fix(sdk): remove duplicate generate_id functions, replace shortuuid with secrets

[moredatarequired]

Addresses WB-7375

Description

• Replace the dependency on shortuuid and prefer the built-in module secrets for token (run id) generation
• Consolidate the two nearly identical generate_id functions, move the callers depending on util to lib.runid

Testing

No additional tests.

Checklist

• Include reference to internal ticket "Fixes WB-NNNN" and/or GitHub issue "Fixes #NNNN" (if applicable)
• Ensure PR title compliance with the conventional commits standards

[codecov]

Codecov Report

Merging #4676 (4088eee) into main (d3e1e9e) will increase coverage by 40.50%.
The diff coverage is 97.95%.

❗ Current head 4088eee differs from pull request most recent head ce54c9b. Consider uploading reports for the commit ce54c9b to get more accurate results

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #4676       +/-   ##
===========================================
+ Coverage   42.80%   83.30%   +40.50%     
===========================================
  Files         279      280        +1     
  Lines       34555    34946      +391     
===========================================
+ Hits        14790    29111    +14321     
+ Misses      19765     5835    -13930     

Flag	Coverage Δ
functest	55.65% <67.34%> (+14.51%)	⬆️
unittest	73.93% <97.95%> (+47.71%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
wandb/util.py	87.26% <ø> (+40.77%)	⬆️
wandb/data_types.py	82.89% <87.50%> (+51.05%)	⬆️
wandb/apis/public.py	81.90% <100.00%> (+51.65%)	⬆️
wandb/filesync/step_checksum.py	92.42% <100.00%> (-0.12%)	⬇️
wandb/sdk/data_types/base_types/json_metadata.py	93.10% <100.00%> (+43.10%)	⬆️
wandb/sdk/data_types/helper_types/image_mask.py	90.76% <100.00%> (+61.08%)	⬆️
wandb/sdk/data_types/html.py	95.31% <100.00%> (+61.46%)	⬆️
wandb/sdk/data_types/image.py	86.01% <100.00%> (+31.74%)	⬆️
wandb/sdk/data_types/molecule.py	89.15% <100.00%> (+63.54%)	⬆️
wandb/sdk/data_types/object_3d.py	96.15% <100.00%> (+65.08%)	⬆️
... and 229 more

[dmitryduev]

very nice, thanks @moredatarequired!

[Linux-cpp-lisp]

This appears to have been missed in the docs...:
https://docs.wandb.ai/guides/track/advanced/resuming#automatic-and-controlled-resuming

Is it correct for client code following this example from the docs to just replace wandb.util.generate_id() with wandb.sdk.lib.runid.generate_id()? And is that second function available in earlier versions before v0.13.8 --- as in, if I make that change, will my code still be correct running on wandb<0.13.8?

Thanks!

[moredatarequired]

@Linux-cpp-lisp , I'm sorry about that! None of the util functions are intended to be used by client code, so that example in our docs shouldn't have used it all along. I'll fix the example and put in a fix PR that restores the original function with a deprecation warning.

In the meantime, I actually suggest replacing the call to generate_id with any code that generates your own id. There's nothing special about our id generation, you just need to create (and save, if following our example) a unique id for each run. I suggest using secrets.token_urlsafe() since that requires no external libraries or implementation code.

[Linux-cpp-lisp]

Hi @moredatarequired --- thanks for the quick response!

Sounds good, I'll replace generate_id() with secrets.tocken_urlsafe().

... a unique id for each run

Is there any advantage to using a user-readable / provided ID? In our training framework (https://github.com/mir-group/nequip) we do force the user to provide a unique run_name, would there be any advantage of using this instad?

[moredatarequired]

Is there any advantage to using a user-readable / provided ID? In our training framework (https://github.com/mir-group/nequip) we do force the user to provide a unique run_name, would there be any advantage of using this instead?

As long as the run id is url-safe and unique it should be fine. My worry about user-provided names would just be a lack of uniqueness, and if that's already enforced then it would be okay. Runs already have names that are more readable (generated names look like rosy-shadow-20), so I don't think there's an advantage to using a readable id, but it certainly won't cause a problem.