protected void setOffset(int partition, long offset) {
if (commitedOffsets[partition] < offset) {
setOffset(zkClient, consumerConfig.groupId(), topicName, partition, offset);
commitedOffsets[partition] = offset;
} else {
LOG.debug(
"Ignoring offset {} for partition {} because it is already committed", offset, partition);
}
}
@Test
public void javaHandleKafkaStorage() {
final ConsumerProperties consumerProperties =
new PropertiesBuilder.Consumer(
brokerList, zooKeepHost, topic, groupId, new StringDecoder(null))
.build()
.readFromEndOfStream()
.consumerTimeoutMs(1234)
.kafkaOffsetsStorage(true);
final ConsumerConfig consumerConfig = consumerProperties.toConsumerConfig();
assertEquals(consumerProperties.zookeeperConnect(), zooKeepHost);
assertEquals(consumerProperties.topic(), topic);
assertEquals(consumerProperties.groupId(), groupId);
assertEquals(
consumerProperties.decoder().getClass().getSimpleName(),
StringDecoder.class.getSimpleName());
assertEquals(consumerConfig.clientId(), groupId);
assertEquals(consumerConfig.autoOffsetReset(), "largest");
assertEquals(consumerConfig.offsetsStorage(), "kafka");
assertEquals(consumerConfig.consumerTimeoutMs(), 1234);
assertEquals(consumerConfig.dualCommitEnabled(), true);
}
@Test
public void javaHandleBaseCase() {
final PropertiesBuilder.Consumer propsBuilder =
new PropertiesBuilder.Consumer(
brokerList, zooKeepHost, topic, groupId, new StringDecoder(null));
assertEquals(propsBuilder.getBrokerList(), brokerList);
assertEquals(propsBuilder.getZooKeeperHost(), zooKeepHost);
final ConsumerProperties consumerProperties = propsBuilder.build();
final ConsumerConfig consumerConfig = consumerProperties.toConsumerConfig();
assertEquals(consumerProperties.zookeeperConnect(), zooKeepHost);
assertEquals(consumerProperties.topic(), topic);
assertEquals(consumerProperties.groupId(), groupId);
assertEquals(
consumerProperties.decoder().getClass().getSimpleName(),
StringDecoder.class.getSimpleName());
assertEquals(consumerConfig.clientId(), groupId);
assertEquals(consumerConfig.autoOffsetReset(), "smallest");
assertEquals(consumerConfig.offsetsStorage(), "zookeeper");
assertEquals(consumerConfig.consumerTimeoutMs(), 1500);
assertEquals(consumerConfig.dualCommitEnabled(), false);
}
/**
* For the @param consumerConfig, specify at least the "groupid" and "zookeeper.connect" string.
* The config will be passed into the Kafka High Level Consumer. For a full list of possible
* values, check this out: https://kafka.apache.org/documentation.html#consumerconfigs
*/
public PersistentKafkaSource(
String topicName,
DeserializationSchema<OUT> deserializationSchema,
ConsumerConfig consumerConfig) {
Preconditions.checkNotNull(topicName);
Preconditions.checkNotNull(deserializationSchema);
Preconditions.checkNotNull(consumerConfig);
this.topicName = topicName;
this.deserializationSchema = deserializationSchema;
this.consumerConfig = consumerConfig;
if (consumerConfig.autoCommitEnable()) {
throw new IllegalArgumentException(
"'auto.commit.enable' is set to 'true'. "
+ "This source can only be used with auto commit disabled because the "
+ "source is committing to zookeeper by itself (not using the KafkaConsumer).");
}
if (!consumerConfig.offsetsStorage().equals("zookeeper")) {
// we can currently only commit to ZK.
throw new IllegalArgumentException(
"The 'offsets.storage' has to be set to 'zookeeper' for this Source to work reliably");
}
}
private void writeObject(ObjectOutputStream out) throws IOException, ClassNotFoundException {
out.defaultWriteObject();
out.writeObject(consumerConfig.props().props());
}
@Override
public void open(Configuration parameters) throws Exception {
super.open(parameters);
ConsumerConnector consumer = Consumer.createJavaConsumerConnector(this.consumerConfig);
// we request only one stream per consumer instance. Kafka will make sure that each consumer
// group
// will see each message only once.
Map<String, Integer> topicCountMap = Collections.singletonMap(topicName, 1);
Map<String, List<KafkaStream<byte[], byte[]>>> streams =
consumer.createMessageStreams(topicCountMap);
if (streams.size() != 1) {
throw new RuntimeException("Expected only one message stream but got " + streams.size());
}
List<KafkaStream<byte[], byte[]>> kafkaStreams = streams.get(topicName);
if (kafkaStreams == null) {
throw new RuntimeException(
"Requested stream not available. Available streams: " + streams.toString());
}
if (kafkaStreams.size() != 1) {
throw new RuntimeException(
"Requested 1 stream from Kafka, bot got " + kafkaStreams.size() + " streams");
}
LOG.info(
"Opening Consumer instance for topic '{}' on group '{}'",
topicName,
consumerConfig.groupId());
this.iteratorToRead = kafkaStreams.get(0).iterator();
this.consumer = consumer;
zkClient =
new ZkClient(
consumerConfig.zkConnect(),
consumerConfig.zkSessionTimeoutMs(),
consumerConfig.zkConnectionTimeoutMs(),
new KafkaZKStringSerializer());
// most likely the number of offsets we're going to store here will be lower than the number of
// partitions.
int numPartitions = getNumberOfPartitions();
LOG.debug("The topic {} has {} partitions", topicName, numPartitions);
this.lastOffsets =
getRuntimeContext().getOperatorState("offset", new long[numPartitions], false);
this.commitedOffsets = new long[numPartitions];
// check if there are offsets to restore
if (!Arrays.equals(lastOffsets.getState(), new long[numPartitions])) {
if (lastOffsets.getState().length != numPartitions) {
throw new IllegalStateException(
"There are "
+ lastOffsets.getState().length
+ " offsets to restore for topic "
+ topicName
+ " but "
+ "there are only "
+ numPartitions
+ " in the topic");
}
LOG.info("Setting restored offsets {} in ZooKeeper", Arrays.toString(lastOffsets.getState()));
setOffsetsInZooKeeper(lastOffsets.getState());
} else {
// initialize empty offsets
Arrays.fill(this.lastOffsets.getState(), -1);
}
Arrays.fill(this.commitedOffsets, 0); // just to make it clear
running = true;
}
public void setConsumerProperty(String key, Object value) {
consumerConfig.props().props().setProperty(checkNotNull(key), checkNotNull(value).toString());
}
