[SPARK-43393][SQL][3.3] Address sequence expression overflow bug. #43821
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
Member
|
Is GitHub Action triggered on this PR, @thepinetree ? |
Contributor
Author
|
@dongjoon-hyun I think this PR is good to go too, not sure why github actions thinks the tests haven't finished (like the others, only the documentation build had issues) |
Member
|
Thank you, @thepinetree . |
dongjoon-hyun
pushed a commit
that referenced
this pull request
Nov 20, 2023
### What changes were proposed in this pull request?
Spark has a (long-standing) overflow bug in the `sequence` expression.
Consider the following operations:
```
spark.sql("CREATE TABLE foo (l LONG);")
spark.sql(s"INSERT INTO foo VALUES (${Long.MaxValue});")
spark.sql("SELECT sequence(0, l) FROM foo;").collect()
```
The result of these operations will be:
```
Array[org.apache.spark.sql.Row] = Array([WrappedArray()])
```
an unintended consequence of overflow.
The sequence is applied to values `0` and `Long.MaxValue` with a step size of `1` which uses a length computation defined [here](https://github.com/apache/spark/blob/16411188c7ba6cb19c46a2bd512b2485a4c03e2c/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/collectionOperations.scala#L3451). In this calculation, with `start = 0`, `stop = Long.MaxValue`, and `step = 1`, the calculated `len` overflows to `Long.MinValue`. The computation, in binary looks like:
```
0111111111111111111111111111111111111111111111111111111111111111
- 0000000000000000000000000000000000000000000000000000000000000000
------------------------------------------------------------------
0111111111111111111111111111111111111111111111111111111111111111
/ 0000000000000000000000000000000000000000000000000000000000000001
------------------------------------------------------------------
0111111111111111111111111111111111111111111111111111111111111111
+ 0000000000000000000000000000000000000000000000000000000000000001
------------------------------------------------------------------
1000000000000000000000000000000000000000000000000000000000000000
```
The following [check](https://github.com/apache/spark/blob/16411188c7ba6cb19c46a2bd512b2485a4c03e2c/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/collectionOperations.scala#L3454) passes as the negative `Long.MinValue` is still `<= MAX_ROUNDED_ARRAY_LENGTH`. The following cast to `toInt` uses this representation and [truncates the upper bits](https://github.com/apache/spark/blob/16411188c7ba6cb19c46a2bd512b2485a4c03e2c/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/collectionOperations.scala#L3457) resulting in an empty length of `0`.
Other overflows are similarly problematic.
This PR addresses the issue by checking numeric operations in the length computation for overflow.
### Why are the changes needed?
There is a correctness bug from overflow in the `sequence` expression.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Tests added in `CollectionExpressionsSuite.scala`.
Closes #43821 from thepinetree/spark-sequence-overflow-3.3.
Authored-by: Deepayan Patra <[email protected]>
Signed-off-by: Dongjoon Hyun <[email protected]>
Member
|
Merged to branch-3.3 for Apache Spark 3.3.4. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
What changes were proposed in this pull request?
Spark has a (long-standing) overflow bug in the
sequenceexpression.Consider the following operations:
The result of these operations will be:
an unintended consequence of overflow.
The sequence is applied to values
0andLong.MaxValuewith a step size of1which uses a length computation defined here. In this calculation, withstart = 0,stop = Long.MaxValue, andstep = 1, the calculatedlenoverflows toLong.MinValue. The computation, in binary looks like:The following check passes as the negative
Long.MinValueis still<= MAX_ROUNDED_ARRAY_LENGTH. The following cast totoIntuses this representation and truncates the upper bits resulting in an empty length of0.Other overflows are similarly problematic.
This PR addresses the issue by checking numeric operations in the length computation for overflow.
Why are the changes needed?
There is a correctness bug from overflow in the
sequenceexpression.Does this PR introduce any user-facing change?
No.
How was this patch tested?
Tests added in
CollectionExpressionsSuite.scala.