Why did DynamoDB start to throttle instead of autoscale?

Question

Hi team, I had a DynamoDB table which was configured to be on-demand mode. I recently made a change to provisioned, with a range of reads of 200 (min) to 1700 (max).

After the transition, the provisioned RCU was sitting at 200, meanwhile the traffic that came in was 450. Instead of auto-scaling it immediately, DynamoDB started to throttle, resulting our services to start throwing 503 errors. The provisioned read only scaled up correctly 15 minutes later. Why is this happening?

Answer

Hi, Betty. Sorry about the hassles. A lot of what you're up against is lag in publishing of metrics to CloudWatch - DynamoDB auto scaling is dependent on this. DynamoDB's throughput limits apply on a per-second basis, but CloudWatch metrics typically depict per-minute averages. So it's hard to say what is happening in your case exactly. If you can share some more information about the way that you've constructed your DDB table, maybe the forum can suggest some options for better scaling around your use case.

Answer

You might find this blog informative on how throttles happen in provisioned mode and what strategies to use concerning them: https://aws.amazon.com/blogs/database/handle-traffic-spikes-with-amazon-dynamodb-provisioned-capacity/

Why did DynamoDB start to throttle instead of autoscale?

Relevant content