[SPARK-45678][CORE] Cover BufferReleasingInputStream.available/reset under tryOrFetchFailedException #43543
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
viirya
changed the title
mridulm
approved these changes
Oct 27, 2023
Ah, I missed it. Thanks @mridulm. |
viirya
changed the title
|
Thank you @mridulm :) |
sunchao
pushed a commit
that referenced
this pull request
Oct 28, 2023
…under tryOrFetchFailedException ### What changes were proposed in this pull request? This patch proposes to wrap `BufferReleasingInputStream.available/reset` under `tryOrFetchFailedException`. So `IOException` during `available`/`reset` call will be rethrown as `FetchFailedException`. ### Why are the changes needed? We have encountered shuffle data corruption issue: ``` Caused by: java.io.IOException: FAILED_TO_UNCOMPRESS(5) at org.xerial.snappy.SnappyNative.throw_error(SnappyNative.java:112) at org.xerial.snappy.SnappyNative.rawUncompress(Native Method) at org.xerial.snappy.Snappy.rawUncompress(Snappy.java:504) at org.xerial.snappy.Snappy.uncompress(Snappy.java:543) at org.xerial.snappy.SnappyInputStream.hasNextChunk(SnappyInputStream.java:450) at org.xerial.snappy.SnappyInputStream.available(SnappyInputStream.java:497) at org.apache.spark.storage.BufferReleasingInputStream.available(ShuffleBlockFetcherIterator.scala:1356) ``` Spark shuffle has capacity to detect corruption for a few stream op like `read` and `skip`, such `IOException` in the stack trace will be rethrown as `FetchFailedException` that will re-try the failed shuffle task. But in the stack trace it is `available` that is not covered by the mechanism. So no-retry has been happened and the Spark application just failed. As the `available`/`reset` op will also involve data decompression and throw `IOException`, we should be able to check it like `read` and `skip` do. ### Does this PR introduce _any_ user-facing change? Yes. Data corruption during `available`/`reset` op is now causing `FetchFailedException` like `read` and `skip` that can be retried instead of `IOException`. ### How was this patch tested? Added test. ### Was this patch authored or co-authored using generative AI tooling? No Closes #43543 from viirya/add_available. Authored-by: Liang-Chi Hsieh <[email protected]> Signed-off-by: Chao Sun <[email protected]>
sunchao
pushed a commit
that referenced
this pull request
Oct 28, 2023
…under tryOrFetchFailedException ### What changes were proposed in this pull request? This patch proposes to wrap `BufferReleasingInputStream.available/reset` under `tryOrFetchFailedException`. So `IOException` during `available`/`reset` call will be rethrown as `FetchFailedException`. ### Why are the changes needed? We have encountered shuffle data corruption issue: ``` Caused by: java.io.IOException: FAILED_TO_UNCOMPRESS(5) at org.xerial.snappy.SnappyNative.throw_error(SnappyNative.java:112) at org.xerial.snappy.SnappyNative.rawUncompress(Native Method) at org.xerial.snappy.Snappy.rawUncompress(Snappy.java:504) at org.xerial.snappy.Snappy.uncompress(Snappy.java:543) at org.xerial.snappy.SnappyInputStream.hasNextChunk(SnappyInputStream.java:450) at org.xerial.snappy.SnappyInputStream.available(SnappyInputStream.java:497) at org.apache.spark.storage.BufferReleasingInputStream.available(ShuffleBlockFetcherIterator.scala:1356) ``` Spark shuffle has capacity to detect corruption for a few stream op like `read` and `skip`, such `IOException` in the stack trace will be rethrown as `FetchFailedException` that will re-try the failed shuffle task. But in the stack trace it is `available` that is not covered by the mechanism. So no-retry has been happened and the Spark application just failed. As the `available`/`reset` op will also involve data decompression and throw `IOException`, we should be able to check it like `read` and `skip` do. ### Does this PR introduce _any_ user-facing change? Yes. Data corruption during `available`/`reset` op is now causing `FetchFailedException` like `read` and `skip` that can be retried instead of `IOException`. ### How was this patch tested? Added test. ### Was this patch authored or co-authored using generative AI tooling? No Closes #43543 from viirya/add_available. Authored-by: Liang-Chi Hsieh <[email protected]> Signed-off-by: Chao Sun <[email protected]>
|
+1, late LGTM. |
|
Thank you @dongjoon-hyun ! |
szehon-ho
pushed a commit
to szehon-ho/spark
that referenced
this pull request
Feb 7, 2024
…under tryOrFetchFailedException ### What changes were proposed in this pull request? This patch proposes to wrap `BufferReleasingInputStream.available/reset` under `tryOrFetchFailedException`. So `IOException` during `available`/`reset` call will be rethrown as `FetchFailedException`. ### Why are the changes needed? We have encountered shuffle data corruption issue: ``` Caused by: java.io.IOException: FAILED_TO_UNCOMPRESS(5) at org.xerial.snappy.SnappyNative.throw_error(SnappyNative.java:112) at org.xerial.snappy.SnappyNative.rawUncompress(Native Method) at org.xerial.snappy.Snappy.rawUncompress(Snappy.java:504) at org.xerial.snappy.Snappy.uncompress(Snappy.java:543) at org.xerial.snappy.SnappyInputStream.hasNextChunk(SnappyInputStream.java:450) at org.xerial.snappy.SnappyInputStream.available(SnappyInputStream.java:497) at org.apache.spark.storage.BufferReleasingInputStream.available(ShuffleBlockFetcherIterator.scala:1356) ``` Spark shuffle has capacity to detect corruption for a few stream op like `read` and `skip`, such `IOException` in the stack trace will be rethrown as `FetchFailedException` that will re-try the failed shuffle task. But in the stack trace it is `available` that is not covered by the mechanism. So no-retry has been happened and the Spark application just failed. As the `available`/`reset` op will also involve data decompression and throw `IOException`, we should be able to check it like `read` and `skip` do. ### Does this PR introduce _any_ user-facing change? Yes. Data corruption during `available`/`reset` op is now causing `FetchFailedException` like `read` and `skip` that can be retried instead of `IOException`. ### How was this patch tested? Added test. ### Was this patch authored or co-authored using generative AI tooling? No Closes apache#43543 from viirya/add_available. Authored-by: Liang-Chi Hsieh <[email protected]> Signed-off-by: Chao Sun <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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?
This patch proposes to wrap
BufferReleasingInputStream.available/resetundertryOrFetchFailedException. SoIOExceptionduringavailable/resetcall will be rethrown asFetchFailedException.Why are the changes needed?
We have encountered shuffle data corruption issue:
Spark shuffle has capacity to detect corruption for a few stream op like
readandskip, suchIOExceptionin the stack trace will be rethrown asFetchFailedExceptionthat will re-try the failed shuffle task. But in the stack trace it isavailablethat is not covered by the mechanism. So no-retry has been happened and the Spark application just failed.As the
available/resetop will also involve data decompression and throwIOException, we should be able to check it likereadandskipdo.Does this PR introduce any user-facing change?
Yes. Data corruption during
available/resetop is now causingFetchFailedExceptionlikereadandskipthat can be retried instead ofIOException.How was this patch tested?
Added test.
Was this patch authored or co-authored using generative AI tooling?
No