Performance Test JMeter The jmeter of the pit of the Bug:tps periodic fluctuation problem

first , the question:

When I was doing performance testing, I used jmeter to make 100 concurrent, with 100TPS pressure to test 10 minutes, but the pressure has been fluctuating, and fluctuations in the JMeter very card, such as:

Periodic TPS fluctuations

Various speculations:

So began to look for various reasons for the environment, at first thought is JMeter connection was "hijacked", or jmeter will not card. So it took a whole afternoon to get rid of the pressure-measuring machine environment, the pressure-tested environment (number of TCP connections, program problems, etc.), but never found the cause. Later, after replacing LR, the pressure measurement is normal. So starting to suspect is JMeter's own problem.

reasons to find :

Then I thought, the script that I was testing was the old script, also on the JMeter2.8, and then I'm using the JMeter is 2.13, is it a scripting compatibility issue?

Problem Solving :

All kinds of substitutions, and finally to the problem of CSV Data set config, as long as I use the 2.8 script CSV Data set config to parameterize, even if this parameter I do not use, a pressure measurement will appear TPS fluctuations; After I disable, I create a new CSV Data Set Config, all data remain unchanged, re-test, ok! This hole is so big! I hope Apache organization can modify, it is definitely the old version of CSV Data Set config in the new version of the JMeter, scheduling problems, causing local pressure instability, and the key is that this led to jmeter too card. The pressure after the problem is resolved as follows: Steady, I want to be stable happiness ~

Very stable, fluctuation in the 3TPS range

There are other pits :

PS: In addition, give everyone to add a pit, is jmeter plug-in resource monitoring problem, the agent put on Linux to monitor resources, CPU and memory are not any problem, but if you monitor TCP connection, you should pay attention to, this monitoring can eat about 15% CPU resources , 4-core single-board Oh, and the SYS CPU is much higher. It should be that this monitoring algorithm is not optimized enough to occupy too much resources. Please jmeter users to pay attention to!

By:Fresh.Qin

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

Performance Test JMeter The jmeter of the pit of the Bug:tps periodic fluctuation problem