An experiment with using property-based testing techniques to test a data access layer that connects to a SQL database, as well as adapting Go fuzz tests to more general property-based testing.
Currently, this is more of a fuzzer; the only property it tests is matching that the database stores the same data as an in-memory map, when the database and map are updated simultaneously.
I'd also like to explore some techniques for easily and quickly running isolated integration tests against a real Postgres database, mostly based on this blog post.
The fuzz test in pkg/storage/student_test.go creates a random sequence of SELECT, INSERT, UPDATE, and DELETE operations. The test then executes these with random parameters, calling the data access methods defined in pkg/storage/student.go. As well as performing these operations against a Postgres database (running in a Docker container), the fuzz test performs equivalent operations against a map and verifies that the database has the same data; the map is used as a test oracle.
To see the effectiveness of fuzz testing, we can intentionally introduce a bug that the fuzz test catches, while the unit tests don't. The GetStudentByID method has an erroneous query commented-out; instead of the correct query, the erroneous query simply returns the student with the lowest-sorted ID. If we comment out the correct query and uncomment the erroneous query, the unit tests still pass, because they only test a single record at a time. However, running the fuzz test catches the bug almost immediately.
Requirements:
- Go (>= 1.22)
- Docker
- The Just task runner
All of these are satisfied by creating a Github Codespace from this repository.
To run tests, first run just up to start a Postgres container and run SQL migrations to update its schema. Then, just test-unit runs the unit tests, while just test-fuzz runs the fuzz test as well.