Performing URL-Based Load Testing with Azure Load Testing

Load testing is a crucial practice for enterprise APIs to ensure optimal performance under varying traffic conditions. Azure Load Testing, a fully managed service, helps evaluate the performance, scalability, and capacity of your applications, particularly APIs. It enables you to simulate high-scale traffic, uncover performance bottlenecks, and optimize system resilience.

Azure Load Testing allows you to create sophisticated load tests using tools like Apache JMeter while also offering the flexibility to perform URL-based load testing without the need for external tools.

In this article, I will demonstrate how to perform URL-based load testing for an API exposed through the Azure API Management service using Azure Load Testing. 

Configuration of Load Test

Go to your Azure Load Testing instance, click on "Test", and then select "Create a URL-based test" to begin setting up your load test.

Specify Test details

In Test plan section click on Add request

Enter the details of your API, including the endpoint URL and request method. Since my API retrieves Star Wars characters by their ID, I configured the id parameter as a variable, allowing the test to simulate different requests dynamically.

Next, I created a CSV file containing multiple rows with different id values to be injected into the id parameter of my API. To simulate an error scenario, I intentionally included an id value of 1500, which does not exist in my API. This helps evaluate how the system handles unexpected inputs and errors under load.

In the Load section, I configured key parameters such as the number of engine instances, concurrent users, and test duration. Since this test is for demonstration purposes only, I selected the minimum values to keep the load minimal while still assessing the API’s response behavior.

For monitoring purposes, I added relevant resources to my load test, allowing me to analyze their performance directly from the load test dashboard. Since my API Management instance and Application Insights are crucial for this test, I included them to gain deeper insights into API performance, request handling, and potential bottlenecks.

Next, I navigated to the Test Criteria section, where I defined success and failure conditions based on key performance metrics. I set the test to fail if:

• The 90th percentile response time exceeds 1000ms
• The error percentage is greater than 5%

Execution of Load Test

With the Load Test configuration complete, it's time to execute the test and analyze the results.

Here are the results I received from the load test.

I got following key statistics

• Load
• Duration
• Response time
• Error percentage
• Throughput

The Load Test has failed because it didn’t meet the test criteria I set. Specifically, the error percentage surpassed the 5% limit.

Analysis of error requests using related resources

Since we've integrated Application Insights as a related resource, we can dive deeper into the analysis. Let's focus on the Failed Requests chart, which will provide insights into the specific requests that failed during the load test, helping us identify potential issues or bottlenecks in the API.

To further investigate, click on the Failed Requests chart in Application Insights, then select Drill into Logs and navigate to Failures.

The failure dashboard reveals that there were 9 "Not Found" (404) errors during the load test.

It tells me that there are no Star Wars characters with 1500 as the id