This is the test framework for evaluating durability of MongoDB Atlas.
A core promise of database system is that even after a potential server failure, any updates are durable and remain in the database. However, does this property also hold for distributed systems like MongoDB Atlas, a popular cloud-based database system? Since durability issues cannot be directly observed by users, so how do we measure them?
Our purpose is to develop a test harness that can detect durability bugs in MongoDB Atlas, and that can be used as a tool for database company to verify the bug and make improvements.
We used this test framework to evaluate 2 versions of MongoDB, one with known durability bug. We can see the buggy version lost writes after a simulated server crash, while the latest MongoDB version tested fine. This tool can indeed help database developers to find the bugs in their code.
Gradle 6.4 has been used for this project
All the source code can be found under the path: /gra_test/src/main/java/gra_test/
To run the code in AWS EC2 VM, copy the repo and type in:
./gradlew run
The result file containing all the records after run should be placed under /gra_test/src/main/java
For data analysis, please use python file: analysis.py and specify the relative path of result file.