I have a few questions about server-cost estimations.
- How do you decide what type of instance is required for X number of concurrent users? Is it totally based on experience or is there a certain rule that you follow for the same?
- I was using JMeter for load testing, and I was wondering, how do you test POST APIs with separate data for each user? Or is there any other platform that you use?
- In the case of POST API calls, do we need to create a separate DB for load testing (which I think, we should)? If yes, should we create a test DB in the same DB instance (i.e., in the same AWS RDS)? And does it needs to have some data present in it? As that might change its performance, right?
- How to load test a workflow? Suppose we need to load test a case where we want 5,000 users to hit Auth API. It will consist of two APIs, one to request an OTP and the other to use that OTP to get the token.
Please help me out, on this. As I am quite new to scaling and was just wondering if someone with experience in this can help. Thanks.
CodePudding user response：
It doesn't look like a single "question" to me going forwards you might want to split it into 4 different ones.
- Just measure it, I don't think it's possible to predict the resources usage, start load test with 1 virtual user and gradually increase the load to the anticipated number of users at the same time looking at resources consumption in AWS CloudWatch or other monitoring solution like JMeter PerfMon Plugin. In case if you detect that CPU or RAM is the bottleneck switch to higher instance and repeat the test.
- There are multiple ways of doing parameterization in JMeter tests, the most commonly used approach is CSV Data Set Config so each user would read the next line from the CSV file containing the test data on each iteration
- DB should live on a separate host as if you place it under the same machine as the application server they will be mutually interfering each other and you might face race conditions. With regards to the database size - if possible make a clone of production data
- You should simulate real usage of the application with 100% accuracy, if user needs to authorize before making an API call your load test script should do the same.