Should the Langfuse classs initialization be cached (does it do any I/O on init?) ?

[shner-elmo]

Hey, I'm calling chain.invoke() in multiple modules.
So I'm wondering if I should be careful with initializing this class only once, and sharing the same instance across different modules, or if it has no real CPU or I/O cost that I can initialize it on the fly, perhaps even each time I invoke the LLM?

Thanks

from langfuse import Langfuse

langfuse = Langfuse(
  secret_key="...",
  public_key="...",
  host="http://localhost:port"
)

[marcklingen]

Do you use the Langchain integration? If yes, I'd create a single CallbackHandler instance for the best performance and scalability. Otherwise, you will create multiple Langfuse client instances which is not super expensive but not necessary.

[shner-elmo]

Correct, with the Langchain integration.
Can you elaborate on whats the cost? Does it verify the credentials or is it simply initializing a class? Thanks

[marcklingen]

It just initializes the class and spawns a background process which handles the queue of events that need to be send to langfuse. Thus this will have some (minor) cpu/memory footprint which can be noticed when creating many instances in parallel.

[shner-elmo]

Got it, so you definitely don't want to create a new one each time, but even one per module should be ok, thank you.