langfuse
chore: Improve timeline view front-end latency for very large traces with high level of nesting
Background: we launched the trace timeline view a few months ago, for users to help identify performance bottlenecks and visualise parallelisation.
However, for deeply nested traces we're hitting UI performance issues in rendering the timeline view.
Set up for debugging: Using our langfuse seeder and likely some clever prompting you could construct a deeply nested large trace object with around 1k spans and some lorem epsum input/output. Current behaviour: When toggling all traces, the view loads for a few seconds to show all nested children (as well as their children). When attempting to untoggle a single child, the ui is very slow to respond and the cursor is not fully predictable. Desired behaviour: initial toggle may remain slow, however single toggle action latency should be improved.
Additional context: Please note that this task requires some more research into front-end rendering performance, and is on the more experimental side. This might involve some virtualisation. This issue will be non-trivial to fix so I'd recommend only if you're excited about front-end performance and a somewhat seasoned front-end developer
Hi @marliessophie! I’m interested in tackling the performance issues in the timeline view for deeply nested traces.
Plan: Debugging Setup: I’ll use the Langfuse seeder to create a large trace object with around 1k spans. Performance Research: Focus on front-end rendering optimization, potentially exploring virtualization techniques to enhance latency during toggling actions. Testing & Optimization: I'll monitor the UI response times for both initial toggles and single child untoggles. Looking forward to improving the user experience! Thanks for the opportunity!
