[1/n] Add LightningFetcher

[tchaton]

What does this PR do?

This PR introduce the LightningFetcher.

Next PR will add fault tolerant training components.

Parts of #8867

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

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 #8890 (92abd0e) into master (e060547) will increase coverage by 4%.
The diff coverage is 89%.

@@           Coverage Diff           @@
##           master   #8890    +/-   ##
=======================================
+ Coverage      89%     93%    +4%     
=======================================
  Files         172     173     +1     
  Lines       14178   14182     +4     
=======================================
+ Hits        12568   13174   +606     
+ Misses       1610    1008   -602     

[carmocca]

Is there some bigger picture somewhere of how this is going to be used, what will this replace, and how will users interact with or override it?

[tchaton]

Is there some bigger picture somewhere of how this is going to be used, what will this replace, and how will users interact with or override it?

Hey @carmocca, yes definitely.

The LightningFetcher was introduced through this RFCs: #8867 and will be used to replace the prefetch_function. This should also be used for validation purposes.

There is a long way before we make this object public API, however it is there to solve 3 primary issues:

• Handle Fault Tolerant dataloader dictionary state dict aggregation as used within this PoC: https://github.com/PyTorchLightning/pytorch-lightning/pull/8476/files and this PR: [Feat] 2/n Add Fault Tolerant Training to LightningFetcher #8891
• Add support for inter batch parallelism
• Provide a way for users to customise data being passed to their training_step.

Best,
T.C

[SeanNaren]

Awesome!

[ethanwharris]

LGTM 😃