Hi,

This is my first post, I hope this is the correct topic for the post.

I'm doing stress tests on a microservices architecture deployed in a aws environment and I have found some throughput contention problem.

I'm using blazemeter as performance testing tool, which have integration with aws cloudwatch metrics. My metrics for aws resources are

• Amazon Elastic Block Store
• Elastic Load Balancing
• Amazon VPC
• Amazon Elastic Compute Cloud

This is the call scheme of my test

Blazemeter -> AWS LB -> ingress-nginx -> SpringBoot service

The scenario that I'm testing is as follow

Number of users: 399.

Number of endpoint/request acceded: 1

Think Time: 4 seg

Rampup: 20 min.

Hold Load: 5 min.

With this scenario, the maximum expected throughput should be 99,75 hits / seconds, but I'm only get about 65 hits / seconds.
The average response time in the 90 percentil for this test is about 93ms and the errors rate is 0%.

Looking at the springboot service resources usage, I see that it has many idle threads that can resolve the load.
It doesn't seem that there are no problems in the ingress-nginx resources usage neither.

Both springboot service and ingress nginx don't show high consumptions in system resources.

What metrics could I look at on cloudwatch to rule out or confirm that the bottleneck is not in any of the aws resource that I use. .

I can share with you additional information if needed.

Thank in advance.