feat: Support ANSI mode sum expr (int inputs) #2600
+815
−53
Conversation
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
Contributor
Author
|
Draft PR to support sum function - WIP |
coderfender
changed the title
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #2600 +/- ##
============================================
+ Coverage 56.12% 59.41% +3.28%
- Complexity 976 1377 +401
============================================
Files 119 167 +48
Lines 11743 15330 +3587
Branches 2251 2546 +295
============================================
+ Hits 6591 9108 +2517
- Misses 4012 4938 +926
- Partials 1140 1284 +144 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
martin-g
reviewed
Nov 5, 2025
coderfender
force-pushed
the
support_overflow_sum_function
branch
from
970d693 to
c0715aa
Compare
Contributor
Author
|
@andygrove , @comphead I believe this should be ready for review (pending CI) |
martin-g
reviewed
Nov 7, 2025
spark/src/test/scala/org/apache/comet/CometExpressionSuite.scala
Outdated
Show resolved
Hide resolved
spark/src/main/scala/org/apache/comet/serde/QueryPlanSerde.scala
Outdated
Show resolved
Hide resolved
Contributor
Thats actually interesting why would window fall back, we planning to fallback on windows in #2726 but the test would still preserve windows, so I think we need to check a fallback reason |
coderfender
force-pushed
the
support_overflow_sum_function
branch
from
6dcf47b to
13b7d68
Compare
Contributor
Author
This is no longer a failing test after I rebased with main branch |
Contributor
Author
|
@andygrove @comphead . All the tests passed and the PR is rest for review |
andygrove
reviewed
Nov 20, 2025
Member
andygrove
left a comment
andygrove
left a comment
There was a problem hiding this comment.
@coderfender This LGTM. I am going to pull the PR locally and do some testing/benchmarking today. Could you rebease/upmerge?
andygrove
reviewed
Nov 20, 2025
spark/src/main/scala/org/apache/comet/serde/QueryPlanSerde.scala
Outdated
Show resolved
Hide resolved
andygrove
reviewed
Nov 20, 2025
andygrove
reviewed
Nov 20, 2025
Member
|
Here are microbenchmark results: Legacy modeANSI mode |
andygrove
reviewed
Nov 20, 2025
andygrove
reviewed
Nov 20, 2025
andygrove
reviewed
Nov 20, 2025
spark/src/test/scala/org/apache/comet/CometExpressionSuite.scala
Outdated
Show resolved
Hide resolved
Contributor
Author
|
@andygrove , Please take a look whenever you get a chance . I made changes per review and fixed merge conflicts |
coderfender
changed the title
Contributor
Author
|
@andygrove I rebased with main and fixed merge conflicts . Please review whenever you get a chance |
Contributor
Author
|
Resolved scalafix error |
37 tasks
Contributor
Author
|
@andygrove , @martin-g could you please take a (hopefully final) look at the PR and let me know if you think we need further changes ? |
Member
The changes in the PR are pretty substantial (~800 new lines of code) so there is a lot to review. I don't see anything that concerns me at this point, but I would like to run some benchmarks to check that there are no regressions in the non ANSI case. Unfortunately, we don't have any existing microbenchmarks that work with ANSI enabled, so we need to create something. |
Contributor
Author
|
Thank you @andygrove let me go ahead and work on generating benchmarks for ANSI supported aggregate expressions |
Contributor
Author
|
Created #2883 to support benchmarking with ANSI mode |
Contributor
Author
|
Ran benchmarking with ANSI mode enabled for SUM aggregated function only and here are the results |
Contributor
Author
|
Rebased with main |
andygrove
reviewed
Dec 16, 2025
|
|
||
| // Check for overflow for early termination | ||
| if self.eval_mode == EvalMode::Try { | ||
| let that_has_all_nulls = states[1].as_boolean().value(0); |
Member
There was a problem hiding this comment.
DF docs for merge_batch state:
/// Updates the accumulator's state from an `Array` containing one
/// or more intermediate values.
should there be a check that the input has more than one element before accessing state[1]?
Contributor
Author
There was a problem hiding this comment.
I think the check here might be an overkill. We are guaranteeing that the accumulator's state is holding sum and has_all_nulls for Try eval mode as per Spark's semantics :
andygrove
reviewed
Dec 16, 2025
| self.resize_helper(total_num_groups); | ||
|
|
||
| let that_sums_is_all_nulls = if self.eval_mode == EvalMode::Try { | ||
| Some(values[1].as_boolean()) |
Member
There was a problem hiding this comment.
Same question as earlier. Is values guaranteed to have length > 1 at this point?
Contributor
Author
There was a problem hiding this comment.
I think the check here might be an overkill. We are guaranteeing that the group accumulator's state is holding sum and has_all_nulls for Try eval mode as per Spark's semantics
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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 #531 .
(Partially closes 531) . The ANSI changes to support
AVGwill be tracked in a new PRRationale for this change
DataFusion's default SUM doesn't match Spark's overflow semantics. This implementation ensures we get the exact same behavior as Spark for integer overflow in all 3 eval modes (ANSI , Legacy and Try mode)
What changes are included in this PR?
This PR adds native Rust implementation for SUM aggregation on integer types (byte, short, int, long)
Native changes (Rust):
(Inspired from
SumDecimaland spark's SUM impl)SumIntegeraggregate function that handles SUM for all integer types (in coherence with Spark)( Implemented code in similar fashion of spark leveraging
Option<i64>to represent NULL and numeric values for sum , and an additional parameter calledhas_all_nullswhich is leveraged in Try mode to distinguish if NULL sum is caused by all NULL inputs or the fact that the sum overflowed. (Spark does this withshouldTrackIsEmptyand assigning NULL to running sum which is a long datatype) )Scala side changes :
CometSumto add ANSI support (ANSI and Try)eval_modeinstead offail_on_errorto supportLegacy, ANSI and Tryeval modesjava.lang.Longto avoid Scala's auto-boxing feature which auto casts objects to primitive types there by casting nulls to 0s) handling in both simple and GROUP BY agg .How are these changes tested?