Database Autonomy Service:Traffic playback and stress testing

Scenarios

• Determining whether the current database instance type needs to be scaled out to handle upcoming short-term peaks in service traffic.

• Verifying the effects of a database schema evolution, especially an index change, in a real-world scenario. This reduces the risk of failures after the changes are published.

• Reproducing a failure scenario on a cloned database to identify the cause. This is useful if the cause of a failure cannot be determined when it occurs.

Feature architecture

Feature introduction

Traditional stress testing solutions often use tools such as sysbench and TPCC. In a given scenario, these tools vary data write and read speeds to measure the effect on performance. They automate repetitive tasks such as data generation, testing, and reporting. However, the SQL used by these tools is very different from real service traffic. As a result, the test results cannot accurately reflect how an instance's performance and stability will change in a real-world scenario.

The traffic playback and stress testing feature in DAS captures real service traffic. It dynamically replicates and amplifies the pressure on the production system. This provides more objective and valuable results to help with O&M decisions. The architecture is not a simple two-point model that consists of a client and a test object. Instead, it uses a three-point architecture: production database, stress testing client, and test object. This architecture captures the pressure on the production database and then realistically simulates that pressure on the test object for a specific period. The traffic can also be replayed at different speeds.

Prerequisites