Using AI to Generate Massive Data for JMeter Performance Tests

Core Idea

Generate test data dynamically inside JMeter threads to eliminate the I/O bottleneck of external CSV files. AI is used to produce the Groovy logic for these dynamic generators.

Solution Comparison

CSV data file – simple and intuitive for small, static data sets (under ten thousand rows). Large files become I/O bottlenecks and are hard to maintain.

JMeter functions – built‑in and lightweight, suitable for simple random numbers or counters. Cannot create complex structured data.

JSR223 scripts (Groovy) – unlimited data volume, high performance, flexible for complex business rules. Requires writing scripts, which AI can assist with.

Scenario 1 – Unique User Registration Data

Requirement: each request must use a unique username, email, and a random password while supporting high concurrency.

Prompt to AI :

请帮我编写一个JMeter JSR223 Sampler（使用Groovy语言）的脚本，用于生成大量唯一的用户注册数据。要求如下：
目标：为每个JMeter线程的每次请求生成唯一的数据。
生成规则：
username：格式为 "perf_user_" + 一个全局唯一且递增的数字ID。
email：格式为 "test_" + 上述唯一ID + "@example.com"。
password：一个随机长度在8-12位，且包含大小写字母和数字的字符串。
关键点：要确保在高并发下ID的唯一性。请将生成的变量存入JMeter变量（如 vars），以便HTTP请求等后续元件引用。

AI‑generated Groovy script :

// JSR223 Sampler (Groovy) – generate unique user data
import java.util.concurrent.atomic.AtomicLong

if (!globalCounter) {
    globalCounter = new AtomicLong(1L)
}

long uniqueId = globalCounter.getAndIncrement()
String username = "perf_user_" + uniqueId
String email = "test_" + uniqueId + "@example.com"

def random = new Random()
def lower = "abcdefghijklmnopqrstuvwxyz"
def upper = lower.toUpperCase()
def digits = "0123456789"
def allChars = lower + upper + digits

def passwordLength = random.nextInt(5) + 8 // 8‑12 characters
def passwordBuilder = new StringBuilder()
// Ensure at least one of each character type
passwordBuilder.append(lower.charAt(random.nextInt(lower.length())))
passwordBuilder.append(upper.charAt(random.nextInt(upper.length())))
passwordBuilder.append(digits.charAt(random.nextInt(digits.length())))
(4..passwordLength).each {
    passwordBuilder.append(allChars.charAt(random.nextInt(allChars.length())))
}
// Shuffle for extra randomness
def password = passwordBuilder.toString().toCharArray()
Collections.shuffle(Arrays.asList(password))
password = new String(password)

vars.put("username", username)
vars.put("email", email)
vars.put("password", password)

return "Data generated successfully."

Usage in JMeter:

Add a Thread Group.

Add a JSR223 Sampler under the Thread Group and paste the script. Select groovy as the language.

Reference the variables in an HTTP Request with ${username}, ${email} and ${password}.

Scenario 2 – Correlated Order Data (Product ID & Quantity)

Requirement: simulate an e‑commerce order where each order contains 1‑3 items, each with a random SKU ID (1‑1000) and quantity (1‑5). The script must output a JSON request body.

Prompt to AI :

请帮我编写一个JMeter JSR223 Sampler（Groovy）脚本，模拟一个电商下单请求的数据。
业务背景：系统有1000种商品，每个商品有固定价格。性能测试需要模拟用户随机购买1‑3件商品。
生成规则：
商品ID (skuId)：从1到1000中随机选取。
购买数量 (quantity)：从1到5中随机选取。
需要生成一个订单项列表 items，列表长度为1到3（随机）。每个列表项是一个JSON对象，包含 skuId和 quantity。
最终，脚本需要输出一个完整的JSON字符串作为请求体，并将其存入JMeter变量 requestBody。

AI‑generated Groovy script :

// JSR223 Sampler (Groovy) – generate order request data
import groovy.json.JsonOutput
import java.util.concurrent.ThreadLocalRandom

def random = ThreadLocalRandom.current()

int itemCount = random.nextInt(1, 4) // 1, 2 or 3
def items = []
(1..itemCount).each {
    def skuId = random.nextInt(1, 1001) // 1‑1000
    def quantity = random.nextInt(1, 6) // 1‑5
    items << [skuId: skuId, quantity: quantity]
}

def requestBodyMap = [items: items]
def jsonRequestBody = JsonOutput.toJson(requestBodyMap)

vars.put("requestBody", jsonRequestBody)

return "Order data generated."

In the HTTP Request sampler set Body Data to ${requestBody} and add a Content‑Type header with value application/json.

Scenario 3 – Realistic Data Distribution (City & Amount)

Requirement: generate city names and order amounts according to weighted distributions that reflect real‑world traffic.

Prompt to AI :

请帮我编写一个Groovy脚本，生成模拟真实分布的数据。
用户城市分布：60% 的用户来自["北京","上海","广州","深圳"]，30% 来自其他省会城市，10% 来自三线城市。请按此权重随机生成城市名。
订单金额分布：70% 的订单金额在 50‑200 元之间，25% 在 200‑1000 元之间，5% 在 1000‑5000 元之间。按此权重生成金额。
将生成的城市和金额存入 JMeter 变量 city 和 amount。

AI‑generated Groovy script :

// JSR223 Sampler (Groovy) – simulate realistic data distribution
import java.util.concurrent.ThreadLocalRandom

def random = ThreadLocalRandom.current()

// 1. Weighted city selection
def tier1Cities = ["北京", "上海", "广州", "深圳"]
def tier2Cities = ["杭州", "成都", "武汉", "南京", "西安", "郑州", "长沙", "沈阳"]
def tier3Cities = ["保定", "扬州", "九江", "临沂", "绵阳"]

def allCities = tier1Cities + tier2Cities + tier3Cities

def weights = []
weights.addAll(Collections.nCopies(tier1Cities.size(), 60 / tier1Cities.size())) // 60%
weights.addAll(Collections.nCopies(tier2Cities.size(), 30 / tier2Cities.size())) // 30%
weights.addAll(Collections.nCopies(tier3Cities.size(), 10 / tier3Cities.size())) // 10%

def city = random.weightedValue(allCities, weights)
vars.put("city", city)

// 2. Weighted amount generation
def range = random.nextDouble()
def amount
if (range < 0.7) {
    amount = random.nextDouble(50, 200) // 70% between 50‑200
} else if (range < 0.95) {
    amount = random.nextDouble(200, 1000) // 25% between 200‑1000
} else {
    amount = random.nextDouble(1000, 5000) // 5% between 1000‑5000
}
amount = Math.round(amount * 100) / 100.0 // keep two decimals
vars.put("amount", amount.toString())

return "Realistic data generated."

Best‑Practice Summary

Language choice : Use Groovy in JSR223 Samplers for best performance (supports optional compilation).

Performance optimisation :

Prefer ThreadLocalRandom over Random for better concurrency.

Avoid creating large objects inside the script loop.

Use the Scripting mode rather than Compilation unless the script is completely static.

Uniqueness & thread‑safety : Use concurrent utilities such as AtomicLong to generate globally unique IDs safely under high load.

Debugging : Insert log.info(...) statements to print generated values; comment them out after debugging to avoid log overload.

Seed control : Set a random seed (e.g., random.setSeed(123L)) for reproducible runs, understanding that it reduces randomness.