Overview

This section contains a general overview of deploying a Cube cluster in production. You can also check platform-specific guides for Cube Cloud and Docker.

If you are moving Cube to production, check out the Production Checklist.

Components

As shown in the diagram below, a typical production deployment of Cube includes the following components:

One or multiple API instances
A Refresh Worker
A Cube Store cluster

API Instances process incoming API requests and query either Cube Store for pre-aggregated data or connected database(s) for raw data. The Refresh Worker builds and refreshes pre-aggregations in the background. Cube Store ingests pre-aggregations built by Refresh Worker and responds to queries from API instances.

API instances and Refresh Workers can be configured via environment variables or the cube.js configuration file. They also need access to the data model files. Cube Store clusters can be configured via environment variables.

You can find an example Docker Compose configuration for a Cube deployment in the platform-specific guide for Docker.

API instances

API instances process incoming API requests and query either Cube Store for pre-aggregated data or connected data sources for raw data. It is possible to horizontally scale API instances and use a load balancer to balance incoming requests between multiple API instances.

The Cube Docker image (opens in a new tab) is used for API Instance.

API instances can be configured via environment variables or the cube.js configuration file, and must have access to the data model files (as specified by schemaPath.

Refresh Worker

A Refresh Worker updates pre-aggregations and invalidates the in-memory cache in the background. They also keep the refresh keys up-to-date for all data models and pre-aggregations. Please note that the in-memory cache is just invalidated but not populated by Refresh Worker. In-memory cache is populated lazily during querying. On the other hand, pre-aggregations are eagerly populated and kept up-to-date by Refresh Worker.

The Cube Docker image (opens in a new tab) can be used for creating Refresh Workers; to make the service act as a Refresh Worker, CUBEJS_REFRESH_WORKER=true should be set in the environment variables.

Cube Store

Cube Store is the purpose-built pre-aggregations storage for Cube.

Cube Store uses a distributed query engine architecture. In every Cube Store cluster:

a one or many router nodes handle incoming connections, manages database metadata, builds query plans, and orchestrates their execution
multiple worker nodes ingest warmed up data and execute queries in parallel
a local or cloud-based blob storage keeps pre-aggregated data in columnar format

Cube Store Router with two Cube Store Workers

By default, Cube Store listens on the port 3030 for queries coming from Cube. The port could be changed by setting CUBESTORE_HTTP_PORT environment variable. In a case of using custom port, please make sure to change CUBEJS_CUBESTORE_PORT environment variable for Cube API Instances and Refresh Worker.

Both the router and worker use the Cube Store Docker image (opens in a new tab). The following environment variables should be used to manage the roles:

Environment Variable	Specify on Router?	Specify on Worker?
`CUBESTORE_SERVER_NAME`	✅ Yes	✅ Yes
`CUBESTORE_META_PORT`	✅ Yes	—
`CUBESTORE_WORKERS`	✅ Yes	✅ Yes
`CUBESTORE_WORKER_PORT`	—	✅ Yes
`CUBESTORE_META_ADDR`	—	✅ Yes

Looking for a deeper dive on Cube Store architecture? Check out this presentation (opens in a new tab) by our CTO, Pavel (opens in a new tab).

Cube Store Router

The Router in a Cube Store cluster is responsible for receiving queries from Cube, managing metadata for the Cube Store cluster, and query planning and distribution for the Workers. It also provides a MySQL-compatible interface that can be used to query pre-aggregations from Cube Store directly. Cube only communicates with the Router, and does not interact with Workers directly.

Cube Store Worker

Workers in a Cube Store cluster receive and execute subqueries from the Router, and directly interact with the underlying distributed storage for insertions, selections and pre-aggregation warmup. Workers do not interact with each other directly, and instead rely on the Router to distribute queries and manage any associated metadata.

Scaling

Although Cube Store can be run in single-instance mode, this is often unsuitable for production deployments. For high concurrency and data throughput, we strongly recommend running Cube Store as a cluster of multiple instances instead. Because the storage layer is decoupled from the query processing engine, you can horizontally scale your Cube Store cluster for as much concurrency as you require.

A sample Docker Compose stack setting Cube Store cluster up might look like:

services:
  cubestore_router:
    image: cubejs/cubestore:latest
    environment:
      - CUBESTORE_WORKERS=cubestore_worker_1:10001,cubestore_worker_2:10002
      - CUBESTORE_REMOTE_DIR=/cube/data
      - CUBESTORE_META_PORT=9999
      - CUBESTORE_SERVER_NAME=cubestore_router:9999
    volumes:
      - .cubestore:/cube/data
    depends_on:
      - cubestore_worker_1
      - cubestore_worker_2
 
  cubestore_worker_1:
    image: cubejs/cubestore:latest
    environment:
      - CUBESTORE_WORKERS=cubestore_worker_1:10001,cubestore_worker_2:10002
      - CUBESTORE_SERVER_NAME=cubestore_worker_1:10001
      - CUBESTORE_WORKER_PORT=10001
      - CUBESTORE_REMOTE_DIR=/cube/data
      - CUBESTORE_META_ADDR=cubestore_router:9999
    volumes:
      - .cubestore:/cube/data
 
  cubestore_worker_2:
    image: cubejs/cubestore:latest
    environment:
      - CUBESTORE_WORKERS=cubestore_worker_1:10001,cubestore_worker_2:10002
      - CUBESTORE_SERVER_NAME=cubestore_worker_2:10002
      - CUBESTORE_WORKER_PORT=10002
      - CUBESTORE_REMOTE_DIR=/cube/data
      - CUBESTORE_META_ADDR=cubestore_router:9999
    volumes:
      - .cubestore:/cube/data

Storage

Cube Store makes use of a separate storage layer for storing metadata as well as for persisting pre-aggregations as Parquet files. Cube Store can use both AWS S3 and Google Cloud, or if desired, a local path on the server if all nodes of a cluster run on a single machine.

A simplified example using AWS S3 might look like:

services:
  cubestore_router:
    image: cubejs/cubestore:latest
    environment:
      - CUBESTORE_SERVER_NAME=cubestore_router:9999
      - CUBESTORE_META_PORT=9999
      - CUBESTORE_WORKERS=cubestore_worker_1:9001
      - CUBESTORE_S3_BUCKET=<BUCKET_NAME_IN_S3>
      - CUBESTORE_S3_REGION=<BUCKET_REGION_IN_S3>
      - CUBESTORE_AWS_ACCESS_KEY_ID=<AWS_ACCESS_KEY_ID>
      - CUBESTORE_AWS_SECRET_ACCESS_KEY=<AWS_SECRET_ACCESS_KEY>
  cubestore_worker_1:
    image: cubejs/cubestore:latest
    environment:
      - CUBESTORE_SERVER_NAME=cubestore_worker_1:9001
      - CUBESTORE_WORKER_PORT=9001
      - CUBESTORE_META_ADDR=cubestore_router:9999
      - CUBESTORE_WORKERS=cubestore_worker_1:9001
      - CUBESTORE_S3_BUCKET=<BUCKET_NAME_IN_S3>
      - CUBESTORE_S3_REGION=<BUCKET_REGION_IN_S3>
      - CUBESTORE_AWS_ACCESS_KEY_ID=<AWS_ACCESS_KEY_ID>
      - CUBESTORE_AWS_SECRET_ACCESS_KEY=<AWS_SECRET_ACCESS_KEY>
    depends_on:
      - cubestore_router

Semantic Catalog Production checklist