Introduction

An Amazon Aurora Serverless v2 cluster with a capacity of 2 Aurora Capacity Units (ACUs) can handle a significant number of concurrent users, though the exact number depends on various factors, including the workload's characteristics and the database engine (e.g., MySQL or PostgreSQL).

Key considerations:

• Connections per ACU:

  While there isn't a fixed, universal number of concurrent users per ACU, AWS documentation and community discussions suggest that a 1 ACU instance can handle around 90 connections, and a 64 ACU instance can handle up to 5,000 connections. Extrapolating this, a 2 ACU instance would likely support a few hundred connections, potentially more depending on the efficiency of those connections.

• Workload Type:

  The nature of the workload (read-heavy vs. write-heavy, complexity of queries, transaction size) heavily influences the actual concurrent user capacity. Simple, fast queries allow for more concurrent connections than complex, long-running transactions.

• max_connections Parameter:

  Aurora Serverless v2's max_connections parameter automatically adjusts based on ACU capacity. While it can go up to 5,000 for higher ACUs, for 2 ACUs, the limit will be lower and determined by the system based on available resources.

• Read Replicas:

  For read-heavy workloads, adding read replicas (up to 15) can significantly increase the total concurrent user capacity by distributing read traffic across multiple instances.

• Data API:

  Using the Data API can also help manage connections, especially for applications that don't require persistent connections.

Conclusion

In summary: A 2 ACU Aurora Serverless v2 cluster offers a good balance of capacity and cost-efficiency for many workloads. While it won't support thousands of concurrent users like a fully provisioned large instance or a cluster with many read replicas, it can comfortably handle hundreds of concurrent users for typical web applications and services. For precise estimations, performance testing with your specific workload is recommended.