[#1341] fix(mr): Fix MR Combiner ArrayIndexOutOfBoundsException Bug. #1666
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
|
@sunshineJK Would you please review this PR? |
Test Results 2 384 files + 84 2 384 suites +84 4h 38m 28s ⏱️ + 37m 10s Results for commit 76e15b8. ± Comparison against base commit 45ad0b8. This pull request removes 1 and adds 3 tests. Note that renamed tests count towards both.♻️ This comment has been updated with latest results. |
|
Could you a ut for this case? |
|
@qijiale76 hi, i made some tests and found out that this PR works fine with input datas-212.txt (212MB) or any file size <= 212MB, but failed with datas-213.txt (213MB ). org.apache.hadoop.mapred.MapTask$NewOutputCollector@66971f6b
org.apache.uniffle.common.exception.RssException: Can't support sequence [4097], the max value should be 4095
at org.apache.hadoop.mapreduce.RssMRUtils.getBlockId(RssMRUtils.java:237)
at org.apache.hadoop.mapred.SortWriteBufferManager.createShuffleBlock(SortWriteBufferManager.java:390)other info: hadoop client: hadoop-3.2.1
nodemanager: hadoop-3.2.1
rss: master branch (Apri 19, 2024) with or without PR 1666 |
|
@qijiale76 Can you add UT? |
zhengchenyu
reviewed
Apr 29, 2024
client-mr/core/src/main/java/org/apache/hadoop/mapred/SortWriteBuffer.java
Outdated
Show resolved
Hide resolved
client-mr/core/src/main/java/org/apache/hadoop/mapred/SortWriteBuffer.java
Outdated
Show resolved
Hide resolved
|
LGTM+1 |
zhengchenyu
approved these changes
Apr 29, 2024
|
This pull request will copy data one more time. Will it only copy data when we don't use the combine? |
|
OK, I got it. |
jerqi
approved these changes
Apr 30, 2024
|
Merged to master & branch 0.9 . Thanks @qijiale76 @Lobo2008 @zhengchenyu |
jerqi
pushed a commit
that referenced
this pull request
Apr 30, 2024
…1666) ### What changes were proposed in this pull request? The current implementation of the SortBufferIterator.getKey() and .getValue() methods in the SortWriteBuffer class assumes that keys and values are always stored within a single buffer. This assumption can lead to runtime exceptions, specifically ArrayIndexOutOfBoundsException, when keys or values span multiple WrappedBuffer instances. In the current implementation, due to the execution of the `compact()`, the data of the key will not span buffers. However, the data of the value may be located on different buffers from the key and may span multiple buffers. This PR update getKey() to use fetchDataFromBuffers() to retrieve the key data. And this PR update getValue() to first adjust the starting index and offset based on the length of the key and then use fetchDataFromBuffers() to retrieve the value data. Another solution to this bug is to modify `addRecord()` method by keeping key and value in the same WrappedBuffer. ### Why are the changes needed? Fix: #1341 ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Tested manually. I'll submit a new integration test later.
zhengchenyu
pushed a commit
to zhengchenyu/incubator-uniffle
that referenced
this pull request
Jun 13, 2024
… Bug. (apache#1666) ### What changes were proposed in this pull request? The current implementation of the SortBufferIterator.getKey() and .getValue() methods in the SortWriteBuffer class assumes that keys and values are always stored within a single buffer. This assumption can lead to runtime exceptions, specifically ArrayIndexOutOfBoundsException, when keys or values span multiple WrappedBuffer instances. In the current implementation, due to the execution of the `compact()`, the data of the key will not span buffers. However, the data of the value may be located on different buffers from the key and may span multiple buffers. This PR update getKey() to use fetchDataFromBuffers() to retrieve the key data. And this PR update getValue() to first adjust the starting index and offset based on the length of the key and then use fetchDataFromBuffers() to retrieve the value data. Another solution to this bug is to modify `addRecord()` method by keeping key and value in the same WrappedBuffer. ### Why are the changes needed? Fix: apache#1341 ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Tested manually. I'll submit a new integration test later.
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.
What changes were proposed in this pull request?
The current implementation of the SortBufferIterator.getKey() and .getValue() methods in the SortWriteBuffer class assumes that keys and values are always stored within a single buffer. This assumption can lead to runtime exceptions, specifically ArrayIndexOutOfBoundsException, when keys or values span multiple WrappedBuffer instances.
In the current implementation, due to the execution of the
compact(), the data of the key will not span buffers. However, the data of the value may be located on different buffers from the key and may span multiple buffers.This PR update getKey() to use fetchDataFromBuffers() to retrieve the key data. And this PR update getValue() to first adjust the starting index and offset based on the length of the key and then use fetchDataFromBuffers() to retrieve the value data.
Another solution to this bug is to modify
addRecord()method by keeping key and value in the same WrappedBuffer.Why are the changes needed?
Fix: #1341
Does this PR introduce any user-facing change?
No.
How was this patch tested?
Tested manually. I'll submit a new integration test later.