I can't believe it's not a queue: Kafka and Spring

BombMQ
“cut the red wire”“cut the blue wire”

Bombfka
“cut the red wire”
“cut the blue wire”
replay 
button

Bombfka
“cut the blue wire”
“cut the red wire”

Agenda
• What is Kafka?
• Kafka + Spring
• Metrics Example
• How Heroku uses Kafka

Kafka is a distributed,
partitioned, replicated
commit log service. It
provides the functionality
of a messaging system, but
with a unique design.

Distributed
Publish
Subscribe
Messaging

“hundreds of thousands to
millions of messages a second
on a small cluster”
Tom Crayford
Heroku Kafka

Know Your Cuts of Kafka
Producers Consumers
Partitions
Groups
Brokers
Messages
Topics
Keys

Messages
• Header
• Key
• Value
(Byte Array)

Each Topic Partition
is an ordered log of
immutable messages,
append-only

• Messages are produced in order
• Messages are consumed in order
• Topics are distributed and replicated
Kafka Guarantees

props.put("bootstrap.servers", “broker1:9092,broker2:9092”);
props.put(“key.serializer”, StringSerializer.class.getName());
props.put(“value.serializer”, StringSerializer.class.getName());
Producer<String, String> producer = new KafkaProducer<>(props);
Producer API
producer.send(new ProducerRecord<>("my-topic", message2));
producer.send(...).get();

Consumer API
Automatic Offset Committing
props.put("bootstrap.servers", “broker1:9092,broker2:9092”);
props.put(“key.deserializer”, StringDeserializer.class.getName());
props.put(“value.deserializer”, StringDeserializer.class.getName());
KafkaConsumer<String, String> consumer = new KafkaConsumer<>(props);
consumer.subscribe(singletonList(“my-topic”));
while (running.get()) {
ConsumerRecords<String, String> records = consumer.poll(100);
for (ConsumerRecord<String, String> record : records) {
// ...
}
}

Consumer API
Manual Offset Committing
props.put("enable.auto.commit", "false");
// ...
KafkaConsumer<String, String> consumer = new KafkaConsumer<>(props);
consumer.subscribe(singletonList(“my-topic”));
final int minBatchSize = 200;
while (running.get()) {
ConsumerRecords<String, String> records = consumer.poll(100);
for (ConsumerRecord<String, String> record : records) {
buffer.add(record);
}
if (buffer.size() >= minBatchSize) {
insertIntoDb(buffer);
consumer.commitSync();
buffer.clear();
}
}

Consumer API
Kafka Consumer is NOT threadsafe

Consumer API
Advanced!
• Per-message offset commit
• Manual Partition Assignment
• Storing Offsets Outside Kafka
• Kafka Streams (since 0.10)

Using Kafka with Spring
Producer
@SpringBootApplication
@EnableKafka
public class SpringApplicationProducer {
@Bean
public KafkaTemplate<Integer, String> kafkaTemplate() {
return new KafkaTemplate<Integer, String>(producerFactory());
}
@Bean
public ProducerFactory<Integer, String> producerFactory() {
return new DefaultKafkaProducerFactory<>(producerConfigs());
}
private Map<String, Object> producerConfigs() {
Map<String, Object> props = new HashMap<>();
props.put(ProducerConfig.BOOTSTRAP_SERVERS_CONFIG, "localhost:9092");
props.put(ProducerConfig.KEY_SERIALIZER_CLASS_CONFIG, IntegerSerializer.class);
// ...
return props;
}
}

Producer
@Autowired
private KafkaTemplate<Integer, String> template;
public void send() throws Exception {
template.send(“my-topic", message);
}

Consumer
@Service
public class MyKafkaListener {
@KafkaListener(topicPattern = “my-topic")
public void listen(String message) {
System.out.println("received: " + message);
}
}

Consumer
@Service
public class MyKafkaListener {
@KafkaListener(id = “my-listener", topicPartitions = {
@TopicPartition(topic = "topic1", partitions = { "0", "1" }),
@TopicPartition(topic = "topic2", partitions = { "0", "1" })
})
public void listen(String message) {
System.out.println("received: " + message);
}
}

Architecture
Web Request
Primary App
Metrics App
Router Logs

POST
Spring App
Log Drain App
(Producer)
Kafka
Cluster
Log Drain (HTTPS)
Architecture
Web Request
Metrics
Aggregator
(Consumer)

Heroku Log Drains
$ heroku drains:add https://<metrics-app>/logs
242 <158>1 2016-06-20T21:56:57.107495+00:00 host heroku router -
at=info method=GET path="/" host=demodayex.herokuapp.com
request_id=1850b395-c7aa-485c-aa04-7d0894b5f276 fwd="68.32.161.89"
dyno=web.1 connect=0ms service=6ms status=200 bytes=1548

@RequestMapping(value = "/logs", method = RequestMethod.POST) 
@ResponseBody 
public String logs(@RequestBody String body) throws IOException { 
 
// "application/logplex-1" does not conform to RFC5424.  
// It leaves out STRUCTURED-DATA but does not replace it with 
// a NILVALUE. To workaround this, we inject empty STRUCTURED-DATA. 
String[] parts = body.split("router - "); 
String log = parts[0] + "router - [] " + (parts.length > 1 ? parts[1] : ""); 
 
RFC6587SyslogDeserializer parser = new RFC6587SyslogDeserializer(); 
InputStream is = new ByteArrayInputStream(log.getBytes()); 
Map<String, ?> messages = parser.deserialize(is); 
ObjectMapper mapper = new ObjectMapper();
String json = mapper.writeValueAsString(messages); 
template.send("logs", json); 
 
return "ok"; 
}

Heroku Kafka
$ heroku addons:create heroku-kafka
$ heroku kafka:create logs Create the Topic
$ heroku plugins:install heroku-kafka
Create the Cluster

Heroku Kafka
$ heroku kafka:info
=== KAFKA_URL
Name: kafka-asymmetrical-77749
Created: 2016-06-20 18:21 UTC
Plan: Beta Dev
Status: available
Version: 0.9.0.1
Topics: 2 topics (see heroku kafka:list)
Connections: 0 consumers (0 applications)
Messages: 32.0 messages/s
Traffic: 2.25 KB/s in / 2.25 KB/s out

Heroku Kafka
$ heroku kafka:topic logs
=== KAFKA_URL :: logs
Producers: 0.0 messages/second (0 Bytes/second)
Consumers: 0 Bytes/second total
Partitions: 32 partitions
Replication Factor: 2 (recommend > 1)
Compaction: Compaction is disabled for logs
Retention: 24 hours

public class Metrics {
// ...
public static void main(String[] args) { /* … */ }
public Metrics() throws Exception {
// ... 
 
URI redisUri = new URI(System.getenv("REDIS_URL")); 
pool = new JedisPool(redisUri); 
}
private void start() {
// ...
running.set(true);
executor = Executors.newSingleThreadExecutor(); 
executor.submit(this::loop); 
stopLatch = new CountDownLatch(1);
}
}
Main Consumer Class

private void loop() {
// ...
consumer = new KafkaConsumer<>(properties); 
consumer.subscribe(singletonList(KafkaConfig.getTopic()));
while (running.get()) { 
ConsumerRecords<String, String> records = consumer.poll(100); 
for (ConsumerRecord<String, String> record : records) { 
try { 
Map<String,String> recordMap =
mapper.readValue(record.value(), typeRef); 
Route route = new Route(recordMap); 
receive(route); 
} catch (IOException e) { 
e.printStackTrace(); 
} 
} 
} 
 
consumer.close(); 
stopLatch.countDown();
}
Main Consumer Method

private void receive(Route route) {
// …
 
jedis.hincrBy(key, "sum", value); 
jedis.hincrBy(key, "count", 1); 
 
Integer sum = Integer.valueOf(jedis.hget(key, "sum")); 
Float count = Float.valueOf(jedis.hget(key, "count")); 
Float avg = sum / count;
jedis.hset(key, "average", String.valueOf(avg));
}
Update Redis

?
POST
Spring App
Log Drain App
(Producer)
Kafka
Cluster
Log Drain (HTTPS)
Architecture
Web Request
Metrics
Aggregator
(Consumer)

POST
Spring App
Log Drain App
(Producer)
Kafka
Cluster
Log Drain (HTTPS)
Architecture
Web Request
Metrics
Aggregator
(Consumer)
Replay
(Consumer) Staging App

Demo App
https://github.com/jkutner/heroku-metrics-spring
$ docker-compose up web

Other Use Cases
• User Activity
• Stream Processing

Metrics
Logs
IoT Data
Kafka
Cluster
Stream Processing

Kafka @ Heroku
• Metrics
• API Event Bus

Heroku Kafka http://heroku.com/kafka
$ heroku addons:create heroku-kafka

@codefinger
Joe Kutner
Thank You!

I can't believe it's not a queue: Kafka and Spring

More Related Content

What's hot

Similar to I can't believe it's not a queue: Kafka and Spring

More from Joe Kutner

Recently uploaded

I can't believe it's not a queue: Kafka and Spring