Fix panic in median "AggregateState is not a scalar aggregate" #4488
Conversation
This file contains 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
github-actions
bot
added
core
alamb
commented
Dec 2, 2022
| @@ -436,6 +436,90 @@ async fn median_test( | |||
| Ok(()) | |||
| } | |||
|
|
|||
| #[tokio::test] | |||
| // test case for https://github.com/apache/arrow-datafusion/issues/3105 | |||
There was a problem hiding this comment.
all of the tests added in this file panic on master
tustvold
approved these changes
Dec 8, 2022
| @@ -720,7 +720,7 @@ impl std::hash::Hash for ScalarValue { | |||
| /// dictionary array | |||
| #[inline] | |||
| fn get_dict_value<K: ArrowDictionaryKeyType>( | |||
| array: &ArrayRef, | |||
| array: &dyn Array, | |||
Co-authored-by: Raphael Taylor-Davies <[email protected]>
alamb
commented
Dec 9, 2022
| @@ -216,6 +216,70 @@ SELECT approx_median(a) FROM median_f64_nan | |||
| ---- | |||
| NaN | |||
|
|
|||
| # median_multi | |||
There was a problem hiding this comment.
I ported the tests to sqllogictest as much of the rest of the aggregate tests had been ported too
jonmmease
pushed a commit
to jonmmease/arrow-datafusion
that referenced
this pull request
Dec 20, 2022
…e#4488) * Fix panic in median "AggregateState is not a scalar aggregate" * Apply suggestions from code review Co-authored-by: Raphael Taylor-Davies <[email protected]> Co-authored-by: Raphael Taylor-Davies <[email protected]> (cherry picked from commit 31bbe6c)
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Outdated suggestions cannot be applied.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.
Which issue does this PR close?
closes #3105
Rationale for this change
I think median as currently implemented is unusable except for everything but the simplest of queries (where there is no intermediate repartitioning)
Median was the only aggregate that stored it state using
AggregateState::Arraywhile all other aggregages useAggregateState::ScalarIt turns out that
AggregateState::Arraydoesn't support partial aggregates where partial state has to be marshalled into anArrayand combined. Thus,medianlikely never worked for any DataFusion plan that has more than one partition (where the data is repartitioned for parallel aggregation).What changes are included in this PR?
I tried a few ways to implement partial state for
AggregateState::Arrayand they all got messy very quickly.Thus, I opted for what I think is the simpler, possibly less performant, approach of using Scalars rather than Arrays.
Median was added in #3009 by @andygrove (and sadly I think I have have suggested the array based approach without fully understanding the implications)
For any real world dataset where the performance different between approaches would be measurable, I believe the the current implementation is going to error anyways.
This approach will also extend nicely to Decimal as well
Are these changes tested?
Yes
Are there any user-facing changes?
Less error
Proposed follow on
If this approach is accepted, I propose removing the AggregateState enum and update the Accumulator trait to simplify it as a follow on PR.
cc @andygrove as he originally wrote this code in #3009
cc @tustvold as I think he is thinking about how to improve the grouping situation overall