Hi all, coming back to Flyte after a couple months...
# flytekit
s
Hi all, coming back to Flyte after a couple months working on other things. Anyone found the FrameProfileRenderer to be really slow? I’m trying to generate statistics for a workflow with <100k rows and its been running for over an hour now. Seems to chew up whatever memory I throw and it but its only using one thread. Any ideas how I could improve this? Or is there something else I should be doing?
h
Not familiar as much but it seems to use https://docs.profiling.ydata.ai/latest/ Under the hood... Maybe the project's page offer some insight?
s
It seems they have a minimal mode but currently no way to configure it for the plugin
h
What would you like to configure?
s
Literally just the flag to tell it to remove the really expensive statistics
Either way, for now looks like I can’t use this profiler on any moderately sized dataset.
h
I see.. would you be willing to open a PR against flytekit? You can copy that renderer into your own code, make the change and use it in your type annotation until we upstream the change
s
Unlikely to have time for it but if it keeps bothering me then I’ll give the minimal mode thing a try
h
Do you know the exact code/flag that needs to be set?
s
Copy code
profile = ProfileReport(large_dataset, minimal=True)
w
s