[SPARK-46700][CORE] Count the last spilling for the shuffle disk spilling bytes metric #44709
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
cloud-fan
commented
Jan 12, 2024
| * @param isLastFile if true, this indicates that we're writing the final output file and that the | ||
| * bytes written should be counted towards shuffle spill metrics rather than | ||
| * shuffle write metrics. | ||
| * @param isFinalFile if true, this indicates that we're writing the final output file and that |
There was a problem hiding this comment.
Rename the flag to be more accurate. It doesn't mean the last spill file, but should be the only spill file so that it will be used as the final shuffle output file.
cloud-fan
commented
Jan 12, 2024
| * bytes written should be counted towards shuffle spill metrics rather than | ||
| * shuffle write metrics. | ||
| * @param isFinalFile if true, this indicates that we're writing the final output file and that | ||
| * the bytes written should be counted towards shuffle write metrics rather |
There was a problem hiding this comment.
the comment was wrong before. If this flag is true, we are writing the final shuffle output file and will increase the shuffle write metrics rather than the spill metrics.
cloud-fan
commented
Jan 12, 2024
| private void writeSortedFile(boolean isFinalFile) { | ||
| // Only emit the log if this is an actual spilling. | ||
| if (!isFinalFile) { | ||
| logger.info( |
There was a problem hiding this comment.
move the logging here so that it applies to the last spilling as well.
cloud-fan
commented
Jan 12, 2024
| writeSortedFile(true); | ||
| // Here we are spilling the remaining data in the buffer. If there is no spill before, this | ||
| // final spill file will be the final shuffle output file. | ||
| writeSortedFile(/* isFinalFile = */spills.isEmpty()); |
There was a problem hiding this comment.
This is the actual change of this PR. We should only set this flag if we have not spilled before.
cloud-fan
commented
Jan 12, 2024
| // final write as bytes spilled (instead, it's accounted as shuffle write). The merge needs | ||
| // to be counted as shuffle write, but this will lead to double-counting of the final | ||
| // SpillInfo's bytes. | ||
| writeMetrics.decBytesWritten(spills[spills.length - 1].file.length()); |
There was a problem hiding this comment.
This hack is not needed anymore.
cloud-fan
commented
Jan 12, 2024
| @@ -425,9 +434,8 @@ private void testMergingSpills( | |||
| assertSpillFilesWereCleanedUp(); | |||
| ShuffleWriteMetrics shuffleWriteMetrics = taskMetrics.shuffleWriteMetrics(); | |||
| assertEquals(dataToWrite.size(), shuffleWriteMetrics.recordsWritten()); | |||
| assertTrue(taskMetrics.diskBytesSpilled() > 0L); | |||
| assertTrue(taskMetrics.diskBytesSpilled() < mergedOutputFile.length()); | |||
There was a problem hiding this comment.
The previous test was simply too relaxed.
|
cc @mridulm |
gengliangwang
approved these changes
Jan 12, 2024
dongjoon-hyun
pushed a commit
that referenced
this pull request
Jan 12, 2024
…ling bytes metric ### What changes were proposed in this pull request? This PR fixes a long-standing bug in ShuffleExternalSorter about the "spilled disk bytes" metrics. When we close the sorter, we will spill the remaining data in the buffer, with a flag `isLastFile = true`. This flag means the spilling will not increase the "spilled disk bytes" metrics. This makes sense if the sorter has never spilled before, then the final spill file will be used as the final shuffle output file, and we should keep the "spilled disk bytes" metrics as 0. However, if spilling did happen before, then we simply miscount the final spill file for the "spilled disk bytes" metrics today. This PR fixes this issue, by setting that flag when closing the sorter only if this is the first spilling. ### Why are the changes needed? make metrics accurate ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? updated tests ### Was this patch authored or co-authored using generative AI tooling? no Closes #44709 from cloud-fan/shuffle. Authored-by: Wenchen Fan <[email protected]> Signed-off-by: Dongjoon Hyun <[email protected]> (cherry picked from commit 4ea3742) Signed-off-by: Dongjoon Hyun <[email protected]>
dongjoon-hyun
pushed a commit
that referenced
this pull request
Jan 12, 2024
…ling bytes metric ### What changes were proposed in this pull request? This PR fixes a long-standing bug in ShuffleExternalSorter about the "spilled disk bytes" metrics. When we close the sorter, we will spill the remaining data in the buffer, with a flag `isLastFile = true`. This flag means the spilling will not increase the "spilled disk bytes" metrics. This makes sense if the sorter has never spilled before, then the final spill file will be used as the final shuffle output file, and we should keep the "spilled disk bytes" metrics as 0. However, if spilling did happen before, then we simply miscount the final spill file for the "spilled disk bytes" metrics today. This PR fixes this issue, by setting that flag when closing the sorter only if this is the first spilling. ### Why are the changes needed? make metrics accurate ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? updated tests ### Was this patch authored or co-authored using generative AI tooling? no Closes #44709 from cloud-fan/shuffle. Authored-by: Wenchen Fan <[email protected]> Signed-off-by: Dongjoon Hyun <[email protected]> (cherry picked from commit 4ea3742) Signed-off-by: Dongjoon Hyun <[email protected]>
|
Merged to master/3.5/3.4. Thank you, @cloud-fan and @gengliangwang . |
szehon-ho
pushed a commit
to szehon-ho/spark
that referenced
this pull request
Feb 7, 2024
…ling bytes metric ### What changes were proposed in this pull request? This PR fixes a long-standing bug in ShuffleExternalSorter about the "spilled disk bytes" metrics. When we close the sorter, we will spill the remaining data in the buffer, with a flag `isLastFile = true`. This flag means the spilling will not increase the "spilled disk bytes" metrics. This makes sense if the sorter has never spilled before, then the final spill file will be used as the final shuffle output file, and we should keep the "spilled disk bytes" metrics as 0. However, if spilling did happen before, then we simply miscount the final spill file for the "spilled disk bytes" metrics today. This PR fixes this issue, by setting that flag when closing the sorter only if this is the first spilling. ### Why are the changes needed? make metrics accurate ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? updated tests ### Was this patch authored or co-authored using generative AI tooling? no Closes apache#44709 from cloud-fan/shuffle. Authored-by: Wenchen Fan <[email protected]> Signed-off-by: Dongjoon Hyun <[email protected]> (cherry picked from commit 4ea3742) Signed-off-by: Dongjoon Hyun <[email protected]>
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?
This PR fixes a long-standing bug in ShuffleExternalSorter about the "spilled disk bytes" metrics. When we close the sorter, we will spill the remaining data in the buffer, with a flag
isLastFile = true. This flag means the spilling will not increase the "spilled disk bytes" metrics. This makes sense if the sorter has never spilled before, then the final spill file will be used as the final shuffle output file, and we should keep the "spilled disk bytes" metrics as 0. However, if spilling did happen before, then we simply miscount the final spill file for the "spilled disk bytes" metrics today.This PR fixes this issue, by setting that flag when closing the sorter only if this is the first spilling.
Why are the changes needed?
make metrics accurate
Does this PR introduce any user-facing change?
no
How was this patch tested?
updated tests
Was this patch authored or co-authored using generative AI tooling?
no