Respect batch size limits when flushing queue

[adam-on-tailwind]

Problem

I noticed that observations are being dropped by the client. When I looked into it further I realized that this happens when we exceed the batch size limit. The LangfuseCore#processQueueItems method is separating queued items into those that should be processed and those that should be skipped. The problem is that the flush method doesn't respect that segregation, despite the fact that it invokes processQueueItems. Therefore it's possible to send an oversized batch to the server. I'm pretty sure that the server is enforcing batch size limits, which is why we're seeing dropped observations.

Changes

I updated the flush method to consume the results of processQueueItems. Now it will only send the processedItems and will re-queue the remainingItems for the next time we flush. I suppose that there is a possibility that the queue could grow without bound. If that's an issue, my suggestion would be to implement a fixed size queue instead.

Release info Sub-libraries affected

langfuse-core

Bump level

• Major
• Minor
• Patch

Libraries affected

• All of them
• langfuse
• langfuse-node

Changelog notes

• Do not flush batches that exceed maximum batch size

[vercel]

@adam-on-tailwind is attempting to deploy a commit to the langfuse Team on Vercel.

A member of the Team first needs to authorize it.

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Disclaimer: Experimental PR review

PR Summary

This PR updates the flush method in LangfuseCoreStateless to respect batch size limits when flushing the queue, addressing the issue of dropped observations.

• Implemented processQueueItems method to separate items into processed and remaining queues
• Updated flush method to only send processed items and re-queue remaining items
• Added test case in langfuse.flush.spec.ts to verify correct handling of oversized batches
• Ensures client respects batch size limits, preventing server-side observation drops
• Potential for unbounded queue growth, suggesting consideration of a fixed-size queue implementation

_{2 file(s) reviewed, no comment(s)
Edit PR Review Bot Settings}