Documentation
Caching

Caching Overview

Request vs Cube caching layers

Cube provides a two-level caching system. The first level is in-memory cache and is active by default.

Cube's in-memory cache acts as a buffer for your database when there's a burst of requests hitting the same data from multiple concurrent users. Pre-aggregations are designed to provide the right balance between time to insight and querying performance.

To reset the in-memory cache in development mode, just restart the server.

The second level of caching is called pre-aggregations, and requires explicit configuration to activate.

We do not recommend changing the default in-memory caching configuration unless it is necessary. To speed up query performance, consider using pre-aggregations.

Pre-aggregations

Pre-aggregations is a layer of the aggregated data built and refreshed by Cube. It can dramatically improve the query performance and provide a higher concurrency.

To start building pre-aggregations, depending on your data source, Cube may require write access to the pre-aggregations schema in the source database. In this case, Cube first builds pre-aggregations as tables in the source database and then exports them into the pre-aggregations storage. Please refer to the documentation for your specific driver to learn more about read-only support and pre-aggregation build strategies.

Pre-aggregations are defined in the data model. You can learn more about defining pre-aggregations in data modeling reference.

YAML
JavaScript
cubes:
  - name: orders
    sql_table: orders
 
    measures:
      - name: total_amount
        sql: amount
        type: sum
 
    dimensions:
      - name: created_at
        sql: created_at
        type: time
 
    pre_aggregations:
      - name: amount_by_created
        measures:
          - total_amount
        time_dimension: created_at
        granularity: month

In-memory cache

Cube caches the results of executed queries using in-memory cache. The cache key is a generated SQL statement with any existing query-dependent pre-aggregations.

Upon receiving an incoming request, Cube first checks the cache using this key. If nothing is found in the cache, the query is executed in the database and the result set is returned as well as updating the cache.

If an existing value is present in the cache and the refresh_key value for the query hasn't changed, the cached value will be returned. Otherwise, an SQL query will be executed against either the pre-aggregations storage or the source database to populate the cache with the results and return them.

Refresh Keys

Cube takes great care to prevent unnecessary queries from hitting your database. The first stage caching system caches query results, but Cube needs a way to know if the data powering that query result has changed. If the underlying data isn't any different, the cached result is valid and can be returned skipping an expensive query, but if there is a difference, the query needs to be re-run and its result cached.

To aid with this, Cube defines a refresh_key for each cube. Refresh keys are evaluated by Cube to assess if the data needs to be refreshed.

The following refresh_key tells Cube to refresh data every 5 minutes:

YAML
JavaScript
cubes:
  - name: orders
    # ...
 
    refresh_key:
      every: 5 minute

With the following refresh_key, Cube will only refresh the data if the value of MAX(created_at) changes. By default, Cube will check this refresh_key every 10 seconds:

YAML
JavaScript
cubes:
  - name: orders
    # ...
 
    refresh_key:
      sql: SELECT MAX(created_at) FROM orders

By default, Cube will check and invalidate the cache in the background when in development mode. In production environments, we recommend running a Refresh Worker as a separate instance.

We recommend enabling background cache invalidation in a separate Cube worker for production deployments. Please consult the Production Checklist for more information.

If background refresh is disabled, Cube will refresh the cache during query execution. Since this could lead to delays in responding to end-users, we recommend always enabling background refresh.

Default Refresh Keys

The default values for refresh_key are

  • every: 2 minute for BigQuery, Athena, Snowflake, and Presto.
  • every: 10 second for all other databases.

You can use a custom SQL query to check if a refresh is required by changing the refresh_key property in a cube. Often, a MAX(updated_at_timestamp) for OLTP data is a viable option, or examining a metadata table for whatever system is managing the data to see when it last ran.

Disabling the cache

There's no straightforward way to disable caching in Cube. The reason is that Cube not only stores cached values but also uses the cache as a point of synchronization and coordination between nodes in a cluster. For the sake of design simplicity, Cube doesn't distinguish client invocations, and all calls to the data load API are idempotent. This provides excellent reliability and scalability but has some drawbacks. One of those load data calls can't be traced to specific clients, and as a consequence, there's no guaranteed way for a client to initiate a new data loading query or know if the current invocation wasn't initiated earlier by another client. Only Refresh Key freshness guarantees are provided in this case.

For situations like real-time analytics or responding to live user changes to underlying data, the refresh_key query cache can prevent fresh data from showing up immediately. For these situations, the cache can effectively be disabled by setting the refresh_key.every parameter to something very low, like 1 second.

Inspecting Queries

To inspect whether the query hits in-memory cache, pre-aggregation, or the underlying data source, you can use the Playground or Cube Cloud (opens in a new tab).

Developer Playground can be used to inspect a single query. To do that, click the "cache" button after executing the query. It will show you the information about the refresh_key for the query and whether the query uses any pre-aggregations. To inspect multiple queries or list existing pre-aggregations, you can use Cube Cloud (opens in a new tab).

To inspect queries in the Cube Cloud, navigate to the "History" page. You can filter queries by multiple parameters on this page, including whether they hit the cache, pre-aggregations, or raw data. Additionally, you can click on the query to see its details, such as time spent in the database, the database queue's size at the point of query execution, generated SQL, query timeline, and more. It will also show you the optimal pre-aggregations that could be used for this query.

To see existing pre-aggregations, navigate to the "Pre-Aggregations" page in the Cube Cloud. The table shows all the pre-aggregations, the last refresh timestamp, and the time spent to build the pre-aggregation. You can also inspect every pre-aggregation's details: the list of queries it serves and all its versions.

Cache type

Any query that is fulfilled by Cube will use one of the following cache types:

  • Pre-aggregations in Cube Store. This is the most advantageous and performant option.
  • Pre-aggregations in Cube Store with a suboptimal query plan. This cache type indicates that queries still benefit from pre-aggregations in Cube Store but it's possible to get a performance boost by using indexes.
  • Pre-aggregations in the data source. This cache type indicates that queries don't benefit from pre-aggregations in Cube Store and it's possible to get a massive performance boost by using Cube Store as pre-aggregation storage.
  • In-memory cache. This cache type indicates that queries don't benefit from pre-aggregations at all. Queries directly hit the upstream data source and in-memory cache is used to speed up the execution of identical queries that arrive within a short period of time.
  • No cache. This cache type indicates queries that directly hit the upstream data source and have the worst performance possible.

In Query History and throughout Cube Cloud, colored bolt icons are used to indicate the cache type. Also, Performance Insights provide an overview of API requests by specific cache types.