Plans to support Langfuse 3.0 Architecture

[covolution]

This issue is for a: (mark with an x)

- [ ] bug report -> please search issues before submitting
- [X ] feature request
- [ ] documentation issue or request
- [ ] regression (a behavior that used to work and stopped in a new release)

I see that Langfuse will be making significant architectural changes in version 3.0: Adding a worker, Redis, and Clickhouse.

What will happen to this project? Are you planning to support the new architecture?

Versions

3.0

[pamelafox]

Ah good to know! I've posted in that thread:
https://github.com/orgs/langfuse/discussions/1902#discussioncomment-9928976

[JungeAlexander]

@pamelafox just a kind reminder that the langfuse team is reaching the last stages of finalizing v3 and is requesting feedback here: https://github.com/orgs/langfuse/discussions/1902#discussioncomment-11268939

Maybe now is a good time to reach out to the langfuse maintainers to start working on a branch to update the template here. Thanks for maintaining this template!

[pamelafox]

It looks like it'll be a lot harder, given its using Clickhouse and S3, which wouldn't be hosted on Azure, right? We do have an Azure Redis. What parts would you expect to be hosted on Azure?

[JungeAlexander]

Good questions. I am unsure about Clickhouse but other Blob Storage options should be supported. Storing data in multiple places and cloud vendors is probably not what people using this template want but instead running it all on Azure.

[Steffen911]

Hey all,
I'm an engineer over at Langfuse and found this issue from the Langfuse v3 discussion thread.

We will add Azure Blob Storage support within the next Langfuse release (likely v2.93.0) coming later this week (see https://langfuse.com/docs/deployment/v3/components/blobstorage#azure-blob-storage-beta for the setup steps). We did a smoke test using Azurite and an Azure account of ours and it seems to perform well across our blob storage use cases.

With Redis on Azure and Azure Blob this leaves the concerns around ClickHouse. Here, a single Docker container with a disk volume might handle smaller use-cases without high availability (ClickHouse docker image). ClickHouse cloud also offers a managed service running on Azure which can be consumed via the Azure marketplace (docs). Would either of those options address the concerns around this samples repository? The third option is a self-hosted ClickHouse cluster on Kubernetes which might be beyond the scope of this deployment sample.

Let me know if you have any questions!

[pamelafox]

Thanks @Steffen911 for the details. For ClickHouse, I don't think we'd want to use the Docker container with disk volume as using disk volumes aren't recommended for Container Apps (https://learn.microsoft.com/en-us/azure/container-apps/storage-mounts?tabs=smb&pivots=azure-cli - either they're too ephemeral or read/write ops are too slow).
So I'll probably try ClickHouse marketplace next, before resorting to Kubernetes.

[marcklingen]

Langfuse v3.0.0 is stable now and we (especially @Steffen911) would love to help build an Azure sample for this.

For reference, tracking this here in Langfuse GitHub discussions as well: https://github.com/orgs/langfuse/discussions/4647

More context on this change and potential upgrade steps: https://langfuse.com/self-hosting/upgrade-guides/upgrade-v2-to-v3

[pamelafox]

Thanks @marcklingen, I need to learn kubernetes. I've talked to some colleagues that work with AKS and should be able to help.

[marcklingen]

I think we can also set up something more Azure-native if K8s is not the best way to go on Azure for this. e.g. on AWS we consider supporting ECS/Fargate, maybe there is an equivalent here that makes Langfuse easier to manage on Azure without needing to know Kubernetes -- for everyone familiar with K8s, there is already the helm chart as a standardized template

[Marilog]

@marcklingen maybe it is too early to ask 🙈but is there any approximate timeline for the release of the solution deployment in Azure?
Is it possible to know which resources are going to be created and deployed to meet the requirements of Langfuse 3.0 architecture?

[marcklingen]

Awesome, @pamelafox please let us know if we can work more directly together on this. @Steffen911 is the core maintainer of Langfuse on the infrastructure side.

@Marilog no ETA yet, in the meantime you can use the helm chart or track progress here as well (pls upvote): https://github.com/orgs/langfuse/discussions/4647