You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "hlteoh37 (via GitHub)" <gi...@apache.org> on 2025/04/14 17:35:36 UTC
Re: [PR] [FLINK-36947][Connectors/Kinesis] Fix issue where excessive GetRecords calls are made on idle source causing high CPU utilisation and throttling. [FLINK-36939][Connectors/Kinesis] Fix issue where excessive BlockingQueue.poll() in FanOutKinesisShardSplitReader and GetRecords in PollingKinesisShardSplitReader calls are made on idle source causing high CPU utilisation [flink-connector-aws]
hlteoh37 commented on code in PR #195:
URL: https://github.com/apache/flink-connector-aws/pull/195#discussion_r2042593349
##########
flink-connector-aws/flink-connector-aws-kinesis-streams/src/main/java/org/apache/flink/connector/kinesis/source/reader/KinesisShardSplitReaderBase.java:
##########
@@ -125,6 +144,52 @@ public RecordsWithSplitIds<Record> fetch() throws IOException {
recordBatch.isCompleted());
}
+ private boolean skipWhenNoAssignedSplit(KinesisShardSplitState splitState) throws IOException {
+ if (splitState == null) {
+ try {
+ // Small sleep to prevent busy polling
+ Thread.sleep(1);
+ return true;
+ } catch (InterruptedException e) {
+ throw new IOException("Sleep was interrupted while skipping no assigned split", e);
+ }
+ }
+
+ return false;
+ }
+
+ private boolean skipUntilScheduledFetchTime(KinesisShardSplitState splitState)
+ throws IOException {
+ if (scheduledFetchTimes.containsKey(splitState)
+ && scheduledFetchTimes.get(splitState) > System.currentTimeMillis()) {
+ try {
+ // Small sleep to prevent busy polling
+ Thread.sleep(1);
+ return true;
+ } catch (InterruptedException e) {
+ throw new IOException(
Review Comment:
Same as above, we should propagate the interrupted status up the thread.
##########
flink-connector-aws/flink-connector-aws-kinesis-streams/src/main/java/org/apache/flink/connector/kinesis/source/reader/KinesisShardSplitReaderBase.java:
##########
@@ -125,6 +144,52 @@ public RecordsWithSplitIds<Record> fetch() throws IOException {
recordBatch.isCompleted());
}
+ private boolean skipWhenNoAssignedSplit(KinesisShardSplitState splitState) throws IOException {
+ if (splitState == null) {
+ try {
+ // Small sleep to prevent busy polling
+ Thread.sleep(1);
+ return true;
+ } catch (InterruptedException e) {
+ throw new IOException("Sleep was interrupted while skipping no assigned split", e);
+ }
+ }
+
+ return false;
+ }
+
+ private boolean skipUntilScheduledFetchTime(KinesisShardSplitState splitState)
+ throws IOException {
+ if (scheduledFetchTimes.containsKey(splitState)
+ && scheduledFetchTimes.get(splitState) > System.currentTimeMillis()) {
+ try {
+ // Small sleep to prevent busy polling
+ Thread.sleep(1);
+ return true;
+ } catch (InterruptedException e) {
+ throw new IOException(
+ "Sleep was interrupted while skipping until scheduled fetch record time",
+ e);
+ }
+ }
+
+ return false;
+ }
+
+ private void scheduleNextFetchTime(KinesisShardSplitState splitState, RecordBatch recordBatch) {
Review Comment:
Let's add a javadoc to explain how we expect "scheduledFetchTimes" to be used.
1/ If split was not active, we will schedule fetch time. We will wait until next fetch time before polling.
2/ If split was active, we will NOT schedule a fetch time. We should immediately poll.
(Num. 2 is not intuitive)
##########
flink-connector-aws/flink-connector-aws-kinesis-streams/src/main/java/org/apache/flink/connector/kinesis/source/reader/KinesisShardSplitReaderBase.java:
##########
@@ -125,6 +144,52 @@ public RecordsWithSplitIds<Record> fetch() throws IOException {
recordBatch.isCompleted());
}
+ private boolean skipWhenNoAssignedSplit(KinesisShardSplitState splitState) throws IOException {
+ if (splitState == null) {
+ try {
+ // Small sleep to prevent busy polling
+ Thread.sleep(1);
+ return true;
+ } catch (InterruptedException e) {
+ throw new IOException("Sleep was interrupted while skipping no assigned split", e);
Review Comment:
We should propagate the interrupted status up the thread.
```
Thread.interrupt()
```
##########
flink-connector-aws/flink-connector-aws-kinesis-streams/src/test/java/org/apache/flink/connector/kinesis/source/reader/PollingKinesisShardSplitReaderTest.java:
##########
@@ -395,6 +399,79 @@ void testMaxRecordsToGetParameterPassed() throws IOException {
assertThat(records.size()).isEqualTo(maxRecordsToGet);
}
+ @ValueSource(longs = {0L, 250L, 1000L})
+ @ParameterizedTest
+ public void testGetRecordsIntervalForIdleSource(long interval) {
+ Configuration configuration = newConfigurationForTest();
+ configuration.set(READER_EMPTY_RECORDS_FETCH_INTERVAL, Duration.ofMillis(interval));
+
+ splitReader =
+ new PollingKinesisShardSplitReader(
+ testStreamProxy, shardMetricGroupMap, configuration);
+
+ testStreamProxy.addShards(TEST_SHARD_ID);
+ splitReader.handleSplitsChanges(
+ new SplitsAddition<>(Collections.singletonList(getTestSplit(TEST_SHARD_ID))));
+
+ await().atLeast(interval, TimeUnit.MILLISECONDS)
+ .atMost(interval + 250, TimeUnit.MILLISECONDS)
+ .pollInSameThread()
+ .pollInterval(Duration.ofMillis(1))
+ .untilAsserted(
+ () -> {
+ assertThat(splitReader.fetch().nextRecordFromSplit()).isNull();
+ assertThat(testStreamProxy.getTotalGetRecordsCall()).isEqualTo(2);
+ });
+ }
+
+ @ValueSource(ints = {1, 8, 64})
+ @ParameterizedTest
+ public void testGetRecordsIntervalForMultipleIdleSource(int shardCount) {
Review Comment:
easy to read :)
##########
flink-connector-aws/flink-connector-aws-kinesis-streams/src/test/java/org/apache/flink/connector/kinesis/source/reader/fanout/FanOutKinesisShardSplitReaderTest.java:
##########
@@ -31,32 +32,48 @@
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
+import org.junit.jupiter.params.ParameterizedTest;
+import org.junit.jupiter.params.provider.ValueSource;
import software.amazon.awssdk.services.kinesis.model.Record;
import java.time.Duration;
import java.util.ArrayList;
import java.util.Collections;
+import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.TimeUnit;
+import java.util.stream.Collectors;
+import java.util.stream.IntStream;
+import static org.apache.flink.connector.kinesis.source.config.KinesisSourceConfigOptions.EFO_CONSUMER_SUBSCRIPTION_TIMEOUT;
+import static org.apache.flink.connector.kinesis.source.config.KinesisSourceConfigOptions.READER_EMPTY_RECORDS_FETCH_INTERVAL;
import static org.apache.flink.connector.kinesis.source.util.TestUtil.CONSUMER_ARN;
+import static org.apache.flink.connector.kinesis.source.util.TestUtil.generateShardId;
import static org.apache.flink.connector.kinesis.source.util.TestUtil.getTestSplit;
import static org.assertj.core.api.AssertionsForClassTypes.assertThatNoException;
import static org.assertj.core.api.AssertionsForInterfaceTypes.assertThat;
+import static org.awaitility.Awaitility.await;
import static org.junit.jupiter.api.Assertions.assertTimeoutPreemptively;
+import static org.mockito.Mockito.spy;
+import static org.mockito.Mockito.times;
+import static org.mockito.Mockito.verify;
Review Comment:
Why do we need to test the await behaviour here? Since we are testing the behaviour of the `KinesisShardSplitReaderBase`, can we instead
1/ Implement a no-op test implementation of the base, and test that? That way we scope it directly to the abstract class being tested.
##########
flink-connector-aws/flink-connector-aws-kinesis-streams/src/test/java/org/apache/flink/connector/kinesis/source/reader/fanout/FanOutKinesisShardSplitReaderTest.java:
##########
@@ -31,32 +32,48 @@
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
+import org.junit.jupiter.params.ParameterizedTest;
+import org.junit.jupiter.params.provider.ValueSource;
import software.amazon.awssdk.services.kinesis.model.Record;
import java.time.Duration;
import java.util.ArrayList;
import java.util.Collections;
+import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.TimeUnit;
+import java.util.stream.Collectors;
+import java.util.stream.IntStream;
+import static org.apache.flink.connector.kinesis.source.config.KinesisSourceConfigOptions.EFO_CONSUMER_SUBSCRIPTION_TIMEOUT;
+import static org.apache.flink.connector.kinesis.source.config.KinesisSourceConfigOptions.READER_EMPTY_RECORDS_FETCH_INTERVAL;
import static org.apache.flink.connector.kinesis.source.util.TestUtil.CONSUMER_ARN;
+import static org.apache.flink.connector.kinesis.source.util.TestUtil.generateShardId;
import static org.apache.flink.connector.kinesis.source.util.TestUtil.getTestSplit;
import static org.assertj.core.api.AssertionsForClassTypes.assertThatNoException;
import static org.assertj.core.api.AssertionsForInterfaceTypes.assertThat;
+import static org.awaitility.Awaitility.await;
import static org.junit.jupiter.api.Assertions.assertTimeoutPreemptively;
+import static org.mockito.Mockito.spy;
+import static org.mockito.Mockito.times;
+import static org.mockito.Mockito.verify;
Review Comment:
We shouldn't be using Mockito - Flink best practices is to implement test classes to mimic interface behavior
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org