Replies: 1 comment 9 replies
-
Hi @pooja1423, thanks for sharing you feedback!
This should not happen as increasing the number of threads is the best way to increase parallelism/throughput of the SDK. Do you self-host Langfuse? Can you share your project id in case you run on Langfuse cloud?
Reducing the event size generally helps but should not be necessary unless you try to log really large events to Langfuse. |
Beta Was this translation helpful? Give feedback.
9 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hello, we are in the process of deploying Langfuse for our internal Langchain LLM pipelines that retrieve information from documents and we're running into long runtimes when Langfuse tracing in enabled. It might be the way we have it setup and would appreciate any feedback.
We're using the python SDK to setup the Langfuse client. A trace is created for each document and a callback handler is passed into the chain invocation using (trace.get_langchain_handler()). Each document is processed in a separate thread.
This pipeline for 100 documents takes 3 seconds, but the final flush() adds an additional 1+ minutes to the runtime. Here are things I have tried:
The flush time linearly increases with the number of documents which is limiting us from scaling and iterating quickly. Do you have any suggestions or things we can try/setup differently? Thank you
Beta Was this translation helpful? Give feedback.
All reactions