Duplicate UserWarning Logs for `lightning-sdk` Version Check

[deependujha]

🐛 Bug

While using either optimize or StreamingDataset, multiple logs for the same warning is shown. It should be at max one.

To Reproduce

Use Optimize function when your machine's lightning-sdk is not the latest version.

Code sample

# Ideally attach a minimal code sample to reproduce the decried issue.
# Minimal means having the shortest code but still preserving the bug.

Expected behavior

Additional context

Environment detail

• PyTorch Version (e.g., 1.0):
• OS (e.g., Linux):
• How you installed PyTorch (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information:

[bhimrazy]

It seems that the lightning_sdk does not have a warning cache. I believe that adding something similar to the WarningCache from litdata in the Lightning SDK could help reduce unnecessary version-related warnings.

For reference, here is the WarningCache implementation in litdata:
https://github.com/Lightning-AI/litdata/blob/4ff18da3bfa04555de7fa92b6287c368c8c27c31/src/litdata/helpers.py#L9-L10

Alternatively, it could be directly utilized from lightning_utilities.
https://github.com/Lightning-AI/utilities/blob/9e88903a7cecd7933da6ef659f1eee22ac652de1/src/lightning_utilities/core/rank_zero.py#L95-L113

Question:
By the way, it seems that rank_zero_only.rank needs to be set in this context. How is that typically configured?

cc: @Borda @tchaton

[Borda]

Yes these warning shall not be duplicated, so let's update the SDK accordingly...

[deependujha]

I think warning cache might not work for multiple workers in Dataloader. Rather we should check for it in StreamingDataloader (litdata version).

But, I doubt if this will work for ddp, as multiple dataloaders get instantiated.

[bhimrazy]

I think warning cache might not work for multiple workers in Dataloader. Rather we should check for it in StreamingDataloader (litdata version).

But, I doubt if this will work for ddp, as multiple dataloaders get instantiated.

yeah, makes sense.