[dai123x]

What does this PR do?

Fixes #9136.

AdvancedProfiler.teardown() cleared profiled_actions unconditionally. If a profiling context was still active when teardown ran, its action had already been dropped from the dict, so the context's finally block then called stop() on an action that no longer existed:

ValueError: Attempting to stop recording an action (run_test_evaluation) which was never started.

That is reachable in normal use: the profiler is torn down at the end of a stage while a profile(...) block around the test evaluation can still be unwinding.

Changes

• teardown() disables any still-running profiles before clearing the dict, so nothing is left enabled once the profiler goes away.
• stop() logs at debug level and returns instead of raising when the action was never started. This path is reachable from teardown() by construction, so the exception was never actionable for the caller.
• Removes the now-inaccurate Raises: entry from the AdvancedProfiler docstring.
• CHANGELOG entry added under the unreleased fixes.

Tests

Two regression tests in tests/tests_pytorch/profilers/test_profiler.py:

• test_advanced_profiler_stop_unstarted_action - stop() on an action that was never started returns without raising.
• test_advanced_profiler_teardown_disables_active_action - an action still active at teardown() is disabled and cleared.

What I ran and what I did not: I did not run the suite locally, because this workspace has no torch environment installed. Those two tests plus the existing profiler tests are what CI needs to run; if either fails I will fix it on this branch.

No breaking changes: the only behaviour removed is an exception that could not be recovered from and always followed a teardown.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79%. Comparing base (84df182) to head (1557458).
✅ All tests successful. No failed tests found.
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

❗ There is a different number of reports uploaded between BASE (84df182) and HEAD (1557458). Click for more details.

HEAD has 6067 uploads less than BASE

Flag	BASE (84df182)	HEAD (1557458)
cpu	1403	49
lightning_fabric	620	0
pytest	703	0
python3.12	583	21
python3.12.7	83	3
python3.13	570	19
python	83	3
lightning	166	5
python3.10	84	3
pytorch_lightning	617	44
pytorch2.6	124	9
pytest-full	700	49
pytorch2.12	83	5
pytorch2.7	42	3
pytorch2.9	81	6
pytorch2.8	81	6
pytorch2.13	123	8
pytorch2.10	83	6
pytorch2.11	83	6

Additional details and impacted files

@@            Coverage Diff            @@
##           master   #22005     +/-   ##
=========================================
- Coverage      87%      79%     -8%     
=========================================
  Files         270      267      -3     
  Lines       24015    23959     -56     
=========================================
- Hits        20855    18827   -2028     
- Misses       3160     5132   +1972     

[dai123x]

The previous Test PyTorch run had two matrix jobs cancelled externally mid-run (no test failures — both were past 74% and passing), which failed pl-cpu-guardian and required-jobs. The docs linkcheck failures are pre-existing NVIDIA docs 404s, also failing on master. The newly pushed retrigger commit (ac08704) is held at action_required — could someone approve and re-run the workflows? Thanks!

[dai123x]

Gentle nudge on this one. All six workflows on the current head (ac08704) are sitting at action_required, so nothing has actually run yet:

• Test PyTorch
• Code check
• Docs builds
• Package
• Check Schema
• Probot

Could someone approve the runs? For context on the earlier attempt: the previous Test PyTorch failure was two matrix jobs cancelled externally mid-run (both past 74%, no test failures), which took down pl-cpu-guardian and required-jobs; the docs linkcheck 404s are pre-existing NVIDIA URLs that also fail on master. The change itself is covered by tests and Codecov reports all modified lines covered.

Thanks!