Extend the API to support Stateful containers

[aleixmorgadas]

After the improved support for Testcontainers in the Spring Boot 3.1, I'm using the feature extensively with several DB modules such as MongoDB and PostgreSQL.

I found that I need to customize a lot the modules definition to make the containers stateful and keep the state between executions without using the reuse feature.

You can see an example of how I'm extending the MongoDBContainer here in order to overcome some issues when restoring the state.

I found that Testcontainers API worked well for its initial purpose, Integration tests, but after being used for more than just test, I'm starting to experiment API gaps such as:

• Grateful container termination to keep the state consistent between runs.
  • Sometimes the data is corrupted due to containers being killed instead of stopped.
• Make it more straight forward to make a container stateful (opt-in) vs stateless (default mode).

Happy to help! 🙌

[mtremsal]

@aleixmorgadas thank you for sharing the idea. What's the rationale for the following point?

keep the state between executions without using the reuse feature.

[aleixmorgadas]

The reuse feature helps me to keep the state between executions while the docker is running. BUT, when the container is stopped/killed (like stopping the computer) the state is lost and I start from 0 when starting the application in dev mode again.

What I'm doing right now is adding volumes and customizing the container class, like MongoDBContainer to overcome some challenges when starting a container with a persisted state.

Here the problem I had with MongoDBContainer, it has a specific configuration on starting:
https://github.com/testcontainers/testcontainers-java/blob/main/modules/mongodb/src/main/java/org/testcontainers/containers/MongoDBContainer.java#L83-L87. This behavior makes the container break when the state is already there. In order to fix that, I needed to override the behavior by disabling sharding and replica set behavior. Then it worked just fine.

Same thing happens with PostgresContainer. Due to ungrateful shutdown, the state is corrupted and you need to change the waitFor strategy to enable the Postgres to recover the state.