Support docarray pydantic v2

[JoanFM]

Support docarray pydantic v2

[TamiasSibiricus]

@JoanFM have find why openapi schema sometime fails. This happened when LegacyDocumentJina appears in schemas list and fall down in pydantic lib with schema generation on two fields:

class LegacyDocumentJina(BaseDoc):
    ...
    chunks: Optional[Union[DocList[LegacyDocumentJina], List[LegacyDocumentJina]]] = None
    matches: Optional[Union[DocList[LegacyDocumentJina], List[LegacyDocumentJina]]] = None
    ...

Seems that both fields contain doclists with docs refered to self. when i comment this fields in my fork everything works fine.

I just do not car about LegacyDocumentJina fields for backward compatibility BTW.

[JoanFM]

Yes, I also discovered this, but then some new things appear. WIll post more as I find, thanks a lot for the help. If you can do a PR to try all your changes it would be great

[MimicTester1307]

Hi @JoanFM if it's is still needed, I'd like to work on this issue or #5999. I think it'll help familiarize me with the codebase.
Additionally, I tried accessing https://learn.jina.ai/ (I had bookmarked it), but it now redirects to the jina.ai home page. Will the learning resources be up again soon?

[JoanFM]

Hey @MimicTester1307 , I do not think these learning resources wlill be back again. Where did you get the link from?

I think you can try to work on the #5999 . This ticket may be complex because we may need to.keep compatibility with 2 pydantic versions at the same time.

[MimicTester1307]

I see, the courses were pretty comprehensive, though. I think I got it from the JinaAI site initially (a couple of months ago). I bookmarked it for later when I was more comfortable to attempt a contribution.

Alright, thanks. I'll take start with #5999. I can assign it to myself, right?

[JoanFM]

I see, the courses were pretty comprehensive, though. I think I got it from the JinaAI site initially (a couple of months ago). I bookmarked it for later when I was more comfortable to attempt a contribution.

I will yes

[Wh1isper]

Hello, may I ask what is stopping us from upgrading to pydantic v2?

I know there are many integration tests that fail due to upgrades in #6132 . But I think a more beneficial approach would be for jina to upgrade a version number, something like 3.23 -> 3.30, and provide security fixes and important backports for 3.2x only. Once the downstream projects are upgraded, we can stop supporting 3.2x.

This is not in conflict with maintaining Pydantic V1 compatibility, as I know more and more projects are based on Pydantic V2, which forces me to do more complex isolation designs when adopting jina.

[Wh1isper]

clip-as-service uses docarray, which allows us to interact with clip by way of microservices👍, but clip-as-service's clip_client needs to rely on jina, which results in the project that relies on the clip_client being tied to the pydantic V1, though I just need a client... 😂

I understand that clip_client can't run without jina, so we may should push jina to get with the times.

[JoanFM]

Hello, may I ask what is stopping us from upgrading to pydantic v2?

I know there are many integration tests that fail due to upgrades in #6132 . But I think a more beneficial approach would be for jina to upgrade a version number, something like 3.23 -> 3.30, and provide security fixes and important backports for 3.2x only. Once the downstream projects are upgraded, we can stop supporting 3.2x.

This is not in conflict with maintaining Pydantic V1 compatibility, as I know more and more projects are based on Pydantic V2, which forces me to do more complex isolation designs when adopting jina.

Right now, what is blocking is mostly the time availability from my side to dedicate to this feature. If someone from community feels like providing this I would love to help them out however.

[jina-bot]

@jina-ai/product This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 14 days

[TamiasSibiricus]

@JoanFM I have tried with latest changes in PR #6150 and everything works like a charm except one thing. when i start flow as service in docker container with grpc protocol only sometimes got error with openapi schema generation related to LegacyDocumentJina. Maybe add error report later. Just to be clear this happens only first time when i rebuild image for docker container based on jina 3.25.1 image. Maybe some artefacts.

[JoanFM]

Hey @TamiasSibiricus ,

Yes I have observed that, but there is some more nuances about having DocList actually be a List and so on. I hope to get some time to work on that again.