Testing and development are inseparable for MySQL developers. Due to the complex internal mechanisms of MySQL, external testers often struggle to grasp its logic and typically can only conduct black-box testing. Given MySQL's complexity, as many test cases as possible are generally required to reduce the workload of regression testing. Therefore, MySQL developers need to understand how to effectively utilize test cases to test their own programs.
Not every problem necessitates designing test cases; sometimes, the complexity of designing test cases exceeds that of fixing the code. For example, with Paxos-based communication within MySQL, there are no specific test cases, primarily relying on extensive manual testing by developers.
The principle of test case design should be to provide test cases where the cost is reasonable, avoiding them where the cost is excessive. Balancing this relationship effectively improves development efficiency.
It's worth noting that the test cases for Group Replication have been found to be too lax, leading to significant challenges in regression testing after refactoring. Excessive and lax test cases can consume substantial development time, making it difficult to determine whether the problem lies with the test cases themselves or with the refactored code.
Testing results can fluctuate due to various environmental factors, necessitating consistent environments to ensure fairness. The introduction of SSDs has significantly reduced MySQL's I/O wait times; however, if SSD performance degradation is not monitored during testing, results may diverge from expectations. Using commands such as fstrim -a can mitigate SSD degradation effects, ensuring tests are less affected.
Linux operating systems utilize I/O caching, which can lead to variability in test results, especially over long intervals between tests. Therefore, it's advisable to minimize the time gap between two tests and repeat the initial test to assess fluctuations. If significant, the test results may become invalid.
In this book, testing tools like BenchmarkSQL or SysBench are typically employed. BenchmarkSQL increases data volume over time, so for fairness and comparability of performance, tests commence after database refactoring to ensure reproducibility of results.
The most efficient testing method is to utilize real online traffic for evaluation directly. However, this can be too risky for databases. A safer approach is to replicate online traffic into a dedicated testing environment.
In Oracle, Database Replay enables testing a system with real production workloads, helping identify potential problems before implementing changes on the production system. Any workload period can be captured with little overhead and used to drive a test system, maintaining the concurrency and load characteristics of the real workload. Maintaining these characteristics is crucial, as current testing solutions often lack synchronization based on data dependencies. Without proper synchronization, the workload does not perform as required, leading to poor coverage and inadequate load, leaving many problems undetected. Database Replay's data-based synchronization makes testing realistic and helps discover potential problems [34].
In MySQL, a common strategy involves taking a MySQL secondary instance offline for testing, configuring the necessary cluster, and replicating online MySQL requests to this new testing primary. The closer the testing primary resembles the production environment, the more accurate the test results. There are various methods to replicate online MySQL requests. This book recommends the open-source tool TCPCopy [60]. By using TCPCopy, many online problems have been successfully effectively resolved, laying a solid foundation for MySQL proxy enhancements [61]. For testing a MySQL cluster, replicating online requests to the testing system using TCPCopy allows us to evaluate whether the modifications achieve the expected outcomes, such as performance improvements, and robustness.
Testing serves not only to verify known problems but also to uncover new ones. Verification of problems is the initial step to ensure that the software behaves as expected. Subsequently, the focus shifts to actively discovering potential problems within the program, preempting their discovery by testers. Adopting this strategy during the process of improving MySQL fundamentally ensures the quality of MySQL modifications. Practical experience has validated this approach as highly effective. Where possible, replicating online traffic for testing provides a robust means to identify potential problems within the software, further enhancing its overall quality.