Reality-Driven Testing using TestContainers
- 2. Oleksii Holub - Working at Bitwarden 🧑💻 - Previously: Sentry - Developer tooling & experience ⚙️ - Open-source maintainer 🐙 - CliWrap (~6M downloads, ~4K stars) - YoutubeExplode (~2M downloads, ~3K stars) - CliFx (~0.6M downloads, ~1.5k stars) - …more - Tech writer & speaker ✍️ - Unit Testing is Overrated (~400k reads) - Prefer Fakes Over Mocks (~50k reads) - Fluent Generics in C# (~15k reads) - …more - Microsoft MVP & GitHub Star ⭐ tyrrrz.me
- 3. Let’s take a step back Why do we need testing?
- 4. Premises of (Automated) Testing - Gain confidence that the software works correctly 😤 - Confidence is a product of authenticity 🪞 - Authentic test processes provide more confidence - Contrived test processes provide less confidence - Some components may be hard to test 🪞 - Because they are not easily accessible (e.g. cloud services) - Because they are non-deterministic (e.g. dependent on outside effects) - Because they are slow (e.g. relying on heavy I/O or computations) - They can be replaced with test doubles - Their behavior can be tested in isolation - This reduces authenticity and thus confidence tyrrrz.me Prioritize this Avoid this
- 6. Reasons to Use Real Dependencies - Overuse of test doubles erodes confidence 🧑 - Less of the actual application surface is tested - Potentially impossible states may be tested - Especially when mucking with internal state - Crucial infrastructural concerns are not tested - Database interactions - Request routing & handling - Mocks make tests more complicated 🧑 - Tests should be trivial to understand - Tests should not rely on implementation details - Tests that contain logic need other tests to test themselves - But we can’t just use real dependencies… 😔 - …or can we? tyrrrz.me
- 8. - Code-first container orchestrator (like Docker Compose) - Single API, multiple platforms (.NET, Java, Go, NodeJS, Python, etc.) - Works anywhere Docker is supported - Open-source and community-maintained - Built-in kits for popular services (e.g. Redis, MSSQL, Kafka) - Single reusable infrastructure setup tyrrrz.me TestContainers
- 10. tyrrrz.me Configure the container Wire the container to the execution context Use the container inside the test
- 11. 📺 Demo time
- 12. Performance Considerations - Docker Pull is the slowest part 🐌 - Mostly just affects CI environments - Many platforms pre-cache popular images - Reducing image size is important - Cold start overhead is constant ❄️ - Test suite becomes more efficient as it grows - More tests means faster relative execution time - Different isolation strategies can yield better results tyrrrz.me
- 13. Test Isolation Strategies - No isolation 🚫 - All tests share the same environment - Environment gets reset after each test - Tests cannot be parallelized - Domain isolation 🛍🧑 - All tests share the same environment - Operations are scoped within a logical scope (e.g. tenant) - Each test gets its own scope - Storage isolation 📝 - All tests share the same environment - Each test gets its own database instance - Environment isolation 📦 - Each test gets its own environment tyrrrz.me
- 14. Summary - Testing against real dependencies… - Is easy (as long as there’s a Docker image) ✅ - Is quite fast ✅ - Simplifies testing code ✅ - Provides more confidence ✅ - TestContainers… - Supports many different development platforms ✅ - Lets you easily share test infra between projects ✅ - Testable software… - Is not just about dependency inversion and interfaces ❌ - Is sometimes just about using the right technologies ✅ tyrrrz.me