perf(arrow-select): gc dictionaries in interleave_fallback_dictionary#9940
Open
asubiotto wants to merge 1 commit intoapache:mainfrom
Open
perf(arrow-select): gc dictionaries in interleave_fallback_dictionary#9940asubiotto wants to merge 1 commit intoapache:mainfrom
asubiotto wants to merge 1 commit intoapache:mainfrom
Conversation
interleave_dictionaries checks if input dictionaries should be merged/GCed, but should_merge_dictionary_values has limited type support due to the bytes interning it does. For other types and in the general fallback case where the heuristic fails, interleave_fallback_dictionary concatenates all the values slices together, resulting in a lot of bloat in cases where the interleave selection is minimal. This happens a lot e.g. in datafusion on a multi-partition sort on a dictionary column. This commit improves these cases by introducing a heuristic to check interleave index coverage assuming uniform value selection. When coverage is less than the number of values, interleave_fallback_dictionary now performs a `take` on the value slices in order to reduce output size bloat. On a real-world datafusion sort that motivated this change, I saw runtime drop from 20 minutes to 7 minutes. The heuristic allows us to avoid a microbenchmark regression. Signed-off-by: Alfonso Subiotto Marques <alfonso.subiotto@polarsignals.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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Which issue does this PR close?
Rationale for this change
interleave_dictionaries checks if input dictionaries should be merged/GCed, but should_merge_dictionary_values has limited type support due to the bytes interning it does. For other types and in the general fallback case where the heuristic fails, interleave_fallback_dictionary concatenates all the values slices together, resulting in a lot of bloat in cases where the interleave selection is minimal. This happens a lot e.g. in datafusion on a multi-partition sort on a dictionary column.
What changes are included in this PR?
This commit improves these cases by introducing a heuristic to check interleave index coverage assuming uniform value selection. When coverage is less than the number of values, interleave_fallback_dictionary now performs a
takeon the value slices in order to reduce output size bloat. On a real-world datafusion sort that motivated this change, I saw runtime drop from 20 minutes to 7 minutes. The heuristic allows us to avoid a microbenchmark regression.Are these changes tested?
Yes, by existing tests and some new tests exercising this path specifically.
Are there any user-facing changes?