Fix profiling AutoTuner RAPIDS property lookup - #2152
Merged
amahussein merged 1 commit intoSep 9, 2026
Merged
Conversation
Fixes NVIDIA#2142 Include RAPIDS properties in the profiling provider's all-properties map so AutoTuner source lookups see values recorded in the event log. Update the three affected expectations, which previously encoded false missing-property comments. Test profiling RAPIDS property preservation Signed-off-by: Ahmed Hussein <ahussein@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #2142.
Profiling builds separate general Spark and RAPIDS property views, but
SingleAppSummaryInfoProvider.getAllPropertiesexposed only the general view. Since the AutoTuner uses that method as its source map, it treated configuredspark.rapids.*values as absent, emitted falsewas not setcomments, and ignored those values during tuning.This change returns both views. They are projections of the same application property map, so overlapping keys carry the same value.
Two behavior changes follow from the AutoTuner now seeing these values, and both are intended. Updated-only filtering now applies correctly to enabled, non-preserved RAPIDS settings, so a
--confline is omitted when the source value already matches the recommendation, consistent with other Spark properties. Event-log values for twospark.rapids.sql.hybrid.*flags can now participate in on-prem off-heap sizing and may change related memory recommendations; target-enforced values already worked.Three existing expected outputs are updated because they encoded the bug: the checked-in event log sets the RAPIDS properties associated with the removed comments, so the old passing assertions expected incorrect comments. Truly absent properties, including
spark.rapids.sql.enabledin this fixture, remain reported as unset. An existing target-cluster test is extended to assert that apreserveentry retains the RAPIDS value from the event log, which fails if the provider change is reverted.Testing:
mvn -Dbuildver=357 verify(992 tests passed)mvn -Dbuildver=357 -Pscala213 verify(992 tests passed)ClusterRecommendationSuiteunder Scala 2.12 and 2.13 (13 tests passed in each profile)spark_rapids profilingbefore/after replay overgpu_oom_eventlog.zstd(only false missing-property comments removed)