[ASK] Any tips to improve performance for a large scale flow?

[tzrfabian]

Hello, i just want to ask, is there any tips to improve a langflow's flow that built on large scale? Example pic attached at bellow.
Because currently each questions asked in these flow process can take around 20 sec - 2 minutes. Should i upgrade my server specs or maybe there's other way to improve flow process?

Thank you.

[andifilhohub]

The root cause of your flow taking 20 seconds to 2 minutes is tied to state-management overhead inside Langflow.

My recommendation is to disable anything that increases state-tracking cost:
• Disable Vertex build recording.
• Disable transactions.
• Disable tracing.

Flow efficiency is directly correlated with Langflow’s state-control mechanisms. Every state-related feature you turn off reduces orchestration overhead and improves latency.

In version 1.7.0, the new lfx serve (stateless) mode will be available. You should consider adopting this option. A stateless execution model can significantly mitigate the delays you’re experiencing.

Review the environment variables in the documentation to identify which state features you can safely keep disabled.

Regarding LFX, it may not be fully documented yet, but you can get ahead by reviewing this PR:
#10498