[Data] Use application-level retries for Parquet metadata tasks

[bveeramani]

Why are these changes needed?

#42044 updated the ParquetDatasource logic so that Ray retries metadata fetching tasks that raise OSErrors.

ray/python/ray/data/datasource/parquet_datasource.py

Lines 295 to 298 in ba3cfa0

	if RETRY_EXCEPTIONS_FOR_META_FETCH_TASK is not None:
	prefetch_remote_args[
	"retry_exceptions"
	] = RETRY_EXCEPTIONS_FOR_META_FETCH_TASK

The problem is that we retry tasks indefinitely, so if a metadata fetching tasks raises a non-transient OSError, your program freezes without warning.

ray/python/ray/data/_internal/remote_fn.py

Line 26 in ba3cfa0

"max_retries": -1,

To fix this issue, this PR replaces the Ray Core-level retry with finite application-level retries. It also updates the retry logic to apply jitter to backoffs that exceed the limit.

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 added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• 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 :(

[c21]

Thanks @bveeramani! Having some comments.

[c21]

LGTM pending the test for the user workload.

[c21]

Error message LGTM