Skip to content

KAFKA-9703:Free up resources when splitting huge batches #8286

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
jiameixie wants to merge 1 commit into from
Closed

KAFKA-9703:Free up resources when splitting huge batches #8286

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 4 additions & 1 deletion clients/src/main/java/org/apache/kafka/clients/producer/internals/ProducerBatch.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -268,14 +268,17 @@ public Deque<ProducerBatch> split(int splitBatchSize) {
// A newly created batch can always host the first message.
if (!batch.tryAppendForSplit(record.timestamp(), record.key(), record.value(), record.headers(), thunk)) {
batches.add(batch);
batch.closeForRecordAppends();

@chia7712 chia7712 Mar 15, 2020

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

batch.closeForRecordAppends does NOT release the byte array hosted by MemoryRecordsBuilder

    public void closeForRecordAppends() {
        if (appendStream != CLOSED_STREAM) {
            try {
                appendStream.close();
            } catch (IOException e) {
                throw new KafkaException(e);
            } finally {
                appendStream = CLOSED_STREAM;
            }
        }
    }

appendStream is a wrap of bufferStream. Not sure whether calling batch.closeForRecordAppends() can resolve the OOM or not.

this.appendStream = new DataOutputStream(compressionType.wrapForOutput(this.bufferStream, magic));

@becketqin becketqin Mar 18, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@chia7712 Thanks for reporting the issue. But I'd like to understand this problem a bit more. Supposedly the split batches will be closed when they are drained out of the RecordAccumulator.
https://github.com/apache/kafka/blob/trunk/clients/src/main/java/org/apache/kafka/clients/producer/internals/RecordAccumulator.java#L608

and ProducerBatch#close() calls ProducerBatch#closeForRecordAppends().

Do you have a heap dump when the OOM happens so we can take a further look?

@jiameixie jiameixie Mar 18, 2020

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@becketqin In my case, the OOM is caused by negative estimatedCompressionRatio. I have committed #8285 to fix it.

@becketqin becketqin Mar 20, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We probably also want to close the last batch for append in line 279.

@jiameixie jiameixie Mar 20, 2020

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@becketqin My negligence! The last batch should also be closed. I have modified it.

batch = createBatchOffAccumulatorForRecord(record, splitBatchSize);
batch.tryAppendForSplit(record.timestamp(), record.key(), record.value(), record.headers(), thunk);
}
}

// Close the last batch and add it to the batch list after split.
if (batch != null)
if (batch != null) {
batches.add(batch);
batch.closeForRecordAppends();
}

produceFuture.set(ProduceResponse.INVALID_OFFSET, NO_TIMESTAMP, new RecordBatchTooLargeException());
produceFuture.done();
Expand Down