[Migration]: upgrade pydantic to v2 with sdk compatibility

[aybruhm]

Description

This PR improves compatibility with the latest version of Pydantic (v2), ensuring the SDK remains up-to-date.

Changes done:

• Improved and ensured the SDK types are compatible with pydantic>=2.
• Updated function_signature to handle type annotations to allow fastapi to translate SDK type.
• Refactored override_schema to align with Pydantic v2 schema definitions.
• Updated the behaviour of the getBodySchemaName function in openapi_parser.ts to return the appropriate schema $ref due to the upgrade to Pydantic version 2.

Minor Changes done:

• Updated Docker templates in docker-assets/ to ensure agenta installation precedes application dependency installation.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
agenta	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Jun 4, 2024 11:29am

[aybruhm]

Thanks for the great work @aybruhm

In its current form, the PR did not work for me. Not sure why it did for you. I created #1736 as a candidate fix that works for me. Not sure why it works and how it works. So please take a long look into it to understand the issue and the fix before merging.

Please also take the time to QA the PR in detail to make sure we don't introduce bugs here.

I tested both the previous and current versions of pydantic with two different variants, and both worked with the candidate fix (thank you). However, I'm unable to reproduce the issue that led to the previous solution that I did in the openapi_parser.ts file. I suspect I might have made a mistake while debugging the pydantic v2 upgrade with the SDK compatibility last week. 😕

[aybruhm]

Hi @bekossy ,

When you see this, could you please check why the frontend tests are failing? To reproduce, update the backend environment, AGENTA_TEMPLATE_REPO, to stage_templates. Thank you.

[bekossy]

@aybruhm I was able to reproduce the issue and it turns out that it is from the code and not the tests
I ran the docker-compose.test.yml file and created a new app from the UI but got stuck here

[aybruhm]

@aybruhm I was able to reproduce the issue and it turns out that it is from the code and not the tests I ran the docker-compose.test.yml file and created a new app from the UI but got stuck here

Thanks, @bekossy! I've reproduced the issue. Could you please pull and test again? I've fixed the problem. @mmabrouk, the issue was the body schema of the LLM application that was created from the UI. Here it is:

It turns out that the body schema for an LLM app created from the UI is different from the one created from the CLI.

Next steps for QA:

• Create and run an app locally from the CLI and UI using the current agenta version (with pydantic v1).
• Checkout this PR branch and repeat the above steps. Before creating an app from the UI, update the backend environment variable AGENTA_TEMPLATE_REPO to stage_templates and start the Docker Compose.

[mmabrouk]

@aybruhm I was able to reproduce the issue and it turns out that it is from the code and not the tests I ran the docker-compose.test.yml file and created a new app from the UI but got stuck here

Thanks, @bekossy! I've reproduced the issue. Could you please pull and test again? I've fixed the problem. @mmabrouk, the issue was the body schema of the LLM application that was created from the UI. Here it is:

It turns out that the body schema for an LLM app created from the UI is different from the one created from the CLI.

Next steps for QA:

• Create and run an app locally from the CLI and UI using the current agenta version (with pydantic v1).
• Checkout this PR branch and repeat the above steps. Before creating an app from the UI, update the backend environment variable AGENTA_TEMPLATE_REPO to stage_templates and start the Docker Compose.

I would say instead of trying to use the templates. A much easier solution is to serve applications from the CLI using the poetry shell with agenta pointing to this branch, then again with the CLI using the production branch and make sure that everything is working.

Specifically:

• All types of parameters are shown and work (for this we should serve both the chat application and the single prompt application as they have all types of parameters)
• The application can be used from the playground (generate endpoint) and in production using the endpoint shown under deployments (after publishing the application). We need to make sure that the deployed endpoint has the correct version of the configuration. For this please change the prompt dramatically (e.g. add answer only in chinese) then deploy the application

[aybruhm]

That's all I have been doing. :-)

@aybruhm I was able to reproduce the issue and it turns out that it is from the code and not the tests I ran the docker-compose.test.yml file and created a new app from the UI but got stuck here

Thanks, @bekossy! I've reproduced the issue. Could you please pull and test again? I've fixed the problem. @mmabrouk, the issue was the body schema of the LLM application that was created from the UI. Here it is:

It turns out that the body schema for an LLM app created from the UI is different from the one created from the CLI.
Next steps for QA:

• Create and run an app locally from the CLI and UI using the current agenta version (with pydantic v1).
• Checkout this PR branch and repeat the above steps. Before creating an app from the UI, update the backend environment variable AGENTA_TEMPLATE_REPO to stage_templates and start the Docker Compose.

I would say instead of trying to use the templates. A much easier solution is to serve applications from the CLI using the poetry shell with agenta pointing to this branch, then again with the CLI using the production branch and make sure that everything is working.

Specifically:

• All types of parameters are shown and work (for this we should serve both the chat application and the single prompt application as they have all types of parameters)
• The application can be used from the playground (generate endpoint) and in production using the endpoint shown under deployments (after publishing the application). We need to make sure that the deployed endpoint has the correct version of the configuration. For this please change the prompt dramatically (e.g. add answer only in chinese) then deploy the application

[aybruhm]

Everything LGTM! 👍🏾

[mmabrouk]

Great job @aybruhm
I think this only needs cleaning up the changes and it's ready to merge. Then let's update the templates (to catch any potential bugs early).

[mmabrouk]

Thanks @aybruhm for solving this challenging issue, great work!