELECT NOW() - INTERVAL '365 day', so the scheduled refresh will build historical partitions for 365 days in the past and will only refresh last week's data.

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        measures:
          - CUBE.count
        time_dimension: CUBE.created_at
        granularity: day
        partition_granularity: month
        build_range_start:
          sql: SELECT NOW() - INTERVAL '365 day'
        build_range_end:
          sql: SELECT NOW()
        refresh_key:
          update_window: 1 week

union_with_source_data

This option allows combining a pre-aggregation with fresh data retrieved from the data source; this is extremely useful in scenarios where latency can be sacrificed for accuracy.

To configure a pre-aggregation to behave in this way, ensure the pre-aggregation is of type rollup_lambda, and then set union_with_source_data to true:

cubes:
  - name: orders
    # ...
 
    pre_aggregations:
      - name: lambda
        type: rollup_lambda
        union_with_source_data: true
        rollups:
          - orders.main

indexes

In case of pre-aggregation tables having significant cardinality, you might want to create indexes for them in databases which support it. This can be done as follows:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: category_and_date
        measures:
          - CUBE.count
        dimensions:
          - CUBE.category
        time_dimension: CUBE.created_at
        granularity: day
        indexes:
          - name: category_index
            columns:
              - CUBE.category
 
    # ...

To maximize performance, you can introduce an index per query, so the set of dimensions used in the query is the same as in the index. There's no need to define measure columns in the index. The index column order is the sort order of all data within the index. Under the hood, Cube Store will create table dataset copy and will sort it according to index specification. Cube Store works best when you define all your equals filter dimensions first and all dimensions from GROUP BY last. For range filter columns like contains, >, <, >=, <= with high selectivity, typically less than 5% of records keep those columns before GROUP BY dimensions. Otherwise, for low selectivity range filters, consider putting these columns after GROUP BY dimensions.

For original_sql pre-aggregations, the original column names as strings can be used:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: main
        type: original_sql
        indexes:
          - name: timestamp_index
            columns:
              - timestamp

Aggregating indexes

Cube Store also supports aggregating indexes that contain only specified columns and pre-aggregated measures in the pre-aggregation definition. Queries that can target aggregating indexes must have the following characteristics:

1. They cannot make use of any filters other than for those dimensions that are included in that index
2. All dimensions used in the query must be defined in the aggregating index

Queries that do not have the above can still make use of regular indexes so performance can still be optimized.

Basically, the aggregating index is a rollup of data in a rollup table itself. Data needs to be downloaded from the source DB as many times as many rollups you have. Given that having a single rollup and multiple aggregate indexes gives you pretty much the same performance from the Cube Store side but multiple times less cost from a data warehouse perspective.

Aggregating indexes can be defined by adding type: 'aggregate' to the index definition:

cubes:
  - name: orders
    sql_table: orders
 
    pre_aggregations:
      - name: category_and_date
        measures:
          - CUBE.count
        dimensions:
          - CUBE.status
        time_dimension: CUBE.created_at
        granularity: day
        indexes:
          - name: category_index
            columns:
              - CUBE.status
              - Products.name
          - name: aggregated_index
            columns:
              - CUBE.status
            type: aggregate
 
    # ...