Generative AI support for Spring Petclinic Microservices

[odedia]

This PR adds a new microservice to support a generative ai chatbot based on Spring AI.
It supports natural language for performing activities like listing vets, listing owners, adding owners and adding pets to owners.
If the microservice is not booted, the chatbot simply response that chat is not available.
Based on https://github.com/spring-projects/spring-petclinic/tree/spring-ai
The code was adapted to keep domain boundaries - the functions now call the relevant microservice to perform the various activities as needed.
It also include an example of using RAG in the listVets function.
Thank you.

Task list:

• Sonarqube quality issues (excluding code duplication warnings)
• Move API key to an env variable.
• Move defaultFunctions() to code instead of properties.
• Remove lombok where possible (VisitDetails), use records instead.
• Change FallbackController to POST only.
• Update spring-petclinic-api-gateway/src/main/resources/application.yml to POST only.
• Declare marked as a webjar
• Upgrade the architecture diagram
• Test local and Docker deployment

[arey]

Hi @odedia, thanks for this contribution I have to review.

Since my last email to you and @showpune, I've reworked a bit your spring-ai branch and published it in a brand new repository: https://github.com/spring-petclinic/spring-petclinic-springai
The chatbot is always active. I will push other improvements trough Pull Request.

Over the next few months I plan to make a version with Lanchain4j.

[odedia]

Thanks!
The purpose of this exercise is to show spring ai naturally integrating with a Microservices architecture.
the chatbot will not do anything if the genai microservice is not running.

[sonarqubecloud]

Quality Gate failed

Failed conditions
6 Security Hotspots
3.1% Duplication on New Code (required ≤ 3%)
B Reliability Rating on New Code (required ≥ A)

See analysis details on SonarCloud

Catch issues before they fail your Quality Gate with our IDE extension SonarLint

[arey]

Thank you @odedia for your contribution to modernizing the architecture by adding a very hype GenAI service in 2024.
I've given you a lot of feedback. Topics are open for discussion.

In order to be able to merge this PR on the main branch, to remain open to as many developers as possible, I think the sample application needs to start without an OpenAI or Azure account. The genai servide should be optional or start in LLM mode.

[arey]

suggestion: most of the configuration properties are located to the spring-petclinic-microservices-config repository, especially the docker profile. I suggest to create another PR to report it.
https://github.com/spring-petclinic/spring-petclinic-microservices-config/

[arey]

question: how could be the credentials managed in a Docker image exported to a registry? We have to answer this question whether we leave the config here or centralize it on the server config.

[odedia]

I'm not sure if we can reliably export the credentials in the config server. Since this is private sensitive data, and the config server is publicly available on Github, the creds.yaml should remain in the source code of the genai service. It is listed for ignore in .gitignore.

[arey]

The config repository could be cloned in a private repository. It's URL could be provided with the CONFIG_SERVER_URL environment variable. I agree: it's not very practical.

I'm not very fan of adding the creds.yaml in a Docker image. We could use a Docker volume. But I'll prefer to use an environment variable like AZURE_OPENAI_KEY and AZURE_OPENAI_ENDPOINT. See spring-petclinic/spring-petclinic-ai@fbaf8dc#diff-54eeffbae371fcd1398d4ca5e89a1b8118208b7bb2f8ddf55c1aa2f7d98ab136R36
Developers can pass those parameteres from their IDE or from the Docker command line. What do you think about?

[odedia]

Yep, that would work.

[arey]

suggestion: we need to offer a solution to developers who don't have an Azure or OpenAI account and don't want to pay for it: either enable fallback by default, or disable chat, or have a local solution with ollama. What do you think @odedia ?

[odedia]

I began with using Llama 3.1, but the chat client became incredibly verbose and not as useful as OpenAI, that's why I switched to it.

[arey]

@odedia I've switched the Maven configuration from Azure OpenAI to OpenAI with the free demo account that OpenAI suggests. To use it, we have to use the gpt-4o-mini model.

[arey]

There are also somme issues detected by the Sonar Quality Gate.

[odedia]

Sonarqube is mostly complaining about Code duplication, because I copied the data classes from existing microservices to the new one.
However this is the correct approach in a microservices architecture, you don't want to have dependencies on other microservices.
A better approach would be to use contract-based DTO generation but that's a rather large undertaking for the entire project.

[odedia]

I don't see a develop branch in the upstream project, is that exposed to all?

[arey]

Sonarqube is mostly complaining about Code duplication, because I copied the data classes from existing microservices to the new one. However this is the correct approach in a microservices architecture, you don't want to have dependencies on other microservices. A better approach would be to use contract-based DTO generation but that's a rather large undertaking for the entire project.

Putting code duplication to one side.
We can focus on the 12 issues. I can help you if you want.

[arey]

I don't see a develop branch in the upstream project, is that exposed to all?

I got confused. I meant the main branch

[odedia]

Any help would be appreciated, I will try to implement the other changes towards the weekend.

[selvasingh]

@arey when should we expect this PR to be merged? We like to use this sample on Azure in docs, demos and workshop.

-- Asir Selvasingh

CC @odedia

[arey]

Hi @selvasingh
My feedback has been slow to be taken into account. I guess @odedia needs some help. Would you be willing to help?

[odedia]

Sorry, this was put in my backlog.
There's quite a backlog of tasks here, can you confirm this is what we require?

1. Sonarqube quality issues (excluding code duplication warnings)
2. Move API key to an env variable.
3. Move defaultFunctions() to code instead of properties.
4. Remove lombok where possible (VisitDetails), use records instead.
5. Change FallbackController to POST only.
6. Update spring-petclinic-api-gateway/src/main/resources/application.yml to POST only.
7. Declare marked as a webjar

Thanks

[arey]

Sorry for the long to-do list. I will also add:

8. Move configuration properties of the spring-petclinic-genai-service module to the spring-petclinic-microservices-config/
9. The aim is for the demo application to work in standalone mode, disconnected from the network, suggest a way to disable the geani-service or provide a local LLM solution
10. Upgrade the architecture diagram
11. Test local and Docker deployment

I hope I haven't forgotten anything.

From your PR, I suggest creating a working feature/genai branch on that repository. This allowed many developers to contribute. Personally, I could work on it next week.

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud