Verbesserungen aus 'rebalance-listener' nach 'counting-consumer' gemerged
authorKai Moritz <kai@juplo.de>
Sat, 13 Aug 2022 09:15:54 +0000 (11:15 +0200)
committerKai Moritz <kai@juplo.de>
Sat, 13 Aug 2022 09:25:55 +0000 (11:25 +0200)
1  2 
src/main/java/de/juplo/kafka/ApplicationConfiguration.java
src/main/java/de/juplo/kafka/EndlessConsumer.java
src/main/java/de/juplo/kafka/KeyCountingRecordHandler.java
src/test/java/de/juplo/kafka/ApplicationTests.java

index 0000000,7a0a8ad..bb219d0
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,76 +1,64 @@@
 -  @Bean
 -  public KeyCountingRebalanceListener keyCountingRebalanceListener(
 -      KeyCountingRecordHandler keyCountingRecordHandler,
 -      ApplicationProperties properties)
 -  {
 -    return new KeyCountingRebalanceListener(
 -        keyCountingRecordHandler,
 -        properties.getClientId());
 -  }
 -
+ package de.juplo.kafka;
+ import org.apache.kafka.clients.consumer.KafkaConsumer;
+ import org.apache.kafka.common.serialization.LongDeserializer;
+ import org.apache.kafka.common.serialization.StringDeserializer;
+ import org.springframework.boot.context.properties.EnableConfigurationProperties;
+ import org.springframework.context.annotation.Bean;
+ import org.springframework.context.annotation.Configuration;
+ import java.util.Properties;
+ import java.util.concurrent.ExecutorService;
+ import java.util.concurrent.Executors;
+ @Configuration
+ @EnableConfigurationProperties(ApplicationProperties.class)
+ public class ApplicationConfiguration
+ {
+   @Bean
+   public KeyCountingRecordHandler keyCountingRecordHandler()
+   {
+     return new KeyCountingRecordHandler();
+   }
 -      KeyCountingRebalanceListener keyCountingRebalanceListener,
+   @Bean
+   public EndlessConsumer<String, Long> endlessConsumer(
+       KafkaConsumer<String, Long> kafkaConsumer,
+       ExecutorService executor,
 -            keyCountingRebalanceListener,
+       KeyCountingRecordHandler keyCountingRecordHandler,
+       ApplicationProperties properties)
+   {
+     return
+         new EndlessConsumer<>(
+             executor,
+             properties.getClientId(),
+             properties.getTopic(),
+             kafkaConsumer,
+             keyCountingRecordHandler);
+   }
+   @Bean
+   public ExecutorService executor()
+   {
+     return Executors.newSingleThreadExecutor();
+   }
+   @Bean(destroyMethod = "close")
+   public KafkaConsumer<String, Long> kafkaConsumer(ApplicationProperties properties)
+   {
+     Properties props = new Properties();
+     props.put("bootstrap.servers", properties.getBootstrapServer());
+     props.put("partition.assignment.strategy", "org.apache.kafka.clients.consumer.CooperativeStickyAssignor");
+     props.put("group.id", properties.getGroupId());
+     props.put("client.id", properties.getClientId());
+     props.put("auto.offset.reset", properties.getAutoOffsetReset());
+     props.put("auto.commit.interval.ms", (int)properties.getCommitInterval().toMillis());
+     props.put("metadata.max.age.ms", "1000");
+     props.put("key.deserializer", StringDeserializer.class.getName());
+     props.put("value.deserializer", LongDeserializer.class.getName());
+     return new KafkaConsumer<>(props);
+   }
+ }
@@@ -22,14 -18,15 +18,14 @@@ import java.util.concurrent.locks.Reent
  
  
  @Slf4j
- public class EndlessConsumer implements Runnable
+ @RequiredArgsConstructor
+ public class EndlessConsumer<K, V> implements Runnable
  {
    private final ExecutorService executor;
-   private final String bootstrapServer;
-   private final String groupId;
    private final String id;
    private final String topic;
-   private final String autoOffsetReset;
+   private final Consumer<K, V> consumer;
 -  private final ConsumerRebalanceListener consumerRebalanceListener;
+   private final RecordHandler<K, V> handler;
  
    private final Lock lock = new ReentrantLock();
    private final Condition condition = lock.newCondition();
    {
      try
      {
-       Properties props = new Properties();
-       props.put("bootstrap.servers", bootstrapServer);
-       props.put("group.id", groupId);
-       props.put("client.id", id);
-       props.put("auto.offset.reset", autoOffsetReset);
-       props.put("metadata.max.age.ms", "1000");
-       props.put("key.deserializer", StringDeserializer.class.getName());
-       props.put("value.deserializer", StringDeserializer.class.getName());
-       this.consumer = new KafkaConsumer<>(props);
        log.info("{} - Subscribing to topic {}", id, topic);
 -      consumer.subscribe(Arrays.asList(topic), consumerRebalanceListener);
 +      consumer.subscribe(Arrays.asList(topic));
  
-       seen = new HashMap<>();
        while (true)
        {
-         ConsumerRecords<String, String> records =
+         ConsumerRecords<K, V> records =
              consumer.poll(Duration.ofSeconds(1));
  
          // Do something with the data...
index 0000000,099dcf7..83b3ff2
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,46 +1,40 @@@
 -  public void addPartition(Integer partition, Map<String, Long> statistics)
 -  {
 -    seen.put(partition, statistics);
 -  }
 -
 -  public Map<String, Long> removePartition(Integer partition)
 -  {
 -    return seen.remove(partition);
 -  }
 -
+ package de.juplo.kafka;
+ import lombok.extern.slf4j.Slf4j;
+ import org.apache.kafka.clients.consumer.ConsumerRecord;
+ import java.util.HashMap;
+ import java.util.Map;
+ @Slf4j
+ public class KeyCountingRecordHandler implements RecordHandler<String, Long>
+ {
+   private final Map<Integer, Map<String, Long>> seen = new HashMap<>();
+   @Override
+   public void accept(ConsumerRecord<String, Long> record)
+   {
+     Integer partition = record.partition();
+     String key = record.key() == null ? "NULL" : record.key().toString();
++
++    if (!seen.containsKey(partition))
++      seen.put(partition, new HashMap<>());
++
+     Map<String, Long> byKey = seen.get(partition);
+     if (!byKey.containsKey(key))
+       byKey.put(key, 0l);
+     long seenByKey = byKey.get(key);
+     seenByKey++;
+     byKey.put(key, seenByKey);
+   }
+   public Map<Integer, Map<String, Long>> getSeen()
+   {
+     return seen;
+   }
+ }
index 0000000,5b13b7d..0909f2c
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,345 +1,342 @@@
 -      KeyCountingRebalanceListener keyCountingRebalanceListener;
 -      @Autowired
+ package de.juplo.kafka;
+ import lombok.extern.slf4j.Slf4j;
+ import org.apache.kafka.clients.consumer.ConsumerRecord;
+ import org.apache.kafka.clients.consumer.KafkaConsumer;
+ import org.apache.kafka.clients.producer.KafkaProducer;
+ import org.apache.kafka.clients.producer.ProducerRecord;
+ import org.apache.kafka.common.TopicPartition;
+ import org.apache.kafka.common.errors.RecordDeserializationException;
+ import org.apache.kafka.common.serialization.*;
+ import org.apache.kafka.common.utils.Bytes;
+ import org.junit.jupiter.api.*;
+ import org.springframework.beans.factory.annotation.Autowired;
+ import org.springframework.boot.autoconfigure.EnableAutoConfiguration;
+ import org.springframework.boot.test.context.ConfigDataApplicationContextInitializer;
+ import org.springframework.boot.test.context.TestConfiguration;
+ import org.springframework.context.annotation.Bean;
+ import org.springframework.context.annotation.Import;
+ import org.springframework.kafka.test.context.EmbeddedKafka;
+ import org.springframework.test.context.TestPropertySource;
+ import org.springframework.test.context.junit.jupiter.SpringJUnitConfig;
+ import java.time.Duration;
+ import java.util.*;
+ import java.util.concurrent.ExecutionException;
+ import java.util.concurrent.ExecutorService;
+ import java.util.function.BiConsumer;
+ import java.util.stream.Collectors;
+ import java.util.stream.IntStream;
+ import static de.juplo.kafka.ApplicationTests.PARTITIONS;
+ import static de.juplo.kafka.ApplicationTests.TOPIC;
+ import static org.assertj.core.api.Assertions.*;
+ import static org.awaitility.Awaitility.*;
+ @SpringJUnitConfig(initializers = ConfigDataApplicationContextInitializer.class)
+ @TestMethodOrder(MethodOrderer.OrderAnnotation.class)
+ @TestPropertySource(
+               properties = {
+                               "consumer.bootstrap-server=${spring.embedded.kafka.brokers}",
+                               "consumer.topic=" + TOPIC,
+                               "consumer.commit-interval=1s" })
+ @EmbeddedKafka(topics = TOPIC, partitions = PARTITIONS)
+ @EnableAutoConfiguration
+ @Slf4j
+ class ApplicationTests
+ {
+       public static final String TOPIC = "FOO";
+       public static final int PARTITIONS = 10;
+       StringSerializer stringSerializer = new StringSerializer();
+       @Autowired
+       Serializer valueSerializer;
+       @Autowired
+       KafkaProducer<String, Bytes> kafkaProducer;
+       @Autowired
+       KafkaConsumer<String, Long> kafkaConsumer;
+       @Autowired
+       KafkaConsumer<Bytes, Bytes> offsetConsumer;
+       @Autowired
+       ApplicationProperties properties;
+       @Autowired
+       ExecutorService executor;
+       @Autowired
 -                                              keyCountingRebalanceListener,
+       KeyCountingRecordHandler keyCountingRecordHandler;
+       EndlessConsumer<String, Long> endlessConsumer;
+       Map<TopicPartition, Long> oldOffsets;
+       Map<TopicPartition, Long> newOffsets;
+       Set<ConsumerRecord<String, Long>> receivedRecords;
+       /** Tests methods */
+       @Test
+       void commitsCurrentOffsetsOnSuccess() throws ExecutionException, InterruptedException
+       {
+               send100Messages((partition, key, counter) ->
+               {
+                       Bytes value = new Bytes(valueSerializer.serialize(TOPIC, counter));
+                       return new ProducerRecord<>(TOPIC, partition, key, value);
+               });
+               await("100 records received")
+                               .atMost(Duration.ofSeconds(30))
+                               .pollInterval(Duration.ofSeconds(1))
+                               .until(() -> receivedRecords.size() >= 100);
+               await("Offsets committed")
+                               .atMost(Duration.ofSeconds(10))
+                               .pollInterval(Duration.ofSeconds(1))
+                               .untilAsserted(() ->
+                               {
+                                       checkSeenOffsetsForProgress();
+                                       compareToCommitedOffsets(newOffsets);
+                               });
+               assertThatExceptionOfType(IllegalStateException.class)
+                               .isThrownBy(() -> endlessConsumer.exitStatus())
+                               .describedAs("Consumer should still be running");
+       }
+       @Test
+       void commitsOffsetOfErrorForReprocessingOnDeserializationError()
+       {
+               send100Messages((partition, key, counter) ->
+               {
+                       Bytes value = counter == 77
+                                       ? new Bytes(stringSerializer.serialize(TOPIC, "BOOM!"))
+                                       : new Bytes(valueSerializer.serialize(TOPIC, counter));
+                       return new ProducerRecord<>(TOPIC, partition, key, value);
+               });
+               await("Consumer failed")
+                               .atMost(Duration.ofSeconds(30))
+                               .pollInterval(Duration.ofSeconds(1))
+                               .until(() -> !endlessConsumer.running());
+               checkSeenOffsetsForProgress();
+               compareToCommitedOffsets(newOffsets);
+               endlessConsumer.start();
+               await("Consumer failed")
+                               .atMost(Duration.ofSeconds(30))
+                               .pollInterval(Duration.ofSeconds(1))
+                               .until(() -> !endlessConsumer.running());
+               checkSeenOffsetsForProgress();
+               compareToCommitedOffsets(newOffsets);
+               assertThat(receivedRecords.size())
+                               .describedAs("Received not all sent events")
+                               .isLessThan(100);
+               assertThatNoException()
+                               .describedAs("Consumer should not be running")
+                               .isThrownBy(() -> endlessConsumer.exitStatus());
+               assertThat(endlessConsumer.exitStatus())
+                               .describedAs("Consumer should have exited abnormally")
+                               .containsInstanceOf(RecordDeserializationException.class);
+       }
+       /** Helper methods for the verification of expectations */
+       void compareToCommitedOffsets(Map<TopicPartition, Long> offsetsToCheck)
+       {
+               doForCurrentOffsets((tp, offset) ->
+               {
+                       Long expected = offsetsToCheck.get(tp) + 1;
+                       log.debug("Checking, if the offset for {} is {}", tp, expected);
+                       assertThat(offset)
+                                       .describedAs("Committed offset corresponds to the offset of the consumer")
+                                       .isEqualTo(expected);
+               });
+       }
+       void checkSeenOffsetsForProgress()
+       {
+               // Be sure, that some messages were consumed...!
+               Set<TopicPartition> withProgress = new HashSet<>();
+               partitions().forEach(tp ->
+               {
+                       Long oldOffset = oldOffsets.get(tp) + 1;
+                       Long newOffset = newOffsets.get(tp) + 1;
+                       if (!oldOffset.equals(newOffset))
+                       {
+                               log.debug("Progress for {}: {} -> {}", tp, oldOffset, newOffset);
+                               withProgress.add(tp);
+                       }
+               });
+               assertThat(withProgress)
+                               .describedAs("Some offsets must have changed, compared to the old offset-positions")
+                               .isNotEmpty();
+       }
+       /** Helper methods for setting up and running the tests */
+       void seekToEnd()
+       {
+               offsetConsumer.assign(partitions());
+               offsetConsumer.seekToEnd(partitions());
+               partitions().forEach(tp ->
+               {
+                       // seekToEnd() works lazily: it only takes effect on poll()/position()
+                       Long offset = offsetConsumer.position(tp);
+                       log.info("New position for {}: {}", tp, offset);
+               });
+               // The new positions must be commited!
+               offsetConsumer.commitSync();
+               offsetConsumer.unsubscribe();
+       }
+       void doForCurrentOffsets(BiConsumer<TopicPartition, Long> consumer)
+       {
+               offsetConsumer.assign(partitions());
+               partitions().forEach(tp -> consumer.accept(tp, offsetConsumer.position(tp)));
+               offsetConsumer.unsubscribe();
+       }
+       List<TopicPartition> partitions()
+       {
+               return
+                               IntStream
+                                               .range(0, PARTITIONS)
+                                               .mapToObj(partition -> new TopicPartition(TOPIC, partition))
+                                               .collect(Collectors.toList());
+       }
+       public interface RecordGenerator<K, V>
+       {
+               public ProducerRecord<String, Bytes> generate(int partition, String key, long counter);
+       }
+       void send100Messages(RecordGenerator recordGenerator)
+       {
+               long i = 0;
+               for (int partition = 0; partition < 10; partition++)
+               {
+                       for (int key = 0; key < 10; key++)
+                       {
+                               ProducerRecord<String, Bytes> record =
+                                               recordGenerator.generate(partition, Integer.toString(partition*10+key%2), ++i);
+                               kafkaProducer.send(record, (metadata, e) ->
+                               {
+                                       if (metadata != null)
+                                       {
+                                               log.debug(
+                                                               "{}|{} - {}={}",
+                                                               metadata.partition(),
+                                                               metadata.offset(),
+                                                               record.key(),
+                                                               record.value());
+                                       }
+                                       else
+                                       {
+                                               log.warn(
+                                                               "Exception for {}={}: {}",
+                                                               record.key(),
+                                                               record.value(),
+                                                               e.toString());
+                                       }
+                               });
+                       }
+               }
+       }
+       @BeforeEach
+       public void init()
+       {
+               seekToEnd();
+               oldOffsets = new HashMap<>();
+               newOffsets = new HashMap<>();
+               receivedRecords = new HashSet<>();
+               doForCurrentOffsets((tp, offset) ->
+               {
+                       oldOffsets.put(tp, offset - 1);
+                       newOffsets.put(tp, offset - 1);
+               });
+               TestRecordHandler<String, Long> captureOffsetAndExecuteTestHandler =
+                               new TestRecordHandler<String, Long>(keyCountingRecordHandler) {
+                                       @Override
+                                       public void onNewRecord(ConsumerRecord<String, Long> record)
+                                       {
+                                               newOffsets.put(
+                                                               new TopicPartition(record.topic(), record.partition()),
+                                                               record.offset());
+                                               receivedRecords.add(record);
+                                       }
+                               };
+               endlessConsumer =
+                               new EndlessConsumer<>(
+                                               executor,
+                                               properties.getClientId(),
+                                               properties.getTopic(),
+                                               kafkaConsumer,
+                                               captureOffsetAndExecuteTestHandler);
+               endlessConsumer.start();
+       }
+       @AfterEach
+       public void deinit()
+       {
+               try
+               {
+                       endlessConsumer.stop();
+               }
+               catch (Exception e)
+               {
+                       log.info("Exception while stopping the consumer: {}", e.toString());
+               }
+       }
+       @TestConfiguration
+       @Import(ApplicationConfiguration.class)
+       public static class Configuration
+       {
+               @Bean
+               Serializer<Long> serializer()
+               {
+                       return new LongSerializer();
+               }
+               @Bean
+               KafkaProducer<String, Bytes> kafkaProducer(ApplicationProperties properties)
+               {
+                       Properties props = new Properties();
+                       props.put("bootstrap.servers", properties.getBootstrapServer());
+                       props.put("linger.ms", 100);
+                       props.put("key.serializer", StringSerializer.class.getName());
+                       props.put("value.serializer", BytesSerializer.class.getName());
+                       return new KafkaProducer<>(props);
+               }
+               @Bean
+               KafkaConsumer<Bytes, Bytes> offsetConsumer(ApplicationProperties properties)
+               {
+                       Properties props = new Properties();
+                       props.put("bootstrap.servers", properties.getBootstrapServer());
+                       props.put("client.id", "OFFSET-CONSUMER");
+                       props.put("group.id", properties.getGroupId());
+                       props.put("key.deserializer", BytesDeserializer.class.getName());
+                       props.put("value.deserializer", BytesDeserializer.class.getName());
+                       return new KafkaConsumer<>(props);
+               }
+       }
+ }