[Tune] Dynamically identify PyTorch Lightning Callback hooks

[amogkam]

Our current PTL integration Callback hooks are hard coded and contain deprecated hooks which log a warning: #27487.

Some of these hooks have already been deprecated in PTL 1.6, others are deprecated in PTL 1.7, and others in 1.8. Instead of maintaining a different PTL integration callback for each PTL version, we instead dynamically identify the available hooks for whatever PTL version the user has installed.

Closes #27487
Closes https://github.com/anyscale/product/issues/14165

Why are these changes needed?

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[Yard1]

Can we test this somehow?

[amogkam]

We already have tests for this @Yard1

[Yard1]

Would it be possible to test the new logic introduced in this PR, though? Eg. use runtime envs to install 2 PTL versions and see that the hooks were detected correctly?

[amogkam]

Hmm but the logic introduced in this pr is already being tested right? There’s no version specific logic introduced in this pr.

Should we be testing all of our integrations against all external library versions?

[Yard1]

I suppose that's true, we don't really have a precedent here. I just find it a bit worrying that the intention of this PR is to make the callback version-independent but we have no way of checking if that was achieved

[amogkam]

I agree, but to be fair, the intention of the callback in general (even before this PR) was also to be version independent.

Ideally, there would be a good way to test this, but we end up having to test against lots of different versions.

[Yard1]

Yeah, that's understandable. The logic looks good to me!

[amogkam]

Failing test is flaky on master, going to merge