* Merge branch 'sumup-adder--ohne--stored-offsets' into sumup-adder.
* In dem gemergten Branch ist es nicht wichtig, wann genau die
Mongo-DB zwischen den Tests zurückgesetzt wird, da sie nur den Zustand
des Consumers enthält.
* Wenn die Offsets mit in der Mongo-DB gespeichert werden, ist es
wesentlich, an zu welchem Zeitpunkt während der Test-Vorbereitung
diese zurückgesetzt wird!
* ROT: Der verbesserte/verschärfte Test deckt Fehler in der Test-Logik auf.
package de.juplo.kafka;
+import org.apache.kafka.clients.consumer.Consumer;
import org.apache.kafka.clients.consumer.KafkaConsumer;
import org.apache.kafka.common.serialization.StringDeserializer;
import org.springframework.boot.context.properties.EnableConfigurationProperties;
ApplicationRecordHandler recordHandler,
AdderResults adderResults,
StateRepository stateRepository,
+ Consumer<String, String> consumer,
ApplicationProperties properties)
{
return new ApplicationRebalanceListener(
adderResults,
stateRepository,
properties.getClientId(),
+ properties.getTopic(),
Clock.systemDefaultZone(),
- properties.getCommitInterval());
+ properties.getCommitInterval(),
+ consumer);
}
@Bean
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("enable.auto.commit", false);
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", StringDeserializer.class.getName());
import lombok.RequiredArgsConstructor;
import lombok.extern.slf4j.Slf4j;
+import org.apache.kafka.clients.consumer.Consumer;
import org.apache.kafka.common.TopicPartition;
import java.time.Clock;
private final AdderResults adderResults;
private final StateRepository stateRepository;
private final String id;
+ private final String topic;
private final Clock clock;
private final Duration commitInterval;
+ private final Consumer<String, String> consumer;
private final Set<Integer> partitions = new HashSet<>();
private Instant lastCommit = Instant.EPOCH;
+ private boolean commitsEnabled = true;
@Override
public void onPartitionsAssigned(Collection<TopicPartition> partitions)
partitions.forEach(tp ->
{
Integer partition = tp.partition();
- log.info("{} - adding partition: {}", id, partition);
this.partitions.add(partition);
StateDocument document =
stateRepository
.findById(Integer.toString(partition))
.orElse(new StateDocument(partition));
+ log.info("{} - adding partition: {}, offset={}", id, partition, document.offset);
+ if (document.offset >= 0)
+ {
+ // Only seek, if a stored offset was found
+ // Otherwise: Use initial offset, generated by Kafka
+ consumer.seek(tp, document.offset);
+ }
recordHandler.addPartition(partition, document.state);
adderResults.addPartition(partition, document.results);
});
partitions.forEach(tp ->
{
Integer partition = tp.partition();
- log.info("{} - removing partition: {}", id, partition);
this.partitions.remove(partition);
- Map<String, AdderResult> state = recordHandler.removePartition(partition);
- for (String key : state.keySet())
+ Long offset = consumer.position(tp);
+ log.info(
+ "{} - removing partition: {}, offset of next message {})",
+ id,
+ partition,
+ offset);
+ if (commitsEnabled)
+ {
+ Map<String, AdderResult> state = recordHandler.removePartition(partition);
+ Map<String, List<AdderResult>> results = adderResults.removePartition(partition);
+ stateRepository.save(new StateDocument(partition, state, results, offset));
+ }
+ else
{
- log.info(
- "{} - Seen {} messages for partition={}|key={}",
- id,
- state.get(key),
- partition,
- key);
+ log.info("Offset commits are disabled! Last commit: {}", lastCommit);
}
- Map<String, List<AdderResult>> results = adderResults.removePartition(partition);
- stateRepository.save(new StateDocument(partition, state, results));
});
}
@Override
public void beforeNextPoll()
{
+ if (!commitsEnabled)
+ {
+ log.info("Offset commits are disabled! Last commit: {}", lastCommit);
+ return;
+ }
+
if (lastCommit.plus(commitInterval).isBefore(clock.instant()))
{
- log.debug("Storing data, last commit: {}", lastCommit);
+ log.debug("Storing data and offsets, last commit: {}", lastCommit);
partitions.forEach(partition -> stateRepository.save(
new StateDocument(
partition,
recordHandler.getState(partition).getState(),
- adderResults.getState(partition))));
+ adderResults.getState(partition),
+ consumer.position(new TopicPartition(topic, partition)))));
lastCommit = clock.instant();
}
}
+
+ @Override
+ public void enableCommits()
+ {
+ commitsEnabled = true;
+ }
+
+ @Override
+ public void disableCommits()
+ {
+ commitsEnabled = false;
+ }
}
try
{
log.info("{} - Subscribing to topic {}", id, topic);
+ rebalanceListener.enableCommits();
consumer.subscribe(Arrays.asList(topic), rebalanceListener);
while (true)
catch(WakeupException e)
{
log.info("{} - RIIING! Request to stop consumption - commiting current offsets!", id);
- consumer.commitSync();
shutdown();
}
catch(RecordDeserializationException e)
offset,
e.getCause().toString());
- consumer.commitSync();
shutdown(e);
}
catch(Exception e)
{
- log.error("{} - Unexpected error: {}", id, e.toString(), e);
+ log.error("{} - Unexpected error: {}, disabling commits", id, e.toString(), e);
+ rebalanceListener.disableCommits();
shutdown(e);
}
finally
public interface PollIntervalAwareConsumerRebalanceListener extends ConsumerRebalanceListener
{
default void beforeNextPoll() {}
+
+ default void enableCommits() {}
+ default void disableCommits() {}
}
{
@Id
public String id;
+ public long offset = -1l;
public Map<String, AdderResult> state;
public Map<String, List<AdderResult>> results;
public StateDocument(
Integer partition,
Map<String, AdderResult> state,
- Map<String, List<AdderResult>> results)
+ Map<String, List<AdderResult>> results,
+ long offset)
{
this.id = Integer.toString(partition);
this.state = state;
this.results = results;
+ this.offset = offset;
}
}
@Autowired
ExecutorService executor;
@Autowired
+ StateRepository stateRepository;
+ @Autowired
MongoClient mongoClient;
@Autowired
MongoProperties mongoProperties;
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);
+ Integer partition = tp.partition();
+ StateDocument document =
+ stateRepository
+ .findById(partition.toString())
+ .orElse(new StateDocument(partition));
+ document.offset = offset;
+ stateRepository.save(document);
});
- // 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();
+ partitions().forEach(tp ->
+ {
+ String partition = Integer.toString(tp.partition());
+ Optional<Long> offset = stateRepository.findById(partition).map(document -> document.offset);
+ consumer.accept(tp, offset.orElse(0l));
+ });
}
List<TopicPartition> partitions()
props.put("value.deserializer", BytesDeserializer.class.getName());
offsetConsumer = new KafkaConsumer<>(props);
+ mongoClient.getDatabase(mongoProperties.getDatabase()).drop();
seekToEnd();
oldOffsets = new HashMap<>();
}
};
- mongoClient.getDatabase(mongoProperties.getDatabase()).drop();
-
endlessConsumer =
new EndlessConsumer<>(
executor,
projects / demos / kafka / training / commitdiff