AWS RDS Aurora V2 setup for low traffic but high availability


We are operating an AWS RDS Aurora V2 (PostgreSQL) setup for a rather low to zero traffic which has to be high available. Currently, we have an Aurora V2 serverless setup with a cluster consisting of a reader and a writer with a configuration of 0.5 to 4 ACUs each. To our surprise, the database causes the highest cost in the cluster. Therefore, we are looking for possibilities to optimize the database for the use case in order to optimize the cost.

One alternative seems to be to replace serverless with the smallest ec2 machine, which seems to be a db.t4g.medium on-demand ec2 instance. However as a result of our investigation and as a db.t4g.medium seems to be the smallest ec2 machine, it looks like that switching to an on-demand ec2 instance will not help. Hence, we are looking for general guidance whether our setup with Aurora V2 serverless with an acu range from 0.5 to 4 can be optimized for our low to zero traffic but highly available setup or whether we already have the best setup. Thank you!

已提問 6 個月前檢視次數 278 次
1 個回答

Hello Customer,

I understand that an Aurora serverless v2 is configured to handle a minimal load. However, this has generated a significant cost. Therefore, you are looking at optimizing the instance and reducing the cost. Below are few items that can be considered.

  1. Please review if the serverless instance is utilizing the max vCPU that is allocated.
  • If there is continuous high usage of ACU, then kindly enable performance insight and review the queries that is generating the load. Further, work with the in-house DBA to fine-tune and reduce the resource consumption.

  • If is intermittent high ACU usage, then identify the timestamp by reviewing the CloudWatch metrics (ServerlessDatabaseCapacity and ACUUtilization) and further utilize performance insight to fine-tune queries.

  • If the instance is not utilizing the maxvCPU allocated then kindly consider reducing the maxvCPU. This would help in reducing abrupt usage of vCPU by rogue queries.

  1. Further, moving to provisioned instances has been already analyzed at your end and it is concluded to be not useful. Hence, you would have to look into fine-tuning the resource usage on serverless cluster.

Based on the above inputs, the only way to reduce the ACU utilization is to reduce the load on the instance. Please note, a single bad query can cause resource saturation and spike the ACU being used. Hence, the main path would be to review the performance insight data and fine-tune the heavy load generating queries.

已回答 6 個月前

您尚未登入。 登入 去張貼答案。