introduce has_len_all_ranks() to check the length of dataloader across ranks

[ninginthecloud]

What does this PR do?

Fixes #9785

This PR has changed following:

1. add new function has_len_all_ranks() (I'm not sure if original has_len() has been widely used by users in public, so I just add a new util function. Feel free to let me know if I just need to replace has_len() directly. ) cc: @tchaton @awaelchli
2. add new attribute to LightningModuel/LightningDataModule allow_zero_length_dataloader_with_ddp
3. check the total length of dataloader across all ranks
4. update tests

test_common.py:test_evaluate[trainer_kwargs2] pytest.param(dict(accelerator="ddp_spawn", gpus=2), marks=RunIf(min_gpus=2))

Does your PR introduce any breaking changes? If yes, please list them.

No.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

[codecov]

Codecov Report

Merging #9827 (aa78bb5) into master (c33df26) will decrease coverage by 4%.
The diff coverage is 92%.

❗ Current head aa78bb5 differs from pull request most recent head 6b07d00. Consider uploading reports for the commit 6b07d00 to get more accurate results

@@           Coverage Diff           @@
##           master   #9827    +/-   ##
=======================================
- Coverage      92%     89%    -4%     
=======================================
  Files         180     181     +1     
  Lines       16147   16170    +23     
=======================================
- Hits        14934   14357   -577     
- Misses       1213    1813   +600     

[tchaton]

quick question. How does Detectron2 handles un-balanced batches as Lightning would definitely hanged otherwise.

[carmocca]

This will conflict with #9764, which one do we want to land first?

[ninginthecloud]

This will conflict with #9764, which one do we want to land first?

thanks for flagging~ #9764 is already ready to go, so I guess it will land first.

[awaelchli]

Hey, just saw this. Sorry for the delay on my response. I won't merge #9764 until 1.5 is released because of our current feature freeze :)

[ninginthecloud]

I noticed that CombinedLoader also relied on has_len() to get the length of dataloader link It's a little complicated to replace has_len() in get_len(). Any suggestions here? cc: @tchaton @awaelchli

[daniellepintz]

nit: I think you wrote the wrong thing under "Does your PR introduce any breaking changes?"

[tchaton]

@ninginthecloud Any chance you can get this PR ready for tomorrow ?