I found a similar discussion titled "How to scale up the dify deployment," which is currently unsolved ^[1].

Regarding your question about horizontal scaling for Dify components (API, Worker, Sandbox), the Dify codebase does include configurations and tools that support horizontal scaling:

1. Configuration Files: The pm2.json file specifies a clustered mode with two instances, indicating support for horizontal scaling ^[2].

2. Environment Variables: There are specific environment variables for optimizing horizontal scaling:

   • SERVER_WORKER_AMOUNT and CELERY_WORKER_AMOUNT for setting the number of workers.
   • CELERY_AUTO_SCALE for enabling autoscaling of Celery workers ^[3].

3. Load Balancing: The documentation mentions a "Round-robin" load balancing strategy by default ^[4].

4. Container Orchestration: Dify supports deployment on Kubernetes using Helm Charts or YAML files, which facilitates horizontal scaling ^[5].

5. Distributed Systems: The use of PostgreSQL and Redis for distributed data storage and caching supports horizontal scaling ^[6].

These elements indicate that Dify does support horizontal scaling across its components, and you can leverage these configurations and tools to achieve it.

To continue talking to Dosu, mention @dosu.

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}