Streams Upgrade Guide

Upgrading from Confluent Platform 6.0.x to Confluent Platform 6.1.15

Compatibility

Kafka Streams applications built with Confluent Platform 6.1.15 are forward and backward compatible with certain Kafka clusters.

Forward-compatible to newer clusters up to Confluent Platform 6.1.15:

Existing Kafka Streams applications built with Confluent Platform 3.0.x, Confluent Platform 3.1.x, Confluent Platform 3.2.x, Confluent Platform 3.3.x, Confluent Platform 4.0.x, Confluent Platform 4.1.x, Confluent Platform 5.0.x, Confluent Platform 5.1.x, Confluent Platform 5.2.x, Confluent Platform 5.3.x, Confluent Platform 5.4.x, or Confluent Platform 5.5.x. will work with upgraded Kafka clusters running Confluent Platform 6.1.15.

Backward-compatible to older clusters down to Confluent Platform 3.1.x:

New Kafka Streams applications built with Confluent Platform 6.1.15 will work with older Kafka clusters running Confluent Platform 3.3.x, Confluent Platform 4.0.x, Confluent Platform 4.1.x, Confluent Platform 5.0.x, Confluent Platform 5.1.x, Confluent Platform 5.2.x, Confluent Platform 5.3.x, Confluent Platform 5.4.x, or Confluent Platform 5.5.x. Kafka clusters running Confluent Platform 3.0.x, Confluent Platform 3.1.x, or Confluent Platform 3.2.x are not compatible with new Confluent Platform 6.1.15 Kafka Streams applications.

Note

As of Confluent Platform 5.2.2 and Confluent Platform 5.3.0, Kafka Streams requires message format 0.11 or higher. Thus, if you kept an older message format when upgrading your brokers to Confluent Platform 3.3 or a later version, Kafka Streams Confluent Platform 5.2.2 or Confluent Platform 5.3.0 won’t work. You will need to upgrade the message format to 0.11 before you upgrade your Kafka Streams application to Confluent Platform 5.2.2, Confluent Platform 5.3.0, or newer.

Note

As of Confluent Platform 4.0.0, Kafka Streams requires message format 0.10 or higher. Thus, if you kept an older message format when upgrading your brokers to Confluent Platform 3.1 or a later version, Kafka Streams Confluent Platform 6.1.15 won’t work. You will need to upgrade the message format to 0.10 before you upgrade your Kafka Streams application to Confluent Platform 6.1.15 or newer.

Compatibility Matrix:

	Kafka Broker (columns)
Streams API (rows)	3.0.x / 0.10.0.x	3.1.x / 0.10.1.x and 3.2.x / 0.10.2.x	3.3.x / 0.11.0.x and 4.0.x / 1.0.x and 4.1.x / 1.1.x and 5.0.x / 2.0.x and 5.1.x / 2.1.x and 5.2.x / 2.2.x and 5.3.x / 2.3.x and 5.4.x / 2.4.x and 5.5.x / 2.5.x and 6.0.x / 2.6.x
3.0.x / 0.10.0.x	compatible	compatible	compatible
3.1.x / 0.10.1.x and 3.2.x / 0.10.2.x		compatible	compatible
3.3.x / 0.11.0.x		compatible with exactly-once turned off (requires broker version Confluent Platform 3.3.x or higher)	compatible
4.0.x / 1.0.x and 4.1.x / 1.1.x and 5.0.x / 2.0.x and 5.1.x / 2.1.x and 5.2.0 / 2.2.0 and 5.2.1 / 2.2.0		compatible with exactly-once turned off (requires broker version Confluent Platform 3.3.x or higher); requires message format 0.10 or higher; message headers are not supported (requires broker version Confluent Platform 3.3.x or higher with message format 0.11 or higher)	compatible; requires message format 0.10 or higher; if message headers are used, message format 0.11 or higher required
5.2.2 / 2.2.1 and 5.3.x / 2.3.x and 5.4.x / 2.4.x and 5.5.x / 2.5.x and 6.0.x / 2.6.x			compatible; requires message format 0.11 or higher; enabling exactly-once v2 requires Confluent Platform 5.4.x or higher

The Streams API is not compatible with Kafka clusters running older Kafka versions (0.7, 0.8, 0.9).

Upgrading your Kafka Streams applications to Confluent Platform 6.1.15

To make use of Confluent Platform 6.1.15, you need to update the Kafka Streams dependency of your application to use the version number 6.1.15-ccs, and you may need to make minor code changes (details below), and then recompile your application.

For example, in your pom.xml file:

<dependency>
    <groupId>org.apache.kafka</groupId>
    <artifactId>kafka-streams</artifactId>
    <!-- update version to 6.1.15-ccs -->
    <version>6.1.15-ccs</version>
</dependency>

As of the 6.1.15 release, Kafka Streams depends on a RocksDB version that requires MacOS 10.14 or higher.

Note

As of Confluent Platform 6.0.0 Kafka Streams introduces a new processing mode "exactly_once_beta" (required broker version Confluent Platform 5.5.x). The exactly_once_beta processing mode is EOS version 2 and is ready for production (i.e., it’s not “beta” software). If you want to upgrade from a lower than Confluent Platform 6.0.0 version and enable this new feature, you must follow a two step upgrade path:

1. Upgrade your application to the new bytecode without a config change (i.e., keeping processing.guarantee="exactly_once")
2. Once all members are on the newer version, change the config to processing.guarantee="exactly_once_beta" and do a second rolling bounce to turn on the new processing mode

For a downgrade, do the reverse: first switch the processing.guarantee config from "exactly_once_beta" to "exatly_once" to disable the feature in your Confluent Platform 6.0.0 application. Afterwards downgrade the bytecode to a lower version.

Note

As of Confluent Platform 5.4.0 Kafka Streams uses the new Incremental Cooperative Rebalancing protocol by default. In order to safely do a rolling upgrade from any version lower than Confluent Platform 5.4.0, you must follow a specific upgrade path that includes two rolling bounces:

1. Set the UPGRADE_FROM config to whichever version you are upgrading from, then do the first rolling bounce and upgrade to the new bytecode
2. Once all members are on the newer version, remove the UPGRADE_FROM config and do a second rolling bounce to turn on cooperative rebalancing

Note that if you wish to continue following the old eager rebalancing protocol, you can simply leave the UPGRADE_FROM config in place. Streams will not use cooperative rebalancing as long as this is set to 2.3 or lower, but it is recommended to make sure it is at least 2.0 as earlier versions will have other side effects required for upgrading from much older versions.

Note

As of Confluent Platform 4.0.0 a topology regression was introduced when source KTable instances were changed to have changelog topics instead of re-using the source topic. As of Confluent Platform 5.0.0 KTable instances re-using the source topic as the changelog topic has been reinstated, but is optional and must be configured by setting StreamsConfig.TOPOLOGY_OPTIMIZATION to StreamsConfig.OPTIMIZE.

This brings up some different scenarios depending on what you are upgrading from and what you are upgrading to.

• If you are upgrading from Kafka 2.0.x-cp1 to Kafka 6.1.15-ccs it’s recommended to keep the existing optimization configuration. Changing the optimization level might make your application vulnerable to data loss for a small window of time.
• If you are upgrading from using StreamsBuilder version Kafka 1.0.x-cp1/1.1.x-cp1 to Kafka 6.1.15-ccs you can enable the optimization, but your topology will change so you’ll need to restart your application with a new application ID. Additionally, if you want to perform a rolling upgrade, it is recommended not to enable the optimization. If you elect not to enable the optimization, then no further changes are required.
• If you are upgrading from using KStreamBuilder version Kafka 1.0.x-cp1/1.1.x-cp1 to StreamsBuilder Kafka 6.1.15-ccs, it’s recommended to enable the optimization as there are no changes to your topology. Please note that if you elect not to enable the optimization, there is a small window of time for possible data loss until the changelog topic contains one record per key.

Additionally, when starting with a new application ID, you can possibly end up reprocessing data, since the application ID has been changed. If you don’t want to reprocess records, you’ll need to create new output topics, so downstream user can cut over in a controlled fashion.

Streams API changes

Sliding Windows

Confluent Platform 6.1 adds SlidingWindows as an option for windowedBy() windowed aggregations as described in KIP-450. Sliding windows are fixed-time and data-aligned windows that allow for flexible and efficient windowed aggregations. Refer to Developer Guide for more details.

TRACE level end-to-end latency metrics

The end-to-end latency metrics introduced in 6.0 have been expanded to include store-level metrics. The new store-level metrics are recorded at the TRACE level, a new metrics recording level. Enabling TRACE level metrics automatically turns on all higher levels, i.e., INFO and DEBUG. For more information, see KIP-613.

Upgrading older Kafka Streams applications to Confluent Platform 6.1.x

Streams API changes in Confluent Platform 6.0.0

• New processing mode that improves application scalability using exactly-once guarantees (KIP-447)
• Highly available state stores (KIP-441)
• New KStream.repartition() operator replaces KStream.through() (KIP-221)
• New end-to-end latency metrics added (KIP-613)
• New --force option in StreamsResetter to force remove left-over members (KIP-571)

Improved exactly-once processing

Starting in Confluent Platform 6.0, a new processing mode "exactly_once_beta" (configurable via parameter processing.guarantee) is available. To use this new feature, your brokers must be on version Confluent Platform 5.5.x or newer. This implementation is more efficient, because it reduces client and broker resource utilization, like client threads and used network connections, and it enables higher throughput and improved scalability. For more information on how this is done inside the brokers and Kafka Streams, see KIP-447.

A switch from "exactly_once" to "exactly_once_beta" (or the other way around) is only possible if the application is on version Confluent Platform 6.0. If you want to upgrade your application from an older version and enable this feature, you first need to upgrade your application to version Confluent Platform 6.0, staying on "exactly_once", and then do second round of rolling bounces to switch to "exactly_once_beta".

Highly available state stores

For more highly available stateful applications, we’ve modified the task assignment algorithm to delay the movement of stateful active tasks to instances that aren’t yet caught up with that task’s state. Instead, to migrate a task from one instance to another (eg when scaling out), Kafka Streams assigns a warmup replica to the target instance so that it can begin restoring the state while the active task stays available on an instance that already had the task. The instances warming up tasks communicate their progress to the group so that, once ready, Kafka Streams can move active tasks to their new owners in the background. Check out KIP-441 for full details, including several new configs for control over this new feature.

End-to-end latency metrics

New end-to-end latency metrics have been added. These task-level metrics are logged at the INFO level and report the minimum and maximum end-to-end latency of a record at the beginning/source node(s) and end/terminal node(s) of a task. For more information, see KIP-613.

Replace through() with repartition() operator

The operator KStream.through() is deprecated in favor of the new KStream.repartition() operator (via KIP-221). KStream.repartition() is similar to KStream.through(), however Kafka Streams will manage the topic for you. If you need to write into and read back from a topic that you manage, you can fall back to use KStream.to() in combination with StreamsBuilder#stream(). If you only want to write into a topic as “side output” and continue processing, you can also fan-out your dataflow via myStream.xxx(); myStream.to() to avoid the need to read back the data from the cluster.

Streams API changes in Confluent Platform 5.5.0

Confluent Platform 5.5 adds new APIs such as KStream.toTable() transforming a KStream to a KTable, a new cogroup() operator that allows to aggregate multiple streams in a single operation, a new Serde type Void to represent null keys or null values from input topic, a new queryMetadataForKey() that exposes metadata on standby replicas, new KafkaStreams.store(StoreQueryParameters) API to allow performing Kafka Streams Interactive Queries for specific partition and specific task type, deprecated UsePreviousTimeOnInvalidTimestamp and replaced it with UsePartitionTimeOnInvalidTimeStamp.

Co-group

Confluent Platform 5.5 adds a new cogroup() operator (via KIP-150) that enables aggregating multiple streams in a single operation. Cogrouped streams can also be windowed before they are aggregated. Check out the Developer Guide for more details.

Querying Stale Data From Standby

Confluent Platform 5.5 adds new APIs that improve performance and high availability of Kafka Streams Interactive Queries. The KafkaStreams.queryMetadataForKey(store, key, serializer) API returns KeyQueryMetadata object containing both active and standby hosting the specific key, while KafkaStreams.allLocalStorePartitionLags() provides a map of LagInfo objects containing lag/staleness information for every store partition on a given streams instance. Together, these APIs help provide granular control on the KafkaStreams.store(StoreQueryParameters) API, to filter stores only for a specific key’s partition or controlling whether stale stores should be queried.

New toTable() API

In the Confluent Platform 5.5 release, you can transform a Kafka Streams directly into a KTable via KIP-523. This API simplifies the scenario where user needs to take an event stream into a changelog stream. Each record from the input KStream is applied to the result KTable as an insert, update, or delete (if the value is null) for the corresponding key. Records with null keys will be dropped.

Deprecated APIs

In Confluent Platform 5.5, the UsePreviousTimeOnInvalidTimestamp was deprecated and replaced with UsePartitionTimeOnInvalidTimeStamp via KIP-530. This is because we have improved on the stream time tracking, and fixed the definition for partition time.

KafkaStreams.store(String, QueryableStoreType) was deprecated and replaced with KafkaStreams.store(StoreQueryParameters) via KIP-562. The new API gives user more options to route their queries to intended replica and partition.

KafkaStreams.metadataForKey(...) API was deprecated and replaced with KafkaStreams.queryMetadataForKey(...) equivalents, via KIP-535. New APIs also provide standby replica information to help route interactive queries for high availability.

Streams API changes in Confluent Platform 5.4.0

Confluent Platform 5.4.x adds new APIs for specifying topologies and new metrics for monitoring your applications. The API of the unit test utilities has been improved, and some APIs have been deprecated. Additionally, Confluent Platform 5.4.x introduces incremental cooperative rebalancing. See details below.

Foreign key KTable-KTable joins

Previously, you could only join tables on the record keys, which means that both tables had to have the same key space.

Starting in Confluent Platform 5.4.0, you can now perform many-to-one KTable-KTable joins on a foreign-key reference. To use the feature, you specify one of the KTable#join overloads that takes a Function<V, KO> foreignKeyExtractor argument. Kafka Streams will automatically set up the necessary repartition operations internally to compute the correct join result.

For more information on the design, see the design document: KIP-213.

Operator Naming

In the Confluent Platform 5.4.x release, you now can name all operators in a Kafka Streams DSL topology via KIP-307. Giving your operators meaningful names makes it easier to understand the topology description (Topology#describe()#toString()) and understand the full context of what your Kafka Streams application is doing.

There are new overloads on most KStream and KTable methods that accept a Named object. Typically, you’ll provide a name for the DSL operation by using Named.as("my operator name"). Naming of repartition topics for aggregation operations will still use Grouped, and join operations will use either Joined or the new StreamJoined object.

Naming State Stores in a KStream-KStream Join

Before the Confluent Platform 5.4.x version of Kafka Streams, users of the DSL could not name the state stores involved in a stream-stream join. If users changed their topology and added an operator before the join, the internal names of the state stores would shift, requiring an application reset when redeploying. In the Confluent Platform 5.4.x release, Kafka Streams adds the StreamJoined class, which gives users the ability to name the join processor, repartition topic(s) (if a repartition is required), and the state stores involved in the join. Also, by naming the state stores, the changelog topics backing the state stores are named as well. It’s important to note that naming the stores will not make them queryable via Kafka Streams Interactive Queries.

Another feature delivered by StreamJoined is that you can now configure the type of state store used in the join. You can elect to use in-memory stores, custom state stores or continue to use the built in RocksDB stores for a stream-stream join. Note that the provided stores will not be available for querying via Kafka Streams Interactive Queries. With the addition of StreamJoined, stream-stream join operations using Joined have been deprecated. We reccommend you to switch over to stream-stream join methods that use the new overloaded methods.

// Joined class, now deprecated

 left.join(right,
            (value1, value2) -> value1 + value2,
            JoinWindows.of(ofMillis(100)),                     
            Joined.with(Serdes.String(), Serdes.Integer(), Serdes.Integer())
                  // only names repartition topic (if needed)
                  .named("my-join"));



 // StreamJoined 

 left.join(right,
            (value1, value2) -> value1 + value2,
            JoinWindows.of(ofMillis(100)),                                            
            StreamJoined.with(Serdes.String(), Serdes.Integer(), Serdes.Integer())
                         // names the state store.
                        .withStoreName("join-store")
                        // names the processor and repartition topic (if needed)
                        .withName("my-join"));

For more information, see KIP-479.

Metric Improvements

Metrics to monitor RocksDB state stores have been added. The newly added RocksDB metrics collect selected statistics provided by the RocksDB instances. The RocksDB metrics are documented in the monitoring section.

Client metrics have been added to monitor Kafka Streams applications on client-level. The new client metrics are documented in the monitoring section.

Testing Improvements

Kafka Streams’ unit test utilities were improved via KIP-470. Instead of using low level ConsumerRecord<byte[],byte[]> and ProducerRecord<byte[],byte[]> that can be generated with ConsumerRecordFactory and verified with the helper class OutputVerifier, new classes TestInputTopic, TestOutputTopic, and TestRecord were added. The new classes offer an improved API that allows to read/write single (or lists of) values, key-value pairs, or instances of TestRecord. The new TestRecord class allows to set or verify record metadata like headers and timestamp. Furthermore, the new API allows you to use any assertion libary of your choice and thus to write idiomatic test code.

// old API

// use a record factory to pipe data through the test driver
ConsumerRecordFactory<String, Integer> factory = new ConsumerRecordFactory<>(
    "input-topic",
    new StringSerializer(),
    new LongSerializer()
);
testDriver.pipe(factory.create("key", 42L));


// get output record from test driver and verify result
ProducerRecord<String, Integer> outputRecord = testDriver.readOutput(
    "output-topic",
    new StringDeserializer(),
    new LongDeserializer()
);
// throws AssertionError if key or value does not match
OutputVerifier.compareKeyValue(outputRecord, "key", 42L);

// advance wall-clock time using primitive long time (in ms)
testDriver.advanceWallClockTime(10L);


// new API

// use a test-input-topic to pipe data through the test driver
TestInputTopic<String, Long> inputTopic = testDriver.createInputTopic(
    "input-topic",
    new StringSerializer(),
    new LongSerializer()); 
inputTopic.pipeInput("key", 42L);

// use a test-output-topic to get output records
TestOutputTopic<String, Long> outputTopic = testDriver.createOutputTopic(
    "result-topic",
    new StringDeserializer(),
    new LongDeserializer());
// user assertion library of your choice to verify output
assertThat(outputTopic.readKeyValue(), equalTo(new KeyValue<>("a", 21L)));

// advance wall-clock time using expressive Duration
testDriver.advanceWallClockTime(Duration.ofMillis(10L));

Rebalancing State

With the introduction of incremental cooperative rebalancing, Streams no longer requires all tasks to be revoked at the beginning of a rebalance. Instead, at the completion of the rebalance only those tasks which are to be migrated to another consumer for overall load balance will need to be revoked and closed. This changes the semantics of the StateListener a bit, as it will not necessarily transition to REBALANCING at the beginning of a rebalance anymore. Note that this means Interactive Queries will now be available at all times except during state restoration, including while a rebalance is in progress. If restoration is occurring when a rebalance begins, we will continue to actively restore the state stores and/or process standby tasks during a cooperative rebalance. Note that with this new rebalancing protocol, you may sometimes see a rebalance be followed by a second short rebalance that ensures all tasks are safely distributed. Cooperative rebalancing will be turned on by default, and requires a specific upgrade path to safely do a rolling upgrade as described in Upgrading your Kafka Streams applications to Confluent Platform 6.1.15. For details on the new cooperative protocol, see KIP-429.

Deprecated APIs

In Confluent Platform 5.4.x, we deprecated WindowStore#put(K key, V value), which should never be used. Use the existing WindowStore#put(K key, V value, long windowStartTimestamp) instead.

For more information, see KIP-474.

// old API

// similar for Transformer interfaces
class MyProcessor implements Processor {
  // omit some methods for brevity

  WindowStore store;

  public void init(ProcessorContext context) {
    store = (WindowStore) context.getStateStore("myStore");
  }

  public void process(K key, V value) {
    store.put(key, value); // deprecated
  }
}


// new API

// similar for Transformer interfaces
class MyProcessor implements Processor {
  // omit some methods for brevity

  WindowStore store;

  public void init(ProcessorContext context) {
    store = (WindowStore) context.getStateStore("myStore");
  }

  public void process(K key, V value) {
    long windowStartTimestamp = ... // compute window start timestamp explicitly
    store.put(key, value, windowStartTimestamp);
  }
}

Furthermore, the PartitionGrouper interface and its corresponding configuration parameter partition.grouper have been deprecated (KIP-528) and will be removed in the next major Apache Kafka release (KAFKA-7785). Hence, this feature won’t be supported in the future any longer and you need to update your code accordingly before upgrading to the next major release. If you use a custom PartitionGrouper and stop to use it, the created tasks might change. You must reset your application to upgrade it.

Streams API changes in Confluent Platform 5.3.0

Several new APIs have been added in Confluent Platform 5.3.x and a few default configurations have changed. Additionally, the RocksDB dependency was updated. See details below.

Scala Suppress Operator

The Suppress operator has been added to the kafka-streams-scala KTable API.

In-memory Window/Session Stores

Streams now offers an in-memory version of the window and the session store, in addition to the persistent ones based on RocksDB. The new public interfaces inMemoryWindowStore() and inMemorySessionStore() are added to Stores and provide the built-in in-memory window or session store.

Serde close() and configure()

We have added default implementation to close() and configure() for Serializer, Deserializer and Serde so that they can be implemented by lambda expression. For more details please read KIP-331.

Store timestamps in RocksDB

To improve operator semantics, new store types are added that allow storing an additional timestamp per key-value pair or window. Some DSL operators (for example KTables) are using those new stores. Hence, you can now retrieve the last update timestamp via Interactive Queries if you specify TimestampedKeyValueStoreType< or TimestampedWindowStoreType as your QueryableStoreType. While this change is mainly transparent, there are some corner cases that may require code changes: Caution: If you receive an untyped store and use a cast, you might need to update your code to cast to the correct type. Otherwise, you might get an exception similar to java.lang.ClassCastException: class org.apache.kafka.streams.state.ValueAndTimestamp cannot be cast to class YOUR-VALUE-TYPE upon getting a value from the store. Additionally, TopologyTestDriver#getStateStore() only returns non-built-in stores and throws an exception if a built-in store is accessed. For more details please read KIP-258.

New flatTransformValues() Operator

To improve type safety, a new operator KStream#flatTransformValues is added. For more details please read KIP-313.

max.poll.interval default

Kafka Streams used to set the configuration parameter max.poll.interval.ms to Integer.MAX_VALUE. This default value is removed and Kafka Streams uses the consumer default value now. For more details please read KIP-442.

Repartition topic defaults

Default configuration for repartition topic was changed: The segment size for index files segment.index.bytes is no longer 50MB, but uses the cluster default. Similarly, the configuration``segment.ms`` in no longer 10 minutes, but uses the cluster default configuration. Lastly, the retention period (retention.ms) is changed from Long.MAX_VALUE to -1` (infinite). For more details please read KIP-443.

RocksDBConfigSetter close()

To avoid memory leaks, RocksDBConfigSetter has a new close() method that is called on shutdown. Users should implement this method to release any memory used by RocksDB config objects, by closing those objects. More details can be found here.

Upgrade RocksDB to v5.18.3

RocksDB dependency was updated to version 5.18.3. The new version allows to specify more RocksDB configurations, including WriteBufferManager that helps to limit RocksDB off-heap memory usage. For more details please read Memory Management.

Streams API changes in Confluent Platform 5.2.0

Confluent Platform 5.2.x adds some new APIs and improves some existing functionality.

Simplified KafkaStreams Transitions

The KafkaStreams#state transition diagram is simplified in Confluent Platform 5.2.0: in older versions the state will transit from CREATED to RUNNING, and then to REBALANCING to get the first stream task assignment, and then back to RUNNING; starting in 5.2.0 it will transit from CREATED directly to REBALANCING and then to RUNNING. If you have registered a StateListener that captures state transition events, you may need to adjust your listener implementation accordingly for this simplification (in practice, your listener logic should be very unlikely to be affected at all).

TimeWindowSerde

In WindowedSerdes, there is a new static constructor to return a TimeWindowSerde with configurable window size. This is to help users construct time window serdes to read directly from a time-windowed store’s changelog. More details can be found in KIP-393.

AutoCloseable

In Confluent Platform 5.2.0 a few public interfaces have been extended, including KafkaStreams to extend AutoCloseable so that they can be used in a try-with-resource statement. For a full list of public interfaces that are affected, see KIP-376.

Streams API changes in Confluent Platform 5.1.0

New config class Grouped

We’ve added a new class Grouped and deprecated Serialized. The intent of adding Grouped is the ability to name repartition topics created when performing aggregation operations. Users can name the potential repartition topic using the Grouped#as() method which takes a String and is used as part of the repartition topic name. The resulting repartition topic name will still follow the pattern of ${application-id}->name<-repartition. The Grouped class is now favored over Serialized in KStream#groupByKey(), KStream#groupBy(), and KTable#groupBy(). Note that Kafka Streams does not automatically create repartition topics for aggregation operations. Additionally, we’ve updated the Joined class with a new method Joined#withName enabling users to name any repartition topics required for performing Stream/Stream or Stream/Table join.

// old API
KStream<String, String> stream = ...

stream.groupByKey(Grouped.with(Serdes.String(), Serdes.String()))


stream.groupBy( (key, value) -> ... , Grouped.with(Serdes.String(), Serdes.String()))


KStream<String, String> streamII = ...
streamII.join(stream, (valueII, valueI) -> ... , JoinWindows.of(20000), Joined.with(Serdes.String(), 
	                                                                             Serdes.String(), 
	                                                                             Serdes.String()))
KTable<String, String> table = ...
table.groupBy( (key, value) -> ... , Grouped.with(Serdes.String(), Serdes.String()))

// new API
KStream<String, String> stream = ...

stream.groupByKey(Grouped.with(Serdes.String(), Serdes.String()))

                            // providing name for possible reparitiion topic
                            // using a name for the repartition topic is optional
stream.groupByKey(Grouped.with("repartition-topic-name", 
                               Serdes.String(), 
                               Serdes.String())) 

stream.groupBy( (key, value) -> ... , Grouped.with(Serdes.String(), Serdes.String()))

                                      // providing name for possible reparitiion topic
                                      // using a name for the repartition topic is optional
stream.groupBy( (key, value) -> ... , Grouped.with("reparition-topic-name", 
                                                   Serdes.String(), 
                                                   Serdes.String())) 
KStream<String, String> streamII = ...


streamII.join(stream, (valueII, valueI) -> ... , JoinWindows.of(20000), Joined.with(Serdes.String(), 
	                                                                             Serdes.String(), 
	                                                                             Serdes.String()))
                                                              
streamII.join(stream, (valueII, valueI) -> ... , JoinWindows.of(20000), Joined.with(Serdes.String(), 
	                                                                             Serdes.String(), 
	                                                                             Serdes.String(), 
	                                                                             "join-repartition-topic-name")) 
                                                                    // providing name for possible reparitiion topic
                                                                    // using a name for the repartiton topic is optional
KTable<String, String> table = ...

table.groupBy( (key, value) -> ... , Grouped.with(Serdes.String(), Serdes.String()))

                                        // providing name for possible reparitiion topic
                                        // using a name for the repartiton topic is optional
table.groupBy( (key, value) -> ... , Grouped.with("repartition-topic-name", 
                                                  Serdes.String(), 
                                                  Serdes.String())) 

New UUID Serde

We added a new serde for UUIDs (Serdes.UUIDSerde) that you can use via Serdes.UUID().

Improved API Semantics

We updated a list of methods that take long arguments as either timestamp (fix point) or duration (time period) and replaced them with Instant and Duration parameters for improved semantics. Some old methods based on long are deprecated, and users are encouraged to update their code.

In particular, aggregation windows (hopping/tumbling/unlimited time windows and session windows) as well as join windows now take Duration arguments to specify window size, hop, and gap parameters. Also, window sizes and retention times are now specified as Duration type in the Stores class. The Window class has new methods #startTime() and #endTime() that return window start/end timestamp as Instant. For Interactive Queries, there are new #fetch(...) overloads taking Instant arguments. Additionally, punctuations are now registerd via ProcessorContext#schedule(Duration interval, ...).

We deprecated KafkaStreams#close(...) and replaced it with KafkaStreams#close(Duration) that accepts a single timeout argument.

Note

The new KafkaStreams#close(Duration) method has improved (but slightly different) semantics than the old one. It does not block forever if the provided timeout is zero and does not accept any negative timeout values. If you want to block, you should pass in Duration.ofMillis(Long.MAX_VALUE).

// old API

KStream stream = ...

// Tumbling/Hopping Windows
stream.groupByKey().
      .windowedBy(TimeWindows.of(5 * 60 * 1000)
                             .advanceBy(TimeUnit.MINUTES.toMillis(1)),
                  ...);

// Unlimited Windows
stream.groupByKey().
      .windowedBy(UnlimitedWindows.startOn(System.currentTimeMillis()), ...);

// Session Windows
stream.groupByKey().
      .windowedBy(SessionWindows.with(100), ...);

// Joining Streams
KStream secondStream = ...
stream.join(stream2, JoinWindows.of(5000)
                                .before(10000)
                                .after(10000), ...)

// Accessing Window start/end-timestamp
KTable<Windowed<K>, V> table = ...
table.toStream()
     .foreach((wKey, value) -> {
       // can still be used for performance reasons (eg, in a Processor)
       long windowStartTimeMs = wKey.window().start();
       long windowEndTimeMs = wKey.window().end();
       ...
     });

// Registering a Punctuation
MyProcessor implements Processor { // same for Transformer[WithKey] and ValueTransformer[WithKey]
  // other methods omitted for brevity

  void init(ProcessorContext context) {
    context.schedule(1000, PunctuationType.STREAM_TIME, timestamp -> {...});
  }
}

// Interactive Queries
KafkaStreams streams = ...
ReadOnlyWindowStore windowStore = streams.store(...);

long now = System.currentTimeMillis();
WindowStoreIterator<V> it = windowStore.fetch(key, now - 5000, now);
WindowStoreIterator<V> all = windowStore.all(now - 5000, now);

// Creating State Stores
Stores.persistentWindowStore("storeName",
                             24 * 3600 * 1000, // retention period in ms
                             3,                // number of segments
                             10 * 60 * 1000,   // window size in ms
                             false);           // retain duplicates
Stores.persistentSessionStore("storeName",
                              24 * 3600 * 1000); // retention period in ms

// Closing KafkaStreams
KafkaStreams streams = ...
streams.close(30, TimeUnit.SECONDS);
streams.close(0, TimeUnit.SECONDS); // block forever



// new API

KStream stream = ...

// Tumbling/Hopping Windows
stream.groupByKey().
      .windowedBy(TimeWindow.of(Duration.ofMinutes(5)
                            .advanceBy(Duration.ofMinutes(1)),
                  ...);

// Unlimited Windows
stream.groupByKey().
      .windowedBy(UnlimitedWindows.startOn(Instant.now()), ...);

// Session Windows
stream.groupByKey().
      .windowedBy(SessionWindows.with(Duration.ofMillis(100)), ...);

// Joining Streams
KStream secondStream = ...
stream.join(stream2, JoinWindows.of(Duration.ofSeconds(5)
                                .before(Duration.ofSeconds(10))
                                .after(Duration.ofSeconds(10)), ...)

// Accessing Window start/end-timestamp
KTable<Windowed<K>, V> table = ...
table.toStream()
     .foreach((wKey, value) -> {
       // better semantics with new API
       Instant windowStartTime = wKey.window().startTime();
       Instant windowEndTime = wKey.window().endTime();
       // can still be used for performance reasons (eg, in a Processor)
       long windowStartTimeMs = wKey.window().start();
       long windowEndTimeMs = wKey.window().end();
       ...
     });

// Registering a Punctuation
MyProcessor implements Processor { // same for Transformer[WithKey] and ValueTransformer[WithKey]
  // other methods omitted for brevity

  void init(ProcessorContext context) {
    context.schedule(Duration.ofSeconds(1), PunctuationType.STREAM_TIME, timestamp -> {...});
  }
}

// Interactive Queries
KafkaStreams streams = ...
ReadOnlyWindowStore windowStore = streams.store(...);

Instant now = Instant.now();
WindowStoreIterator<V> it = windowStore.fetch(key, now.minus(Duration.ofSeconds(5), now);
WindowStoreIterator<V> all = windowStore.all(now.minus(Duration.ofSeconds(5), now);

// Creating State Stores
Stores.persistentWindowStore("storeName",
                             Duration.ofDay(1),      // retention period
                             // number of segments is removed
                             Duration.ofMinutes(10), // window size
                             false);                 // retain duplicates
Stores.persistentSessionStore("storeName",
                              Duration.ofDay(1)); // retention period

// Closing KafkaStreams
KafkaStreams streams = ...
streams.close(Duration.ofSeconds(30));
streams.close(Duration.ofMillis(Long.MAX_VALUE)); // block forever

Simplified Store Segments

We deprecated the notion of number of segments in window stores and replaced it with segment interval. In the old API, the segment interval was computed as min( retention-period / (number-of-segments - 1) , 60_000L) (with 60 seconds being a lower segment size bound). In the new API, there is no lower bound on the segment interval (i.e., minimum segment interval is 1 millisecond) and the number of segments is 1 + (retention-period / segment-interval).

If you used 3 segments in your store with a retention time of 24 hours, you should now use a segment interval of 12 hours (12 = 24 / (3 - 1)) in the new API to get the same behavior.

Method Windows#segments() and variable Windows#segments were deprecated. Similarly, WindowBytesStoreSupplier#segments() was deprecated and replaced with WindowBytesStoreSupplier#segmentInterval(). If you implement custom windows or custom window stores, be aware that you will need to update your code when those methods are removed.

Finally, Stores#persistentWindowStore(...) was deprecated and replaced with a new overload that does not allow specifying the number of segments any longer because those are computed automatically based on retention time and segment interval. If you create persistent window stores explicitly, you should update your code accordingly.

// Old API

Stores.persistentWindowStore("storeName",
                             24 * 3600 * 1000, // retention period in ms
                             3,                // number of segments
                             10 * 60 * 1000,   // window size in ms
                             false);           // retain duplicates
// New API

Stores.persistentWindowStore("storeName",
                             Duration.ofDay(1),      // retention period (updated from `long` to `Duration`)
                             // number of segments is removed
                             Duration.ofMinutes(10), // window size (updated from `long` to `Duration`)
                             false);                 // retain duplicates

Out-of-Order Data Handling

We added a new config (max.task.idle.ms) to allow users to specify how to handle out-of-order data within a task that may be processing multiple Kafka topic partitions (see the Out-of-Order Handling section for more details). The default value is set to 0, to favor minimized processing latency. If users would like to wait on processing when only part of the topic partitions of a given task have data available in order to reduce risks of handling out-of-order data, they can override this config to a larger value.

AdminClient Metrics Exposed

The newly exposed AdminClient metrics are now included with other available metrics when calling the KafkaStream#metrics() method. For more details on monitoring streams applications check out Monitoring Streams Applications.

Topology Description Improved

We updated the <code>TopologyDescription</code> API to allow for better runtime checking. Users are encouraged to use <code>#topicSet()</code> and <code>#topicPattern()</code> accordingly on <code>TopologyDescription.Source</code> nodes, instead of using <code>#topics()</code>, which has since been deprecated. Similarly, use <code>#topic()</code> and <code>#topicNameExtractor()</code> to get descriptions of <code>TopologyDescription.Sink</code> nodes.

// old API

TopologyDescription.Source source = ... // get Source node from a TopologyDescription
TopologyDescription.Sink sink = ... // get Sink node from a TopologyDescription

String topics = source.topics(); // comma separated list of topic names or pattern (as String)

String topic = sink.topic(); // return the output topic name (never null)


// new API

TopologyDescription.Source source = ... // get Source node from a TopologyDescription
TopologyDescription.Sink sink = ... // get Sink node from a TopologyDescription


Set<String> topics = source.topicSet(); // set of all topic names (can be null if pattern subscription is uses)
// or
Pattern pattern = source.topicPattern(); // topic pattern (can be null)

String topic = sink.topic(); // return the output topic name (can be null if dynamic topic routing is used)
// or
TopicNameExtractor topicNameExtractor = sink.topicNameExtractor(); // return the use TopicNameExtractor (can be null)

StreamsBuilder#build Method Overload

We’ve added an overloaded StreamsBuilder#build method that accepts an instance of java.util.Properties with the intent of using the``StreamsConfig#TOPOLOGY_OPTIMIZATION`` config added in Kafka Streams 2.0. Before 2.1, when building a topology with the DSL, Kafka Streams writes the physical plan as the user makes calls on the DSL. Now, by providing a java.util.Properties instance when executing a StreamsBuilder#build call, Kafka Streams can optimize the physical plan of the topology, provided the StreamsConfig#TOPOLOGY_OPTIMIZATION config is set to StreamsConfig#OPTIMIZE. By setting StreamsConfig#OPTIMIZE in addition to the KTable optimization of reusing the source topic as the changelog topic, the topology may be optimized to merge redundant repartition topics into one repartition topic. The original no parameter version of StreamsBuilder#build is still available if you don’t want to optimize your topology. Note that enabling optimization of the topology may require you to do an application reset when redeploying the application. For more details, see Optimizing Kafka Streams

Streams API changes in Confluent Platform 5.0

A few new Streams configurations and public interfaces are added into Confluent Platform 5.0.x release. Additionally, some deprecated APIs are removed in Confluent Platform 5.0.x release.

Skipped Records Metrics Refactored

Starting with Confluent Platform 5.0.0, Kafka Streams does not report the skippedDueToDeserializationError-rate and skippedDueToDeserializationError-total metrics.

Deserialization errors, and all other causes of record skipping, are now accounted for in the pre-existing metrics skipped-records-rate and skipped-records-total. When a record is skipped, the event is now logged at WARN level. Note these metrics are mainly for monitoring unexpected events; If there are systematic issues that caused too many unprocessable records to be skipped, and hence the resulted warning logs become burdensome, you should consider filtering our these unprocessable records instead of depending on record skipping semantics. For more details, see KIP-274.

As of right now, the potential causes of skipped records are:

• null keys in table sources.
• null keys in table-table inner/left/outer/right joins.
• null keys or values in stream-table joins.
• null keys or values in stream-stream joins.
• null keys or values in aggregations / reductions / counts on grouped streams.
• null keys in aggregations / reductions / counts on windowed streams.
• null keys in aggregations / reductions / counts on session-windowed streams.
• Errors producing results, when the configured default.production.exception.handler decides to CONTINUE (the default is to FAIL and throw an exception).
• Errors deserializing records, when the configured default.deserialization.exception.handler decides to CONTINUE (the default is to FAIL and throw an exception). This was the case previously captured in the skippedDueToDeserializationError metrics.
• Fetched records having a negative timestamp.

New Functions in Window Store Interface

Confluent Platform now supports methods in ReadOnlyWindowStore which allows you to query the key-value pair of a single window. If you have customized window store implementations on the above interface, you must update your code to implement the newly added method. For more details, see KIP-261.

Simplified KafkaStreams Constructor

The KafakStreams constructor was simplfied. Instead of requiring the user to create a boilderplate StreamsConfig object, the constructor now directly accepts the Properties object that specifies the actual user configuration.

StreamsBuilder builder = new StreamsBuilder();
// define processing logic
Topology topology = builder.build();

// or

Topology topology = new Topology();
// define processing logic


Properties props = new Properties();
// define configuration


// old API

KafkaStream stream = new KafkaStreams(topology, new StreamsConfig(props));
KafkaStream stream = new KafkaStreams(topology, new StreamsConfig(props), /* pass in KafkaClientSupplier or Time */);


// new API

KafkaStream stream = new KafkaStreams(topology, props);
KafkaStream stream = new KafkaStreams(topology, props, /* pass in KafkaClientSupplier or Time */);

Support Dynamic Routing at Sink

In this release you can now dynamically route records to Kafka topics. More specifically, in both the lower-level Topology#addSink and higher-level KStream#to APIs, we have added variants that take a TopicNameExtractor instance instead of a specific String topic name. For each record received from the upstream processor, the TopicNameExtractor will dynamically determine which Kafka topic to write to based on the record’s key and value, as well as record context. Note that all output Kafka topics are still considered user topics and hence must be pre-created. Also, we have modified the StreamPartitioner interface to add the topic name parameter since the topic name now may not be known beforehand; users who have customized implementations of this interface would need to update their code while upgrading their application.

Support Message Headers

In this release there is message header support in the Processor API. In particular, we have added a new API ProcessorContext#headers() which returns a Headers object that keeps track of the headers of the source topic’s message that is being processed. Through this object, users can manipulate the headers map that is being propagated throughout the processor topology as well, for example Headers#add(String key, byte[] value) and Headers#remove(String key). When Streams DSL is used, users can call process or transform in which they can also access the ProcessorContext to access and manipulate the message header; if user does not manipulate the header, it will still be preserved and forwarded while the record traverses through the processor topology. When the resulted record is sent to the sink topics, the preserved message header will also be encoded in the sent record.

KTable Now Supports Transform Values

In this release another new API, KTable#transformValues, was added. For more information, see KIP-292.

Improved Windowed Serde Support

We added helper class WindowedSerdes that allows you to create time- and session-windowed serdes without the need to know the details how windows are de/serialized. The created window serdes wrap a user-provided serde for the inner key- or value-data type. Furthermore, two new configs default.windowed.key.serde.inner and default.windowed.value.serde.inner were added that allow to specify the default inner key- and value-serde for windowed types. Note, these new configs are only effective, if default.key.serde or default.value.serde specifies a windowed serde (either WindowedSerdes.TimeWindowedSerde or WindowedSerdes.SessionWindowedSerde).

Allow Timestamp Manipulation

Using the Processor API, it is now possible to set the timestamp for output messages explicitly. This change implies updates to the ProcessorContext#forward() method. Some existing methods were deprecated and replaced by new ones. In particular, it is not longer possible to send records to a downstream processor based on its index.

// old API

public class MyProcessor implements Processor<String, Integer> {
  private ProcessorContext context;

  @Override
  public void init(ProcessorContext context) {
    this.context = context;
  }

  @Override
  public void process(String key, Integer value) {
    // do some computation

    // send record to all downstream processors
    context.forward(newKey, newValue);
    // send record to particular downstream processors (if it exists; otherwise drop record)
    context.forward(newKey, newValue, "downstreamProcessorName");
    // send record to particular downstream processors per index (throws if index is invalid)
    int downStreamProcessorIndex = 2;
    context.forward(newKey, newValue, downstreamProcessorIndex);
  }

  @Override
  public void close() {} // nothing to do
}


// new API

public class MyProcessor implements Processor<String, Integer> {
  // omit other methods that don't change for brevity

  @Override
  public void process(String key, Integer value) {
    // do some computation

    // send record to all downstream processors
    context.forward(newKey, newValue); // same as old API
    context.forward(newKey, newValue, To.all()); // new; same as line above
    // send record to particular downstream processors (if it exists; otherwise drop record)
    context.forward(newKey, newValue, To.child("downstreamProcessorName"));
    // send record to particular downstream processors per index (throws if index is invalid)
    // -> not supported in new API

    // new: set record timestamp
    long outputRecordTimestamp = 42L;
    context.forward(newKey, newValue, To.all().withTimestamp(outputRecordTimestamp));
    context.forward(newKey, newValue, To.child("downstreamProcessorName").withTimestamp(outputRecordTimestamp));
  }
}

Public Test-Utils Artifact

Confluent Platform now ships with a kafka-streams-test-uitls artifact that contains utility classes to unit test your Kafka Streams application. Check out Testing Streams Code section for more details.

Scala API

Confluent Platform now ships with the Apache Kafka Scala API for Kafka Streams. You can add the dependency for Scala 2.11 or 2.12 artifacts:

<dependency>
    <groupId>org.apache.kafka</groupId>
    <artifactId>kafka-streams-scala_2.11</artifactId>
    <!-- or Scala 2.12
    <artifactId>kafka-streams-scala_2.12</artifactId>
    -->
    <version>2.0.0-cp1</version>
</dependency>

Deprecated APIs are Removed

The following deprecated APIs are removed in Confluent Platform 5.0.0:

1. KafkaStreams#toString no longer returns the topology and runtime metadata; to get topology metadata you can call Topology#describe(), and to get thread runtime metadata you can call KafkaStreams#localThreadsMetadata (deprecated since Confluent Platform 4.0.0). For detailed guidance on how to update your code please read here.
2. TopologyBuilder and KStreamBuilder are removed and replaced by Topology and StreamsBuidler respectively (deprecated since Confluent Platform 4.0.0).
3. StateStoreSupplier are removed and replaced with StoreBuilder (deprecated since Confluent Platform 4.0.0); and the corresponding Stores#create and KStream, KTable, KGroupedStream’s overloaded functions that use it have also been removed.
4. KStream, KTable, KGroupedStream overloaded functions that requires serde and other specifications explicitly are removed and replaced with simpler overloaded functions that use Consumed, Produced, Serialized, Materialized, Joined (deprecated since Confluent Platform 4.0.0).
5. Processor#punctuate, ValueTransformer#punctuate, ValueTransformer#punctuate and RecordContext#schedule(long) are removed and replaced by RecordContext#schedule(long, PunctuationType, Punctuator) (deprecated since Confluent Platform 4.0.0).
6. The second boolean typed parameter loggingEnabled in ProcessorContext#register has been removed; you can now use StoreBuilder#withLoggingEnabled, #withLoggingDisabled to specify the behavior when they create the state store (deprecated since Confluent Platform 3.3.0).
7. KTable#writeAs, #print, #foreach, #to, #through are removed as their semantics are more confusing than useful, you can call KTable#tostream()#writeAs etc instead for the same purpose (deprecated since Confluent Platform 3.3.0).
8. StreamsConfig#KEY_SERDE_CLASS_CONFIG, #VALUE_SERDE_CLASS_CONFIG, #TIMESTAMP_EXTRACTOR_CLASS_CONFIG are removed and replaced with StreamsConfig#DEFAULT_KEY_SERDE_CLASS_CONFIG, #DEFAULT_VALUE_SERDE_CLASS_CONFIG, #DEFAULT_TIMESTAMP_EXTRACTOR_CLASS_CONFIG respectively (deprecated since Confluent Platform 3.3.0).
9. StreamsConfig#ZOOKEEPER_CONNECT_CONFIG is removed as we do not need ZooKeeper dependency in Streams any more (deprecated since Confluent Platform 3.2.0).

Streams API changes in Confluent Platform 4.1

A few new Streams configurations and public interfaces are added into Confluent Platform 4.1.x release.

Changes in bin/kafka-streams-application-reset

Added options to specify input topics offsets to reset according to KIP-171.

Embedded Admin Client Configuration

You can now customize the embedded admin client inside your Streams application which would be used to send all the administrative requests to Kafka brokers, such as internal topic creation, etc. This is done via the additional KafkaClientSupplier#getAdminClient(Map<String, Object>) interface; for example, users can provide their own AdminClient implementations to override the default ones in their integration testing. In addition, users can also override the configs that are passed into KafkaClientSupplier#getAdminClient(Map<String, Object>) to configure the returned AdminClient. Such overridden configs can be specified via the StreamsConfig by adding the admin configs with the prefix as defined by StreamsConfig#adminClientPrefix(String). Any configs that aren’t admin client configs will be ignored.

For example:

Properties streamProps = ...;
// use retries=10 for the embedded admin client
streamsProps.put(StreamsConfig.adminClientPrefix("retries"), 10);

Streams API changes in Confluent Platform 4.0

Kafka Streams and its API were improved and modified in the Confluent Platform 4.0.x release. All of these changes are backward compatible, thus it’s not require to update the code of your Kafka Streams applications immediately. However, some methods were deprecated and thus it is recommend to update your code eventually to allow for future upgrades. In this section we focus on deprecated APIs.

Building and running a topology

The two main classes to specify a topology, KStreamBuilder and TopologyBuilder, were deprecated and replaced by StreamsBuilder and Topology. Note, that both new classes are in package org.apache.kafka.streams and that StreamsBuilder does not extend Topology, i.e., the class hierarchy is different now. This change also affects KafkaStreams constructors that now only accept a Topology. If you use StreamsBuilder you can obtain the constructed topology via StreamsBuilder#build().

The new classes have basically the same methods as the old ones to build a topology via DSL or Processor API. However, some internal methods that were public in KStreamBuilder and TopologyBuilder, but not part of the actual API, are no longer included in the new classes.

// old API

KStreamBuilder builder = new KStreamBuilder(); // for DSL
// or
TopologyBuilder builder = new TopologyBuilder(); // for Processor API

Properties props = new Properties();
KafkaStreams streams = new KafkaStreams(builder, props);

// new API

StreamsBuilder builder = new StreamsBuilder(); // for DSL
// ... specify computational logic
Topology topology = builder.build();
// or
Topology topology = new Topology(); // for Processor API

Properties props = new Properties();
KafkaStreams streams = new KafkaStreams(topology, props);

Describing topology and stream task metadata

KafkaStreams#toString() and KafkaStreams#toString(final String indent), which were previously used to retrieve the user-specified processor topology information as well as runtime stream tasks metadata, are deprecated in 4.0.0. Instead, a new method of KafkaStreams, namely localThreadsMetadata() is added which returns an org.apache.kafka.streams.processor.ThreadMetadata object for each of the local stream threads that describes the runtime state of the thread as well as its current assigned tasks metadata. Such information will be very helpful in terms of debugging and monitoring your streams applications. For retrieving the specified processor topology information, users can now call Topology#describe() which returns an org.apache.kafka.streams.TopologyDescription object that contains the detailed description of the topology (for DSL users they would need to call StreamsBuilder#build() to get the Topology object first).

Merging KStreams:

As mentioned above, KStreamBuilder was deprecated in favor of StreamsBuilder. Additionally, KStreamBuilder#merge(KStream...) was replaced by KStream#merge(KStream) and thus StreamsBuilder does not have a merge() method. Note: instead of merging an arbitrary number of KStream instances into a single KStream as in the old API, the new #merge() method only accepts a single KStream and thus merges two KStream instances into one. If you want to merge more than two KStream instances, you can call KStream#merge() multiple times.

// old API

KStreamBuilder builder = new KStreamBuilder();

KStream<Long, String> firstStream = ...;
KStream<Long, String> secondStream = ...;
KStream<Long, String> thirdStream = ...;

KStream<Long, String> mergedStream = builder.merge(
  firstStream,
  secondStream,
  thirdStream);

// new API

StreamsBuilder builder = new StreamsBuilder();

KStream<Long, String> firstStream = ...;
KStream<Long, String> secondStream = ...;
KStream<Long, String> thirdStream = ...;

KStream<Long, String> mergedStream = firstStream.merge(secondStream)
                                                .merge(thirdStream);

Punctuation functions

The Processor API was extended to allow users to schedule punctuate functions either based on event-time (i.e. PunctuationType.STREAM_TIME) or wall-clock-time (i.e. PunctuationType.WALL_CLOCK_TIME). Before this, users could only schedule based on event-time and hence the punctuate function was data-driven only. As a result, the original ProcessorContext#schedule is deprecated with a new overloaded function. In addition, the punctuate function inside Processor is also deprecated, and is replaced by the newly added Punctuator#punctuate interface.

// old API (punctuate defined in Processor, and schedule only with stream-time)

public class WordCountProcessor implements Processor<String, String> {

    private ProcessorContext context;
    private KeyValueStore<String, Long> kvStore;

    @Override
    @SuppressWarnings("unchecked")
    public void init(ProcessorContext context) {
        // keep the processor context locally because we need it in punctuate() and commit()
        this.context = context;

        // call this processor's punctuate() method every 1000 milliseconds
        this.context.schedule(1000);

        // retrieve the key-value store named "Counts"
        kvStore = (KeyValueStore) context.getStateStore("Counts");
    }

    @Override
    public void punctuate(long timestamp) {
        KeyValueIterator<String, Long> iter = this.kvStore.all();
        while (iter.hasNext()) {
            KeyValue<String, Long> entry = iter.next();
            context.forward(entry.key, entry.value.toString());
        }
        iter.close();

        // commit the current processing progress
        context.commit();
    }

    // .. other functions
}


// new API (punctuate defined in Punctuator, and schedule can be either stream-time or wall-clock-time)

public class WordCountProcessor implements Processor<String, String> {

    private ProcessorContext context;
    private KeyValueStore<String, Long> kvStore;

    @Override
    @SuppressWarnings("unchecked")
    public void init(ProcessorContext context) {
        // keep the processor context locally because we need it in punctuate() and commit()
        this.context = context;

        // retrieve the key-value store named "Counts"
        kvStore = (KeyValueStore) context.getStateStore("Counts");

        // schedule a punctuate() method every 1000 milliseconds based on stream time
        this.context.schedule(1000, PunctuationType.STREAM_TIME, (timestamp) -> {
            KeyValueIterator<String, Long> iter = this.kvStore.all();
            while (iter.hasNext()) {
                KeyValue<String, Long> entry = iter.next();
                context.forward(entry.key, entry.value.toString());
            }
            iter.close();

            // commit the current processing progress
            context.commit();
        });
    }

    // .. other functions
}

Streams Configuration

You can now override the configs that are used to create internal repartition and changelog topics. You provide these configs via the StreamsConfig by adding the topic configs with the prefix as defined by StreamsConfig#topicPrefix(String). Any properties in the StreamsConfig with the prefix will be applied when creating internal topics. Any configs that aren’t topic configs will be ignored. If you are already using StateStoreSupplier or Materialized to provide configs for changelogs, then they will take precedence over those supplied in the config.

For example:

Properties streamProps = ...;
// use cleanup.policy=delete for internal topics
streamsProps.put(StreamsConfig.topicPrefix("cleanup.policy"), "delete");

New classes for optional DSL parameters

Several new classes were introduced, i.e., Serialized, Consumed, Produced etc. to enable us to reduce the overloads in the DSL. These classes mostly have a static method with to create an instance, i.e., Serialized.with(Serdes.Long(), Serdes.String()).

Scala users should be aware that they will need to surround with with backticks.

For example:

// When using Scala: enclose "with" with backticks
Serialized.`with`(Serdes.Long(), Serdes.String())

Streams API changes in Confluent Platform 3.3

Kafka Streams and its API were improved and modified since the release of Confluent Platform 3.2.x. All of these changes are backward compatible, thus it’s not require to update the code of your Kafka Streams applications immediately. However, some methods and configuration parameters were deprecated and thus it is recommend to update your code eventually to allow for future upgrades. In this section we focus on deprecated APIs.

Streams Configuration

The following configuration parameters were renamed and their old names were deprecated.

• key.serde renamed to default.key.serde
• value.serde renamed to default.value.serde
• timestamp.extractor renamed to default.timestamp.extractor

Thus, StreamsConfig#KEY_SERDE_CONFIG, StreamsConfig#VALUE_SERDE_CONFIG, and StreamsConfig#TIMESTAMP_EXTRACTOR_CONFIG were deprecated, too.

Additionally, the following method changes apply:

• method keySerde() was deprecated and replaced by defaultKeySerde()
• method valueSerde() was deprecated and replaced by defaultValueSerde()
• new method defaultTimestampExtractor() was added

Local timestamp extractors

The Streams API was extended to allow users to specify a per stream/table timestamp extractor. This simplifies the usage of different timestamp extractor logic for different streams/tables. Before, users needed to apply an if-then-else pattern within the default timestamp extractor to apply different logic to different input topics. The old behavior introduced unnecessary dependencies and thus limited code modularity and code reuse.

To enable the new feature, the methods KStreamBuilder#stream(), KStreamBuilder#table(), KStream#globalTable(), TopologyBuilder#addSource(), and TopologyBuilder#addGlobalStore() have new overloads that allow to specify a “local” timestamp extractor that is solely applied to the corresponding input topics.

// old API (single default TimestampExtractor that is applied globally)

public class MyTimestampExtractor implements TimestampExtractor {
    @Override
    public long extract(ConsumerRecord record, long previousTimestamp) {
        long timestamp;

        String topic = record.topic();
        switch (topic) {
            case "streamInputTopic":
                timestamp = record.value().getDataTimestamp(); // assuming that value type has a method #getDataTimestamp()
                break;
            default:
                timestamp = record.timestamp();
        }

        if (timestamp < 0) {
            throw new RuntimeException("Invalid negative timestamp.");
        }

        return timestamp;
    }
}

KStreamBuilder builder = new KStreamBuilder();
KStream stream = builder.stream(keySerde, valueSerde, "streamInputTopic");
KTable table= builder.table("tableInputTopic");

Properties props = new Properties(); // omitting mandatory configs for brevity
// set MyTimestampExtractor as global default extractor for all topics
config.set("default.timestamp.extractor", MyTimestampExtractor.class);

KafkaStreams streams = new KafkaStreams(builder, props);


// new API (custom TimestampExtractor for topic "streamInputTopic" only; returns value embedded timestamp)

public class StreamTimestampExtractor implements TimestampExtractor {
    @Override
    public long extract(ConsumerRecord record, long previousTimestamp) {
        long timestamp = record.value().getDataTimestamp(); // assuming that value type has a method #getDataTimestamp()

        if (timestamp < 0) {
            throw new RuntimeException("Invalid negative timestamp.");
        }

        return timestamp;
    }
}

KStreamBuilder builder = new KStreamBuilder();
// set StreamTimestampExtractor explicitly for "streamInputTopic"
KStream stream = builder.stream(new StreamTimestampExtractor(), keySerde, valueSerde, "streamInputTopic");
KTable table= builder.table("tableInputTopic");

Properties props = new Properties(); // omitting mandatory configs for brevity

KafkaStreams streams = new KafkaStreams(builder, props);

KTable Changes

The following methods have been deprecated on the KTable interface

• void foreach(final ForeachAction<? super K, ? super V> action)
• void print()
• void print(final String streamName)
• void print(final Serde<K> keySerde, final Serde<V> valSerde)
• void print(final Serde<K> keySerde, final Serde<V> valSerde, final String streamName)
• void writeAsText(final String filePath)
• void writeAsText(final String filePath, final Serde<K> keySerde, final Serde<V> valSerde)
• void writeAsText(final String filePath, final String streamName)
• void writeAsText(final String filePath, final String streamName, final Serde<K> keySerde, final Serde<V> valSerde)

These methods have been deprecated in favor of using the Interactive Queries API.

If you want to query the current content of the state store backing the KTable, use the following approach:

• Make a call to KafkaStreams.store(String storeName, QueryableStoreType<T> queryableStoreType) followed by a call to ReadOnlyKeyValueStore.all() to iterate over the keys of a KTable.

If you want to view the changelog stream of the KTable then you could do something along the lines of the following:

• Call KTable.toStream() then call KStream#print().

Streams API changes in Confluent Platform 3.2

Kafka Streams and its API were improved and modified since the release of Confluent Platform 3.1.x. Some of these changes are breaking changes that require you to update the code of your Kafka Streams applications. In this section we focus on only these breaking changes.

Handling Negative Timestamps and Timestamp Extractor Interface

Kafka Streams behavior with regard to invalid (i.e., negative) timestamps was improved. By default you will still get an exception on an invalid timestamp. However, you can reconfigure your application to react more gracefully to invalid timestamps which was not possible before.

Even if you do not use a custom timestamp extractor, you need to recompile your application code, because the TimestampExtractor interface was changed in an incompatible way.

The internal behavior of Kafka Streams with regard to negative timestamps was changed. Instead of raising an exception if the timestamp extractor returns a negative timestamp, the corresponding record will be dropped silently and not be processed. This allows to process topic for which only a few records cannot provide a valid timestamp. Furthermore, the TimestampExtractor interface was changed and now has one additional parameter. This parameter provides a timestamp that can be used, for example, to return an estimated timestamp, if no valid timestamp can be extracted from the current record.

The old default timestamp extractor ConsumerRecordTimestampExtractor was replaced with FailOnInvalidTimestamp, and two new extractors which both extract a record’s built-in timestamp were added (LogAndSkipOnInvalidTimestamp and UsePreviousTimeOnInvalidTimestamp). The new default extractor (FailOnInvalidTimestamp) raises an exception in case of a negative built-in record timestamp such that Kafka Streams’ default behavior is kept (i.e., fail-fast on negative timestamp). The two newly added extractors allow to handle negative timestamp more gracefully by implementing a log-and-skip and timestamp-estimation strategy.

// old interface
public class TimestampExtractor {
  // returning -1 results in an exception
  long extract(ConsumerRecord<Object, Object> record);
}

// new interface
public class TimestampExtractor {
  // provides a timestamp that could be used as a timestamp estimation,
  // if no valid timestamp can be extracted from the current record
  //
  // allows to return -1, which tells Kafka Streams to not process the record (it will be dropped silently)
  long extract(ConsumerRecord<Object, Object> record, long previousTimestamp);
}

Metrics

If you provide custom metrics by implementing interface StreamsMetrics you need to update your code as the interface has many new methods allowing to register finer grained metrics than before. More details are available in KIP-114.

// old interface
public interface StreamsMetrics {
  // Add the latency sensor.
  Sensor addLatencySensor(String scopeName, String entityName, String operationName, String... tags);

  // Record the given latency value of the sensor.
  void recordLatency(Sensor sensor, long startNs, long endNs);
}

// new interface
public interface StreamsMetrics {
  // Get read-only handle on global metrics registry.
  Map<MetricName, ? extends Metric> metrics();

  // Add a latency and throughput sensor for a specific operation
  Sensor addLatencyAndThroughputSensor(final String scopeName,
                                       final String entityName,
                                       final String operationName,
                                       final Sensor.RecordingLevel recordingLevel,
                                       final String... tags);

  // Record the given latency value of the sensor.
  void recordLatency(final Sensor sensor,
                     final long startNs,
                     final long endNs);

  // Add a throughput sensor for a specific operation:
  Sensor addThroughputSensor(final String scopeName,
                             final String entityName,
                             final String operationName,
                             final Sensor.RecordingLevel recordingLevel,
                             final String... tags);

  // Record the throughput value of a sensor.
  void recordThroughput(final Sensor sensor,
                        final long value);

  // Generic method to create a sensor.
  Sensor addSensor(final String name,
                   final Sensor.RecordingLevel recordingLevel);

  // Generic method to create a sensor with parent sensors.
  Sensor addSensor(final String name,
                   final Sensor.RecordingLevel recordingLevel,
                   final Sensor... parents);

  // Remove a sensor.
  void removeSensor(final Sensor sensor);
}

Scala

Starting with 0.10.2.0, if your application is written in Scala, you may need to declare types explicitly in order for the code to compile. The StreamToTableJoinScalaIntegrationTest has an example where the types of return variables are explicitly declared.

Streams API changes in Confluent Platform 3.1

Stream grouping and aggregation

Grouping (i.e., repartitioning) and aggregation of the KStream API was significantly changed to be aligned with the KTable API. Instead of using a single method with many parameters, grouping and aggregation is now split into two steps. First, a KStream is transformed into a KGroupedStream that is a repartitioned copy of the original KStream. Afterwards, an aggregation can be performed on the KGroupedStream, resulting in a new KTable that contains the result of the aggregation.

Thus, the methods KStream#aggregateByKey(...), KStream#reduceByKey(...), and KStream#countByKey(...) were replaced by KStream#groupBy(...) and KStream#groupByKey(...) which return a KGroupedStream. While KStream#groupByKey(...) groups on the current key, KStream#groupBy(...) sets a new key and re-partitions the data to build groups on the new key. The new class KGroupedStream provides the corresponding methods aggregate(...), reduce(...), and count(...).

KStream stream = builder.stream(...);
Reducer reducer = new Reducer() { /* ... */ };

// old API
KTable newTable = stream.reduceByKey(reducer, name);

// new API, Group by existing key
KTable newTable = stream.groupByKey().reduce(reducer, name);
// or Group by a different key
KTable otherTable = stream.groupBy((key, value) -> value).reduce(reducer, name);

Auto Repartitioning

Previously when performing KStream#join(...), KStream#outerJoin(...) or KStream#leftJoin(...) operations after a key changing operation, i.e, KStream#map(...), KStream#flatMap(...), KStream#selectKey(...) the developer was required to call KStream#through(...) to repartition the mapped KStream This is no longer required. Repartitioning now happens automatically for all join operations.

KStream streamOne = builder.stream(...);
KStream streamTwo = builder.stream(...);
KeyValueMapper selector = new KeyValueMapper() { /* ... */ };
ValueJoiner joiner = new ValueJoiner { /* ... */ };
JoinWindows windows = JoinWindows.of("the-join").within(60 * 1000);

// old API
KStream oldJoined = streamOne.selectKey(selector)
                             .through("repartitioned-topic")
                             .join(streamTwo,
                                   joiner,
                                   windows);

// new API
KStream newJoined = streamOne.selectKey((key,value) -> value)
                             .join(streamTwo,
                                   joiner,
                                   windows);

TopologyBuilder

Two public method signatures have been changed on TopologyBuilder, TopologyBuilder#sourceTopics(String applicationId) and TopologyBuilder#topicGroups(String applicationId). These methods no longer take applicationId as a parameter and instead you should call TopologyBuilder#setApplicationId(String applicationId) before calling one of these methods.

TopologyBuilder builder = new TopologyBuilder();
...

// old API
Set<String> topics = topologyBuilder.sourceTopics("applicationId");
Map<Integer, TopicsInfo> topicGroups = topologyBuilder.topicGroups("applicationId");

// new API
topologyBuilder.setApplicationId("applicationId");
Set<String> topics = topologyBuilder.sourceTopics();
Map<Integer, TopicsInfo> topicGroups = topologyBuilder.topicGroups();

DSL: New parameters to specify state store names

Apache Kafka 0.10.1 introduces Kafka Streams Interactive Queries, which allow you to directly query state stores of a Kafka Streams application. This new feature required a few changes to the operators in the DSL. Starting with Kafka 0.10.1, state stores must be always be “named”, which includes both explicitly used state stores (e.g., defined by the user) and internally used state stores (e.g., created behind the scenes by operations such as count()). This naming is a prerequisite to make state stores queryable. As a result of this, the previous “operator name” is now the state store name. This change affects KStreamBuilder#table(...) and windowed aggregates KGroupedStream#count(...), #reduce(...), and #aggregate(...).

// old API
builder.table("topic");
builder.table(keySerde, valSerde, "topic");

table2 = table1.through("topic");

stream.countByKey(TimeWindows.of("windowName", 1000)); // window has a name

// new API
builder.table("topic", "storeName"); // requires to provide a store name to make KTable queryable
builder.table(keySerde, valSerde, "topic", "storeName"); // requires to provide a store name to make KTable queryable

table2 = table1.through("topic", "storeName"); // requires to provide a store name to make KTable queryable

// for changes of countByKey() -> groupByKey().count(...), please see example above
// for changes of TimeWindows.of(...), please see example below
stream.groupByKey().count(TimeWindows.of(1000), "countStoreName"); // window name removed, store name added

Windowing

The API for JoinWindows was improved. It is not longer possible to define a window with a default size (of zero). Furthermore, windows are not named anymore. Rather, any such naming is now done for state stores. See section DSL: New parameters to specify state store names above).

// old API
JoinWindows.of("name"); // defines window with size zero
JoinWindows.of("name").within(60 * 1000L);

TimeWindows.of("name", 60 * 1000L);
UnlimitedWindows.of("name", 60 * 1000L);

// new API, no name, requires window size
JoinWindows.of(0); // no name; set window size explicitly to zero
JoinWindows.of(60 * 1000L); // no name

TimeWindows.of(60 * 1000L); // not required to specify a name anymore
UnlimitedWindows.of(); // not required to specify a name anymore