A simple way is to obtain the proportions of various interface calls according to the proportions through online data analysis.
When actually generating pressure, according to the proportion, set the rate to randomly call and distribute the pressure.
The so-called Tencent robot is actually a pressure tool, similar to nss_bench???
pressure test:
Of course, the heterogeneous situation of cluster machines should be considered, and benchmark models should be selected for testing.
General indicators:
CPU should not exceed 70%
Response time should not exceed 1s
Success rate is greater than 99.9%
Stability test:
The pressure generating capacity is medium capacity, the running time is more than 10 hours, the server has no memory leaks, no process restarts, no errors, and the success rate of various request interfaces or transactions is above 99.9%
How to automatically locate performance bottlenecks:
Recommended performance solutions:
Make some of the company's internal solution experience into an answer database and recommend it to the corresponding personnel