Skip to content

[FLINK-40519][network] Deliver priority barrier to a credited remote reader while the subpartition is blocked during recovery #29056

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

Open
1996fanrui wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[FLINK-40519][network] Deliver priority barrier to a credited remote reader while the subpartition is blocked during recovery #29056

Show file tree
Hide file tree
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
15 changes: 12 additions & 3 deletions ...me/src/main/java/org/apache/flink/runtime/io/network/partition/PipelinedSubpartition.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -472,7 +472,7 @@ BufferAndBacklog pollBuffer() {
// When blocked (e.g. by RECOVERY_COMPLETION event), only allow priority buffers
// (e.g. unaligned checkpoint barriers) to be polled. Regular buffers remain blocked
// until resumeConsumption() is called. See needNotifyPriorityEvent() for details.
if (isBlocked && buffers.getNumPriorityElements() == 0) {
if (isBlockedForDelivery()) {
return null;
}

Expand Down Expand Up @@ -619,11 +619,20 @@ public ResultSubpartitionView.AvailabilityWithBacklog getAvailabilityAndBacklog(
}
}

/**
* Blocked for delivery when blocked (e.g. by RECOVERY_COMPLETION) with no priority element
* queued; priority buffers (e.g. unaligned barriers) are still delivered while blocked.
*/
@GuardedBy("buffers")
private boolean isBlockedForDelivery() {
return isBlocked && buffers.getNumPriorityElements() == 0;
}

@GuardedBy("buffers")
private boolean isDataAvailableUnsafe() {
assert Thread.holdsLock(buffers);

return !isBlocked && (flushRequested || getNumberOfFinishedBuffers() > 0);
return !isBlockedForDelivery() && (flushRequested || getNumberOfFinishedBuffers() > 0);
}

private Buffer.DataType getNextBufferTypeUnsafe() {
Expand Down Expand Up @@ -751,7 +760,7 @@ private void increaseBuffersInBacklog(BufferConsumer buffer) {
@SuppressWarnings("FieldAccessNotGuarded")
@Override
public int getBuffersInBacklogUnsafe() {
if (isBlocked || buffers.isEmpty()) {
if (isBlockedForDelivery() || buffers.isEmpty()) {
return 0;
}

Expand Down
30 changes: 30 additions & 0 deletions ...rc/test/java/org/apache/flink/runtime/io/network/partition/PipelinedSubpartitionTest.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@
import org.apache.flink.runtime.io.network.buffer.Buffer;
import org.apache.flink.runtime.io.network.buffer.BufferBuilderTestUtils;
import org.apache.flink.runtime.io.network.buffer.BufferConsumer;
import org.apache.flink.runtime.io.network.partition.consumer.EndOfOutputChannelStateEvent;
import org.apache.flink.runtime.io.network.util.TestConsumerCallback;
import org.apache.flink.runtime.io.network.util.TestProducerSource;
import org.apache.flink.runtime.io.network.util.TestSubpartitionConsumer;
Expand Down Expand Up @@ -472,6 +473,35 @@ void testConcurrentTimeoutableCheckpointBarrier() throws Exception {
.isInstanceOf(ExecutionException.class);
}

@TestTemplate
void testPriorityBarrierAvailableToCreditedReaderWhileBlocked() throws Exception {
PipelinedSubpartition subpartition = createSubpartition();
subpartition.setChannelStateWriter(ChannelStateWriter.NO_OP);

// Block the subpartition, mirroring the RECOVERY_COMPLETION event emitted during recovery.
subpartition.add(
EventSerializer.toBufferConsumer(EndOfOutputChannelStateEvent.INSTANCE, false));
pollBufferAndCheckType(subpartition, Buffer.DataType.RECOVERY_COMPLETION);

// While blocked and without any priority element, a credited reader sees no data.
assertThat(subpartition.getAvailabilityAndBacklog(true).isAvailable()).isFalse();
assertThat(subpartition.pollBuffer()).isNull();

// Enqueue an unaligned checkpoint barrier as a priority element.
CheckpointOptions options =
CheckpointOptions.unaligned(
CheckpointType.CHECKPOINT, CheckpointStorageLocationReference.getDefault());
subpartition.add(
EventSerializer.toBufferConsumer(
new CheckpointBarrier(1L, System.currentTimeMillis(), options), true));

// The credited-reader availability check must now report available so the remote reader is
// enqueued and the priority barrier is delivered even though the subpartition
// stays blocked.
assertThat(subpartition.getAvailabilityAndBacklog(true).isAvailable()).isTrue();
pollBufferAndCheckType(subpartition, Buffer.DataType.PRIORITIZED_EVENT_BUFFER);
}

private BufferConsumer getTimeoutableBarrierBuffer(long checkpointId) throws IOException {
CheckpointOptions checkpointOptions =
CheckpointOptions.alignedWithTimeout(
Expand Down