[FLINK-31386][network] Fix the potential deadlock issue of blocking shuffle #22148
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
wsry
pushed a commit
to wsry/flink
that referenced
this pull request
Mar 10, 2023
…huffle This closes apache#22148. Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing.
wsry
force-pushed
the
release-1.17-FLINK-31386
branch
from
782cd9d to
e7de8c4
Compare
wsry
pushed a commit
to wsry/flink
that referenced
this pull request
Mar 10, 2023
…huffle Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing. This closes apache#22148.
wsry
force-pushed
the
release-1.17-FLINK-31386
branch
from
e7de8c4 to
ed627bf
Compare
reswqa
reviewed
Mar 10, 2023
Comment on lines
192
to
199
| try { | ||
| for (int i = 0; i < bufferPool.getNumberOfRequiredMemorySegments(); ++i) { | ||
| freeSegments.add(checkNotNull(bufferPool.requestMemorySegmentBlocking())); | ||
| } | ||
| } catch (Throwable throwable) { | ||
| freeSegments.forEach(bufferPool::recycle); | ||
| throw new IOException("Failed to reserve required buffers.", throwable); | ||
| } |
There was a problem hiding this comment.
Can we extract this logic to a method to reuse this, because requestNetworkBuffers also request buffers to guaranteed.
…huffle Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing. This closes apache#22148.
wsry
force-pushed
the
release-1.17-FLINK-31386
branch
from
ed627bf to
7b5e869
Compare
|
@reswqa Thanks for the review and feedback. I have updated the PR. |
wsry
pushed a commit
to wsry/flink
that referenced
this pull request
Mar 10, 2023
…huffle Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing. This closes apache#22148.
wsry
pushed a commit
that referenced
this pull request
Mar 11, 2023
…huffle Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing. This closes #22148.
reswqa
pushed a commit
to reswqa/flink
that referenced
this pull request
Mar 14, 2023
…huffle Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing. This closes apache#22148.
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 is the purpose of the change
Currently, the SortMergeResultPartition may allocate more network buffers than the guaranteed size of the LocalBufferPool. As a result, some result partitions may need to wait other result partitions to release the over-allocated network buffers to continue. However, the result partitions which have allocated more than guaranteed buffers relies on the processing of input data to trigger data spilling and buffer recycling. The input data further relies on batch reading buffers used by the SortMergeResultPartitionReadScheduler which may already taken by those blocked result partitions that are waiting for buffers. Then deadlock occurs. This patch fixes the deadlock issue by reserving the guaranteed buffers on initializing.
Brief change log
Verifying this change
This change added tests.
Does this pull request potentially affect one of the following parts:
@Public(Evolving): (yes / no)Documentation