This is an automated email from the ASF dual-hosted git repository.
bowenli86 pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/flink-connector-kafka.git
The following commit(s) were added to refs/heads/main by this push:
new 904e8821 [FLINK-39979] DynamicKafkaSource should mark reader idle when
metadata removal leaves a subtask with no active subreaders (#276)
904e8821 is described below
commit 904e88213ac4988087499c5621fe5c4c2a6ca4f9
Author: bowenli86 <[email protected]>
AuthorDate: Fri Jun 26 09:08:49 2026 -0700
[FLINK-39979] DynamicKafkaSource should mark reader idle when metadata
removal leaves a subtask with no active subreaders (#276)
DynamicKafkaSource can remain active after a metadata update removes the
last active Kafka cluster/split from a reader subtask. Since no sub-reader
remains to report idleness, downstream watermark progress can stall.
When the reader is already waiting on isAvailable(), resetting availability
to zero sub-readers also leaves the cached availability future incomplete, so
pollNext() is not invoked to release pending split outputs or mark the output
idle.
---
.../source/reader/DynamicKafkaSourceReader.java | 26 ++++
.../DynamicKafkaSourceReaderIdlenessTest.java | 158 +++++++++++++++++++++
.../reader/DynamicKafkaSourceReaderTest.java | 8 +-
3 files changed, 191 insertions(+), 1 deletion(-)
diff --git
a/flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReader.java
b/flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReader.java
index e2a3a572..3d4dd040 100644
---
a/flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReader.java
+++
b/flink-connector-kafka/src/main/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReader.java
@@ -106,6 +106,7 @@ public class DynamicKafkaSourceReader<T> implements
SourceReader<T, DynamicKafka
private int availabilityHelperSize;
private boolean isActivelyConsumingSplits;
private boolean isNoMoreSplits;
+ private boolean dynamicOutputIdle;
private AtomicBoolean restartingReaders;
private ReaderOutput<T> latestReaderOutput;
@@ -132,6 +133,7 @@ public class DynamicKafkaSourceReader<T> implements
SourceReader<T, DynamicKafka
new
MultipleFuturesAvailabilityHelper(this.availabilityHelperSize = 0);
this.isNoMoreSplits = false;
this.isActivelyConsumingSplits = false;
+ this.dynamicOutputIdle = false;
this.restartingReaders = new AtomicBoolean();
this.clustersProperties = new HashMap<>();
this.pendingSplitOutputReleases = new HashSet<>();
@@ -157,6 +159,7 @@ public class DynamicKafkaSourceReader<T> implements
SourceReader<T, DynamicKafka
public InputStatus pollNext(ReaderOutput<T> readerOutput) throws Exception
{
latestReaderOutput = readerOutput;
releasePendingSplitOutputs(readerOutput);
+ maybeUpdateNoActiveSplitOutputIdleness(readerOutput);
// at startup, do not return end of input if metadata event has not
been received
if (clusterReaderMap.isEmpty()) {
@@ -605,6 +608,12 @@ public class DynamicKafkaSourceReader<T> implements
SourceReader<T, DynamicKafka
this.availabilityHelperSize = clusterReaderMap.size());
syncAvailabilityHelperWithReaders();
+ if (getNumberOfActiveSplits() == 0) {
+ restartingReaders.set(false);
+ cachedPreviousFuture.complete(null);
+ return;
+ }
+
// We cannot immediately complete the previous future here. We must
complete it only when
// the new readers have finished handling the split assignment.
Completing the future too
// early can cause WakeupException (implicitly woken up by invocation
to pollNext()) if the
@@ -709,6 +718,23 @@ public class DynamicKafkaSourceReader<T> implements
SourceReader<T, DynamicKafka
pendingSplits.removeIf(split -> split.isRetained() &&
!split.isRetained(currentTimeMillis));
}
+ private int getNumberOfActiveSplits() {
+ return clusterReaderMap.values().stream()
+
.mapToInt(KafkaSourceReader::getNumberOfCurrentlyAssignedSplits)
+ .sum();
+ }
+
+ private void maybeUpdateNoActiveSplitOutputIdleness(ReaderOutput<T>
readerOutput) {
+ boolean hasActiveSplits = getNumberOfActiveSplits() > 0;
+ if (!hasActiveSplits && isActivelyConsumingSplits &&
!dynamicOutputIdle) {
+ readerOutput.markIdle();
+ dynamicOutputIdle = true;
+ } else if (hasActiveSplits && dynamicOutputIdle) {
+ readerOutput.markActive();
+ dynamicOutputIdle = false;
+ }
+ }
+
static Configuration toConfiguration(Properties props) {
Configuration config = new Configuration();
props.stringPropertyNames().forEach(key -> config.setString(key,
props.getProperty(key)));
diff --git
a/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderIdlenessTest.java
b/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderIdlenessTest.java
new file mode 100644
index 00000000..0a34194f
--- /dev/null
+++
b/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderIdlenessTest.java
@@ -0,0 +1,158 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.flink.connector.kafka.dynamic.source.reader;
+
+import org.apache.flink.api.common.eventtime.Watermark;
+import org.apache.flink.connector.kafka.dynamic.metadata.ClusterMetadata;
+import org.apache.flink.connector.kafka.dynamic.metadata.KafkaStream;
+import org.apache.flink.connector.kafka.dynamic.source.MetadataUpdateEvent;
+import
org.apache.flink.connector.kafka.dynamic.source.split.DynamicKafkaSourceSplit;
+import
org.apache.flink.connector.kafka.source.reader.deserializer.KafkaRecordDeserializationSchema;
+import org.apache.flink.connector.kafka.source.split.KafkaPartitionSplit;
+import org.apache.flink.connector.testutils.source.reader.TestingReaderContext;
+import org.apache.flink.connector.testutils.source.reader.TestingReaderOutput;
+import org.apache.flink.core.io.InputStatus;
+
+import org.apache.kafka.clients.consumer.ConsumerConfig;
+import org.apache.kafka.common.TopicPartition;
+import org.apache.kafka.common.serialization.ByteArrayDeserializer;
+import org.junit.jupiter.api.Test;
+
+import java.util.ArrayList;
+import java.util.Collections;
+import java.util.List;
+import java.util.Properties;
+import java.util.concurrent.CompletableFuture;
+
+import static org.assertj.core.api.Assertions.assertThat;
+
+/** Tests idleness transitions for {@link DynamicKafkaSourceReader}. */
+class DynamicKafkaSourceReaderIdlenessTest {
+ private static final String TOPIC = "topic";
+ private static final String REMAINING_TOPIC = "remaining-topic";
+ private static final String KAFKA_CLUSTER_ID = "cluster-1";
+
+ @Test
+ void testMetadataRemovalMarksReaderIdleWhenAllActiveSplitsRemoved() throws
Exception {
+ TestingReaderContext context = new TestingReaderContext();
+ try (DynamicKafkaSourceReader<byte[]> reader =
+ new DynamicKafkaSourceReader<>(
+ context,
+
KafkaRecordDeserializationSchema.valueOnly(ByteArrayDeserializer.class),
+ getRequiredProperties())) {
+ reader.start();
+ reader.handleSourceEvents(getMetadataUpdateEvent());
+
+ DynamicKafkaSourceSplit split =
+ new DynamicKafkaSourceSplit(
+ KAFKA_CLUSTER_ID,
+ new KafkaPartitionSplit(
+ new TopicPartition(TOPIC, 0),
+ 0L,
+ KafkaPartitionSplit.NO_STOPPING_OFFSET));
+ reader.addSplits(Collections.singletonList(split));
+
+ CompletableFuture<Void> availabilityFuture = reader.isAvailable();
+ assertThat(availabilityFuture).isNotDone();
+
+ reader.handleSourceEvents(getMetadataUpdateEvent(REMAINING_TOPIC));
+ assertThat(reader.getAvailabilityHelperSize()).isEqualTo(1);
+ assertThat(availabilityFuture).isDone();
+
+ TrackingReaderOutputWithIdleness<byte[]> readerOutput =
+ new TrackingReaderOutputWithIdleness<>();
+
assertThat(reader.pollNext(readerOutput)).isEqualTo(InputStatus.NOTHING_AVAILABLE);
+
assertThat(reader.pollNext(readerOutput)).isEqualTo(InputStatus.NOTHING_AVAILABLE);
+
assertThat(readerOutput.releasedSplitIds()).containsExactly(split.splitId());
+ assertThat(readerOutput.idleCount()).isEqualTo(1);
+ assertThat(readerOutput.activeCount()).isZero();
+ assertThat(readerOutput.isIdle()).isTrue();
+ }
+ }
+
+ private static MetadataUpdateEvent getMetadataUpdateEvent() {
+ return getMetadataUpdateEvent(TOPIC);
+ }
+
+ private static MetadataUpdateEvent getMetadataUpdateEvent(String topic) {
+ Properties clusterProperties = new Properties();
+ clusterProperties.setProperty(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG,
"localhost:9092");
+ ClusterMetadata clusterMetadata =
+ new ClusterMetadata(Collections.singleton(topic),
clusterProperties);
+ return new MetadataUpdateEvent(
+ Collections.singleton(
+ new KafkaStream(
+ TOPIC,
+ Collections.singletonMap(KAFKA_CLUSTER_ID,
clusterMetadata))));
+ }
+
+ private static Properties getRequiredProperties() {
+ Properties properties = new Properties();
+ properties.setProperty(
+ ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG,
+ ByteArrayDeserializer.class.getName());
+ properties.setProperty(
+ ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG,
+ ByteArrayDeserializer.class.getName());
+ return properties;
+ }
+
+ private static class TrackingReaderOutputWithIdleness<E> extends
TestingReaderOutput<E> {
+ private int idleCount;
+ private int activeCount;
+ private boolean idle;
+ private final List<String> releasedSplitIds = new ArrayList<>();
+
+ @Override
+ public void emitWatermark(Watermark watermark) {}
+
+ @Override
+ public void markIdle() {
+ idleCount++;
+ idle = true;
+ }
+
+ @Override
+ public void markActive() {
+ activeCount++;
+ idle = false;
+ }
+
+ @Override
+ public void releaseOutputForSplit(String splitId) {
+ releasedSplitIds.add(splitId);
+ }
+
+ private int idleCount() {
+ return idleCount;
+ }
+
+ private int activeCount() {
+ return activeCount;
+ }
+
+ private boolean isIdle() {
+ return idle;
+ }
+
+ private List<String> releasedSplitIds() {
+ return releasedSplitIds;
+ }
+ }
+}
diff --git
a/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderTest.java
b/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderTest.java
index 180c3392..fea90b72 100644
---
a/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderTest.java
+++
b/flink-connector-kafka/src/test/java/org/apache/flink/connector/kafka/dynamic/source/reader/DynamicKafkaSourceReaderTest.java
@@ -86,6 +86,12 @@ public class DynamicKafkaSourceReaderTest extends
SourceReaderTestBase<DynamicKa
private static class TrackingReaderOutput<E> extends
TestingReaderOutput<E> {
private final List<String> releasedSplitIds = new ArrayList<>();
+ @Override
+ public void markIdle() {}
+
+ @Override
+ public void markActive() {}
+
@Override
public void releaseOutputForSplit(String splitId) {
releasedSplitIds.add(splitId);
@@ -314,7 +320,7 @@ public class DynamicKafkaSourceReaderTest extends
SourceReaderTestBase<DynamicKa
void testNoSubReadersInputStatus() throws Exception {
try (DynamicKafkaSourceReader<Integer> reader =
(DynamicKafkaSourceReader<Integer>) createReader()) {
- TestingReaderOutput<Integer> readerOutput = new
TestingReaderOutput<>();
+ TrackingReaderOutput<Integer> readerOutput = new
TrackingReaderOutput<>();
InputStatus inputStatus = reader.pollNext(readerOutput);
assertEquals(
InputStatus.NOTHING_AVAILABLE,
