support launching Lightning ddp with traditional command

[awaelchli]

What does this PR do?

Fixes #7003

The Lightning environment provides a convenient way to launch DDP multi-gpu experiments. It launches the required number of processes automatically under the hood as explained in the docs. However, there is currently only a hacky way for the user to prevent this if they wish to launch all processes manually through the command line or with utilities like torch.distributed.launch. This PR adds detection of LOCAL_RANK variable in the environment and determines accordingly if processes need to be launched or not.

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 update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is free to review the PR once the tests have passed.
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?

I made sure I had fun coding 🙃

[codecov]

Codecov Report

Merging #7480 (bb5964f) into master (6ce77a1) will decrease coverage by 0%.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master   #7480   +/-   ##
======================================
- Coverage      93%     92%   -0%     
======================================
  Files         216     216           
  Lines       14115   14112    -3     
======================================
- Hits        13088   13017   -71     
- Misses       1027    1095   +68     

[tchaton]

Nice :)

[stale]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. If you need further help see our docs: https://pytorch-lightning.readthedocs.io/en/latest/generated/CONTRIBUTING.html#pull-request or ask the assistance of a core contributor here or on Slack. Thank you for your contributions.

[carmocca]

LGTM. Some comments:

Can you merge master? To make sure none of the recent changes have any impact.
Should we add a special test using torch.distributed.launch?

[awaelchli]

@carmocca adding a test for torch.distributed.launch/run would not really cover anything in this PR, as this would anyway invoke the torchelastic environment. What this PR is about is the manual way to launch individual processes (say in different terminals or a script).

I added such calls to the special_tests.sh file, please have a look.

It adds about 10s of special test time.

[tchaton]

LGTM !

[awaelchli]

Yesterday GPU tests passed flawlessly. Of course it fails today when I get reviews.
RIP
my prs are cursed, always.