Kafka Streams is a client library for processing and analyzing data stored in Kafka and either write the resulting data back to Kafka or send the final output to an external system. It builds upon important stream processing concepts such as properly distinguishing between event time and processing time, windowing support, and simple yet efficient management of application state.
Kafka Streams has a <b>low barrier to entry</b>: You can quickly write and run a small-scale proof-of-concept on a single machine; and you only need to run additional instances of your application on multiple machines to scale up to high-volume production workloads. Kafka Streams transparently handles the load balancing of multiple instances of the same application by leveraging Kafka's parallelism model.
</p>
<p>
Some highlights of Kafka Streams:
</p>
<ul>
<li>Designed as a <b>simple and lightweight client library</b>, which can be easily embedded in any Java application and integrated with any existing packaging, deployment and operational tools that users have for their streaming applications.</li>
<li>Has <b>no external dependencies on systems other than Apache Kafka itself</b> as the internal messaging layer; notably, it uses Kafka's partitioning model to horizontally scale processing while maintaining strong ordering guarantees.</li>
<li>Employs <b>one-record-at-a-time processing</b> to achieve millisecond processing latency, and supports <b>event-time based windowing operations</b> with late arrival of records.</li>
<li>A <b>stream</b> is the most important abstraction provided by Kafka Streams: it represents an unbounded, continuously updating data set. A stream is an ordered, replayable, and fault-tolerant sequence of immutable data records, where a <b>data record</b> is defined as a key-value pair.</li>
<li>A <b>stream processing application</b> is any program that makes use of the Kafka Streams library. It defines its computational logic through one or more <b>processor topologies</b>, where a processor topology is a graph of stream processors (nodes) that are connected by streams (edges).</li>
<li>A <b>stream processor</b> is a node in the processor topology; it represents a processing step to transform data in streams by receiving one input record at a time from its upstream processors in the topology, applying its operation to it, and may subsequently produce one or more output records to its downstream processors. </li>
<li><b>Source Processor</b>: A source processor is a special type of stream processor that does not have any upstream processors. It produces an input stream to its topology from one or multiple Kafka topics by consuming records from these topics and forward them to its down-stream processors.</li>
<li><b>Sink Processor</b>: A sink processor is a special type of stream processor that does not have down-stream processors. It sends any received records from its up-stream processors to a specified Kafka topic.</li>
Kafka Streams offers two ways to define the stream processing topology: the <ahref="#streams_dsl"><b>Kafka Streams DSL</b></a> provides
the most common data transformation operations such as <code>map</code>, <code>filter</code>, <code>join</code> and <code>aggregations</code> out of the box; the lower-level <ahref="#streams_processor"><b>Processor API</b></a> allows
developers define and connect custom processors as well as to interact with <ahref="#streams_state">state stores</a>.
A processor topology is merely a logical abstraction for your stream processing code.
At runtime, the logical topology is instantiated and replicated inside the application for parallel processing (see <ahref="#streams_architecture_tasks">Stream Partitions and Tasks</a> for details).
<li><b>Event time</b> - The point in time when an event or data record occurred, i.e. was originally created "at the source". <b>Example:</b> If the event is a geo-location change reported by a GPS sensor in a car, then the associated event-time would be the time when the GPS sensor captured the location change.</li>
<li><b>Processing time</b> - The point in time when the event or data record happens to be processed by the stream processing application, i.e. when the record is being consumed. The processing time may be milliseconds, hours, or days etc. later than the original event time. <b>Example:</b> Imagine an analytics application that reads and processes the geo-location data reported from car sensors to present it to a fleet management dashboard. Here, processing-time in the analytics application might be milliseconds or seconds (e.g. for real-time pipelines based on Apache Kafka and Kafka Streams) or hours (e.g. for batch pipelines based on Apache Hadoop or Apache Spark) after event-time.</li>
<li><b>Ingestion time</b> - The point in time when an event or data record is stored in a topic partition by a Kafka broker. The difference to event time is that this ingestion timestamp is generated when the record is appended to the target topic by the Kafka broker, not when the record is created "at the source". The difference to processing time is that processing time is when the stream processing application processes the record. <b>For example,</b> if a record is never processed, there is no notion of processing time for it, but it still has an ingestion time.</li>
The choice between event-time and ingestion-time is actually done through the configuration of Kafka (not Kafka Streams): From Kafka 0.10.x onwards, timestamps are automatically embedded into Kafka messages. Depending on Kafka's configuration these timestamps represent event-time or ingestion-time. The respective Kafka configuration setting can be specified on the broker level or per topic. The default timestamp extractor in Kafka Streams will retrieve these embedded timestamps as-is. Hence, the effective time semantics of your application depend on the effective Kafka configuration for these embedded timestamps.
Kafka Streams assigns a <b>timestamp</b> to every data record
via the <code>TimestampExtractor</code> interface.
Concrete implementations of this interface may retrieve or compute timestamps based on the actual contents of data records such as an embedded timestamp field
to provide event-time semantics, or use any other approach such as returning the current wall-clock time at the time of processing,
thereby yielding processing-time semantics to stream processing applications.
Developers can thus enforce different notions of time depending on their business needs. For example,
per-record timestamps describe the progress of a stream with regards to time (although records may be out-of-order within the stream) and
are leveraged by time-dependent operations such as joins.
Finally, whenever a Kafka Streams application writes records to Kafka, then it will also assign timestamps to these new records. The way the timestamps are assigned depends on the context:
<li> When new output records are generated via processing some input record, for example, <code>context.forward()</code> triggered in the <code>process()</code> function call, output record timestamps are inherited from input record timestamps directly.</li>
<li> When new output records are generated via periodic functions such as <code>punctuate()</code>, the output record timestamp is defined as the current internal time (obtained through <code>context.timestamp()</code>) of the stream task.</li>
<li> For aggregations, the timestamp of a resulting aggregate update record will be that of the latest arrived input record that triggered the update.</li>
Some stream processing applications don't require state, which means the processing of a message is independent from
the processing of all other messages.
However, being able to maintain state opens up many possibilities for sophisticated stream processing applications: you
can join input streams, or group and aggregate data records. Many such stateful operators are provided by the <ahref="#streams_dsl"><b>Kafka Streams DSL</b></a>.
Kafka Streams allows direct read-only queries of the state stores by methods, threads, processes or applications external to the stream processing application that created the state stores. This is provided through a feature called <b>Interactive Queries</b>. All stores are named and Interactive Queries exposes only the read operations of the underlying implementation.
Kafka Streams simplifies application development by building on the Kafka producer and consumer libraries and leveraging the native capabilities of
Kafka to offer data parallelism, distributed coordination, fault tolerance, and operational simplicity. In this section, we describe how Kafka Streams works underneath the covers.
<p>
The picture below shows the anatomy of an application that uses the Kafka Streams library. Let's walk through some details.
<h3><aid="streams_architecture_tasks"href="#streams_architecture_tasks">Stream Partitions and Tasks</a></h3>
<p>
The messaging layer of Kafka partitions data for storing and transporting it. Kafka Streams partitions data for processing it.
In both cases, this partitioning is what enables data locality, elasticity, scalability, high performance, and fault tolerance.
Kafka Streams uses the concepts of <b>partitions</b> and <b>tasks</b> as logical units of its parallelism model based on Kafka topic partitions.
There are close links between Kafka Streams and Kafka in the context of parallelism:
</p>
<ul>
<li>Each <b>stream partition</b> is a totally ordered sequence of data records and maps to a Kafka <b>topic partition</b>.</li>
<li>A <b>data record</b> in the stream maps to a Kafka <b>message</b> from that topic.</li>
<li>The <b>keys</b> of data records determine the partitioning of data in both Kafka and Kafka Streams, i.e., how data is routed to specific partitions within topics.</li>
</ul>
<p>
An application's processor topology is scaled by breaking it into multiple tasks.
More specifically, Kafka Streams creates a fixed number of tasks based on the input stream partitions for the application,
with each task assigned a list of partitions from the input streams (i.e., Kafka topics). The assignment of partitions to tasks
never changes so that each task is a fixed unit of parallelism of the application. Tasks can then instantiate their own processor topology
based on the assigned partitions; they also maintain a buffer for each of its assigned partitions and process messages one-at-a-time from
these record buffers. As a result stream tasks can be processed independently and in parallel without manual intervention.
</p>
<p>
It is important to understand that Kafka Streams is not a resource manager, but a library that “runs” anywhere its stream processing application runs.
Multiple instances of the application are executed either on the same machine, or spread across multiple machines and tasks can be distributed automatically
by the library to those running application instances. The assignment of partitions to tasks never changes; if an application instance fails, all its assigned
tasks will be automatically restarted on other instances and continue to consume from the same stream partitions.
</p>
<p>
The following diagram shows two tasks each assigned with one partition of the input streams.
Kafka Streams allows the user to configure the number of <b>threads</b> that the library can use to parallelize processing within an application instance.
Each thread can execute one or more tasks with their processor topologies independently. For example, the following diagram shows one stream thread running two stream tasks.
Starting more stream threads or more instances of the application merely amounts to replicating the topology and having it process a different subset of Kafka partitions, effectively parallelizing processing.
It is worth noting that there is no shared state amongst the threads, so no inter-thread coordination is necessary. This makes it very simple to run topologies in parallel across the application instances and threads.
The assignment of Kafka topic partitions amongst the various stream threads is transparently handled by Kafka Streams leveraging <ahref="https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Client-side+Assignment+Proposal">Kafka's coordination</a> functionality.
</p>
<p>
As we described above, scaling your stream processing application with Kafka Streams is easy: you merely need to start additional instances of your application,
and Kafka Streams takes care of distributing partitions amongst tasks that run in the application instances. You can start as many threads of the application
as there are input Kafka topic partitions so that, across all running instances of an application, every thread (or rather, the tasks it runs) has at least one input partition to process.
</p>
<br>
<h3><aid="streams_architecture_state"href="#streams_architecture_state">Local State Stores</a></h3>
<p>
Kafka Streams provides so-called <b>state stores</b>, which can be used by stream processing applications to store and query data,
which is an important capability when implementing stateful operations. The <ahref="#streams_dsl">Kafka Streams DSL</a>, for example, automatically creates
and manages such state stores when you are calling stateful operators such as <code>join()</code> or <code>aggregate()</code>, or when you are windowing a stream.
</p>
<p>
Every stream task in a Kafka Streams application may embed one or more local state stores that can be accessed via APIs to store and query data required for processing.
Kafka Streams offers fault-tolerance and automatic recovery for such local state stores.
</p>
<p>
The following diagram shows two stream tasks with their dedicated local state stores.
Kafka Streams builds on fault-tolerance capabilities integrated natively within Kafka. Kafka partitions are highly available and replicated; so when stream data is persisted to Kafka it is available
even if the application fails and needs to re-process it. Tasks in Kafka Streams leverage the fault-tolerance capability
offered by the <ahref="https://www.confluent.io/blog/tutorial-getting-started-with-the-new-apache-kafka-0.9-consumer-client/">Kafka consumer client</a> to handle failures.
If a task runs on a machine that fails, Kafka Streams automatically restarts the task in one of the remaining running instances of the application.
</p>
<p>
In addition, Kafka Streams makes sure that the local state stores are robust to failures, too. For each state store, it maintains a replicated changelog Kafka topic in which it tracks any state updates.
These changelog topics are partitioned as well so that each local state store instance, and hence the task accessing the store, has its own dedicated changelog topic partition.
<ahref="/{{version}}/documentation/#compaction">Log compaction</a> is enabled on the changelog topics so that old data can be purged safely to prevent the topics from growing indefinitely.
If tasks run on a machine that fails and are restarted on another machine, Kafka Streams guarantees to restore their associated state stores to the content before the failure by
replaying the corresponding changelog topics prior to resuming the processing on the newly started tasks. As a result, failure handling is completely transparent to the end user.
</p>
<p>
Note that the cost of task (re)initialization typically depends primarily on the time for restoring the state by replaying the state stores' associated changelog topics.
To minimize this restoration time, users can configure their applications to have <b>standby replicas</b> of local states (i.e. fully replicated copies of the state).
When a task migration happens, Kafka Streams then attempts to assign a task to an application instance where such a standby replica already exists in order to minimize
the task (re)initialization cost. See <code>num.standby.replicas</code> at the <ahref="/{{version}}/documentation/#streamsconfigs">Kafka Streams Configs</a> Section.
There is a <ahref="/{{version}}/documentation/#quickstart_kafkastreams">quickstart</a> example that provides how to run a stream processing program coded in the Kafka Streams library.
As mentioned in the <ahref="#streams_concepts">Core Concepts</a> section, a stream processor is a node in the processor topology that represents a single processing step.
With the <code>Processor</code> API developers can define arbitrary stream processors that process one received record at a time, and connect these processors with
their associated state stores to compose the processor topology that represents their customized processing logic.
<li>In the <code>init</code> method, schedule the punctuation every 1 second and retrieve the local state store by its name "Counts".</li>
<li>In the <code>process</code> method, upon each received record, split the value string into words, and update their counts into the state store (we will talk about this feature later in the section).</li>
<li>In the <code>punctuate</code> method, iterate the local state store and send the aggregated counts to the downstream processor, and commit the current stream state.</li>
<li>First of all a source node named "SOURCE" is added to the topology using the <code>addSource</code> method, with one Kafka topic "src-topic" fed to it.</li>
<li>Three processor nodes are then added using the <code>addProcessor</code> method; here the first processor is a child of the "SOURCE" node, but is the parent of the other two processors.</li>
<li>Finally three sink nodes are added to complete the topology using the <code>addSink</code> method, each piping from a different parent processor node and writing to a separate topic.</li>
Note that the <code>Processor</code> API is not limited to only accessing the current records as they arrive in the <code>process()</code> method, but can also maintain processing states
that keep recently arrived records to use in stateful processing operations such as windowed joins or aggregation.
To take advantage of these states, users can define a state store by implementing the <code>StateStore</code> interface (the Kafka Streams library also has a few extended interfaces such as <code>KeyValueStore</code>);
in practice, though, users usually do not need to customize such a state store from scratch but can simply use the <code>Stores</code> factory to define a state store by specifying whether it should be persistent, log-backed, etc.
In the following example, a persistent key-value store named “Counts” with key type <code>String</code> and value type <code>Long</code> is created.
To build a processor topology using the Streams DSL, developers can apply the <code>KStreamBuilder</code> class, which is extended from the <code>TopologyBuilder</code>.
A simple example is included with the source code for Kafka in the <code>streams/examples</code> package. The rest of this section will walk
through some code to demonstrate the key steps in creating a topology using the Streams DSL, but we recommend developers to read the full example source
Before we discuss concepts such as aggregations in Kafka Streams we must first introduce tables, and most importantly the relationship between tables and streams:
the so-called <ahref="https://engineering.linkedin.com/distributed-systems/log-what-every-software-engineer-should-know-about-real-time-datas-unifying/">stream-table duality</a>.
Essentially, this duality means that a stream can be viewed as a table, and vice versa. Kafka's log compaction feature, for example, exploits this duality.
The <b>stream-table duality</b> describes the close relationship between streams and tables.
<ul>
<li><b>Stream as Table</b>: A stream can be considered a changelog of a table, where each data record in the stream captures a state change of the table. A stream is thus a table in disguise, and it can be easily turned into a “real” table by replaying the changelog from beginning to end to reconstruct the table. Similarly, in a more general analogy, aggregating data records in a stream – such as computing the total number of pageviews by user from a stream of pageview events – will return a table (here with the key and the value being the user and its corresponding pageview count, respectively).</li>
<li><b>Table as Stream</b>: A table can be considered a snapshot, at a point in time, of the latest value for each key in a stream (a stream's data records are key-value pairs). A table is thus a stream in disguise, and it can be easily turned into a “real” stream by iterating over each key-value entry in the table.</li>
Let's illustrate this with an example. Imagine a table that tracks the total number of pageviews by user (first column of diagram below). Over time, whenever a new pageview event is processed, the state of the table is updated accordingly. Here, the state changes between different points in time – and different revisions of the table – can be represented as a changelog stream (second column).
The same mechanism is used, for example, to replicate databases via change data capture (CDC) and, within Kafka Streams, to replicate its so-called state stores across machines for fault-tolerance.
The stream-table duality is such an important concept that Kafka Streams models it explicitly via the <ahref="#streams_kstream_ktable">KStream, KTable, and GlobalKTable</a> interfaces, which we describe in the next sections.
<h5><aid="streams_kstream_ktable"href="#streams_kstream_ktable">KStream, KTable, and GlobalKTable</a></h5>
The DSL uses three main abstractions. A <b>KStream</b> is an abstraction of a record stream, where each data record represents a self-contained datum in the unbounded data set.
A <b>KTable</b> is an abstraction of a changelog stream, where each data record represents an update. More precisely, the value in a data record is considered to be an update of the last value for the same record key,
If these records a KStream and the stream processing application were to sum the values it would return <code>4</code>. If these records were a KTable or GlobalKTable, the return would be <code>3</code>, since the last record would be considered as an update.
Either a <b>record stream</b> (defined as <code>KStream</code>) or a <b>changelog stream</b> (defined as <code>KTable</code> or <code>GlobalKTable</code>)
can be created as a source stream from one or more Kafka topics (for <code>KTable</code> and <code>GlobalKTable</code> you can only create the source stream
A stream processor may need to divide data records into time buckets, i.e. to <b>window</b> the stream by time. This is usually needed for join and aggregation operations, etc. Kafka Streams currently defines the following types of windows:
<ul>
<li><b>Hopping time windows</b> are windows based on time intervals. They model fixed-sized, (possibly) overlapping windows. A hopping window is defined by two properties: the window's size and its advance interval (aka "hop"). The advance interval specifies by how much a window moves forward relative to the previous one. For example, you can configure a hopping window with a size 5 minutes and an advance interval of 1 minute. Since hopping windows can overlap a data record may belong to more than one such windows.</li>
<li><b>Tumbling time windows</b> are a special case of hopping time windows and, like the latter, are windows based on time intervals. They model fixed-size, non-overlapping, gap-less windows. A tumbling window is defined by a single property: the window's size. A tumbling window is a hopping window whose window size is equal to its advance interval. Since tumbling windows never overlap, a data record will belong to one and only one window.</li>
<li><b>Sliding windows</b> model a fixed-size window that slides continuously over the time axis; here, two data records are said to be included in the same window if the difference of their timestamps is within the window size. Thus, sliding windows are not aligned to the epoch, but on the data record timestamps. In Kafka Streams, sliding windows are used only for join operations, and can be specified through the <code>JoinWindows</code> class.</li>
<li><b>Session windows</b> are used to aggregate key-based events into sessions.
Sessions represent a period of activity separated by a defined gap of inactivity.
Any events processed that fall within the inactivity gap of any existing sessions are merged into the existing sessions.
If the event falls outside of the session gap, then a new session will be created.
Session windows are tracked independently across keys (e.g. windows of different keys typically have different start and end times) and their sizes vary (even windows for the same key typically have different sizes);
as such session windows can't be pre-computed and are instead derived from analyzing the timestamps of the data records.
In the Kafka Streams DSL users can specify a <b>retention period</b> for the window. This allows Kafka Streams to retain old window buckets for a period of time in order to wait for the late arrival of records whose timestamps fall within the window interval.
If a record arrives after the retention period has passed, the record cannot be processed and is dropped.
Late-arriving records are always possible in real-time data streams. However, it depends on the effective <ahref="#streams_time">time semantics</a> how late records are handled. Using processing-time, the semantics are “when the data is being processed”,
which means that the notion of late records is not applicable as, by definition, no record can be late. Hence, late-arriving records only really can be considered as such (i.e. as arriving “late”) for event-time or ingestion-time semantics. In both cases,
Kafka Streams is able to properly handle late-arriving records.
A <b>join</b> operation merges two streams based on the keys of their data records, and yields a new stream. A join over record streams usually needs to be performed on a windowing basis because otherwise the number of records that must be maintained for performing the join may grow indefinitely. In Kafka Streams, you may perform the following join operations:
<ul>
<li><b>KStream-to-KStream Joins</b> are always windowed joins, since otherwise the memory and state required to compute the join would grow infinitely in size. Here, a newly received record from one of the streams is joined with the other stream's records within the specified window interval to produce one result for each matching pair based on user-provided <code>ValueJoiner</code>. A new <code>KStream</code> instance representing the result stream of the join is returned from this operator.</li>
<li><b>KTable-to-KTable Joins</b> are join operations designed to be consistent with the ones in relational databases. Here, both changelog streams are materialized into local state stores first. When a new record is received from one of the streams, it is joined with the other stream's materialized state stores to produce one result for each matching pair based on user-provided ValueJoiner. A new <code>KTable</code> instance representing the result stream of the join, which is also a changelog stream of the represented table, is returned from this operator.</li>
<li><b>KStream-to-KTable Joins</b> allow you to perform table lookups against a changelog stream (<code>KTable</code>) upon receiving a new record from another record stream (<code>KStream</code>). An example use case would be to enrich a stream of user activities (<code>KStream</code>) with the latest user profile information (<code>KTable</code>). Only records received from the record stream will trigger the join and produce results via <code>ValueJoiner</code>, not vice versa (i.e., records received from the changelog stream will be used only to update the materialized state store). A new <code>KStream</code> instance representing the result stream of the join is returned from this operator.</li>
<li><b>KStream-to-GlobalKTable Joins</b> allow you to perform table lookups against a fully replicated changelog stream (<code>GlobalKTable</code>) upon receiving a new record from another record stream (<code>KStream</code>).
Joins with a <code>GlobalKTable</code> don't require repartitioning of the input <code>KStream</code> as all partitions of the <code>GlobalKTable</code> are available on every KafkaStreams instance.
The <code>KeyValueMapper</code> provided with the join operation is applied to each KStream record to extract the join-key that is used to do the lookup to the GlobalKTable so non-record-key joins are possible.
An example use case would be to enrich a stream of user activities (<code>KStream</code>) with the latest user profile information (<code>GlobalKTable</code>).
Only records received from the record stream will trigger the join and produce results via <code>ValueJoiner</code>, not vice versa (i.e., records received from the changelog stream will be used only to update the materialized state store).
A new <code>KStream</code> instance representing the result stream of the join is returned from this operator.</li>
Depending on the operands the following join operations are supported: <b>inner joins</b>, <b>outer joins</b> and <b>left joins</b>.
Their <ahref="https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Streams+Join+Semantics">semantics</a> are similar to the corresponding operators in relational databases.
<h5><aid="streams_dsl_aggregations"href="#streams_dsl_aggregations">Aggregate a stream</a></h5>
An <b>aggregation</b> operation takes one input stream, and yields a new stream by combining multiple input records into a single output record. Examples of aggregations are computing counts or sum. An aggregation over record streams usually needs to be performed on a windowing basis because otherwise the number of records that must be maintained for performing the aggregation may grow indefinitely.
<p>
In the Kafka Streams DSL, an input stream of an aggregation can be a <code>KStream</code> or a <code>KTable</code>, but the output stream will always be a <code>KTable</code>.
This allows Kafka Streams to update an aggregate value upon the late arrival of further records after the value was produced and emitted.
When such late arrival happens, the aggregating <code>KStream</code> or <code>KTable</code> simply emits a new aggregate value. Because the output is a <code>KTable</code>, the new value is considered to overwrite the old value with the same key in subsequent processing steps.
Besides join and aggregation operations, there is a list of other transformation operations provided for <code>KStream</code> and <code>KTable</code> respectively.
Specifying the configuration in Kafka Streams is similar to the Kafka Producer and Consumer clients. Typically, you create a <code>java.util.Properties</code> instance,
set the necessary parameters, and construct a <code>StreamsConfig</code> instance from the <code>Properties</code> instance.
Apart from Kafka Streams' own configuration parameters you can also specify parameters for the Kafka consumers and producers that are used internally,
depending on the needs of your application. Similar to the Streams settings you define any such consumer and/or producer settings via <code>StreamsConfig</code>.
Note that some consumer and producer configuration parameters do use the same parameter name. For example, <code>send.buffer.bytes</code> or <code>receive.buffer.bytes</code> which
are used to configure TCP buffers; <code>request.timeout.ms</code> and <code>retry.backoff.ms</code> which control retries for client request (and some more).
If you want to set different values for consumer and producer for such a parameter, you can prefix the parameter name with <code>consumer.</code> or <code>producer.</code>:
Introduced in 0.11.0 is a new broker config that is particularly relevant to Kafka Streams applications, <code>group.initial.rebalance.delay.ms</code>.
This config specifies the time, in milliseconds, that the <code>GroupCoordinator</code> will delay the initial consumer rebalance.
The rebalance will be further delayed by the value of <code>group.initial.rebalance.delay.ms</code> as each new member joins the consumer group, up to a maximum of the value set by <code>max.poll.interval.ms</code>.
The net benefit is that this should reduce the overall startup time for Kafka Streams applications with more than one thread.
The default value for <code>group.initial.rebalance.delay.ms</code> is 3 seconds.
</p>
<p>
In practice this means that if you are starting up your Kafka Streams app from a cold start, then when the first member joins the group there will be at least a 3 second delay before it is assigned any tasks.
If any other members join the group within the initial 3 seconds, then there will be a further 3 second delay.
Once no new members have joined the group within the 3 second delay, or <code>max.poll.interval.ms</code> is reached, then the group rebalance can complete and all current members will be assigned tasks.
The benefit of this approach, particularly for Kafka Streams applications, is that we can now delay the assignment and re-assignment of potentially expensive tasks as new members join.
So we can avoid the situation where one instance is assigned all tasks, begins restoring/processing, only to shortly after be rebalanced, and then have to start again with half of the tasks and so on.
</p>
<h4><aid="streams_execute"href="#streams_execute">Executing Your Kafka Streams Application</a></h4>
At this point, internal structures have been initialized, but the processing is not started yet. You have to explicitly start the Kafka Streams thread by calling the <code>start()</code> method:
To catch any unexpected exceptions, you may set an <code>java.lang.Thread.UncaughtExceptionHandler</code> before you start the application. This handler is called whenever a stream thread is terminated by an unexpected exception:
Now it's time to execute your application that uses the Kafka Streams library, which can be run just like any other Java application – there is no special magic or requirement on the side of Kafka Streams.
For example, you can package your Java application as a fat jar file and then start the application via:
When the application instance starts running, the defined processor topology will be initialized as one or more stream tasks that can be executed in parallel by the stream threads within the instance.
If the processor topology defines any state stores, these state stores will also be (re-)constructed, if possible, during the initialization
period of their associated stream tasks.
It is important to understand that, when starting your application as described above, you are actually launching what Kafka Streams considers to be one instance of your application.
More than one instance of your application may be running at a time, and in fact the common scenario is that there are indeed multiple instances of your application running in parallel (e.g., on another JVM or another machine).
In such cases, Kafka Streams transparently re-assigns tasks from the existing instances to the new instance that you just started.
See <ahref="#streams_architecture_tasks">Stream Partitions and Tasks</a> and <ahref="#streams_architecture_threads">Threading Model</a> for details.
If you want to upgrade from 0.10.2.x to 0.11.0 you don't need to do any code changes as the public API is fully backward compatible.
However, some configuration parameters were deprecated and thus it is recommend to update your code eventually to allow for future upgrades.
See <ahref="#streams_api_changes_0110">below</a> a complete list of 0.11.0 API and semantical changes that allow you to advance your application and/or simplify your code base, including the usage of new features.
If you want to upgrade from 0.10.1.x to 0.10.2, see the <ahref="/{{version}}/documentation/#upgrade_1020_streams">Upgrade Section for 0.10.2</a>.
It highlights incompatible changes you need to consider to upgrade your code and application.
See <ahref="#streams_api_changes_0102">below</a> a complete list of 0.10.2 API and semantical changes that allow you to advance your application and/or simplify your code base, including the usage of new features.
If you want to upgrade from 0.10.0.x to 0.10.1, see the <ahref="/{{version}}/documentation/#upgrade_1010_streams">Upgrade Section for 0.10.1</a>.
It highlights incompatible changes you need to consider to upgrade your code and application.
See <ahref="#streams_api_changes_0101">below</a> a complete list of 0.10.1 API changes that allow you to advance your application and/or simplify your code base, including the usage of new features.
<p> New methods in <code>TopologyBuilder</code>: </p>
<ul>
<li> added overloads for <code>#addSource()</code> that allow to define a <code>TimestampExtractor</code> per source node </li>
<li> added overloads for <code>#addGlobalStore()</code> that allow to define a <code>TimestampExtractor</code> per source node associated with the global store </li>
</ul>
<p> New methods in <code>KStreamBuilder</code>: </p>
<li> The default config values of embedded producer's <code>retries</code> and consumer's <code>max.poll.interval.ms</code> have been changed to improve the resiliency of a Kafka Streams application </li>
<li> set a listener to react on application state change via <code>#setStateListener(StateListener listener)</code></li>
<li> retrieve the current application state via <code>#state()</code></li>
<li> retrieve the global metrics registry via <code>#metrics()</code></li>
<li> apply a timeout when closing an application via <code>#close(long timeout, TimeUnit timeUnit)</code></li>
<li> specify a custom indent when retrieving Kafka Streams information via <code>#toString(String indent)</code></li>
</ul>
<p>
Parameter updates in <code>StreamsConfig</code>:
</p>
<ul>
<li> parameter <code>zookeeper.connect</code> was deprecated; a Kafka Streams application does no longer interact with Zookeeper for topic management but uses the new broker admin protocol
<li> added overloads for <code>#stream()</code> and <code>#table()</code> that allow to define a <code>auto.offset.reset</code> policy per input stream/table </li>
<li> added method <code>#globalKTable()</code> to create a <code>GlobalKTable</code></li>
<li> added overloads for <code>#join()</code> to join with <code>KTable</code></li>
<li> added overloads for <code>#join()</code> and <code>leftJoin()</code> to join with <code>GlobalKTable</code></li>
<li> note, join semantics in 0.10.2 were improved and thus you might see different result compared to 0.10.0.x and 0.10.1.x
(cf. <ahref="https://cwiki.apache.org/confluence/display/KAFKA/Kafka+Streams+Join+Semantics">Kafka Streams Join Semantics</a> in the Apache Kafka wiki)
<li> like all other KTable operations, <code>KTable-KTable</code> joins do not throw an exception on <code>null</code> key records anymore, but drop those records silently </li>
<p> Relaxed type constraints of many DSL interfaces, classes, and methods (cf. <ahref="https://cwiki.apache.org/confluence/display/KAFKA/KIP-100+-+Relax+Type+constraints+in+Kafka+Streams+API">KIP-100</a>). </p>
<h3><aid="streams_api_changes_0101"href="#streams_api_changes_0101">Streams API changes in 0.10.1.0</a></h3>
<p> Stream grouping and aggregation split into two methods: </p>
<ul>
<li> old: KStream #aggregateByKey(), #reduceByKey(), and #countByKey() </li>
<li> new: KStream#groupByKey() plus KGroupedStream #aggregate(), #reduce(), and #count() </li>
<li> Example: stream.countByKey() changes to stream.groupByKey().count() </li>
</ul>
<p> Auto Repartitioning: </p>
<ul>
<li> a call to through() after a key-changing operator and before an aggregation/join is no longer required </li>
<li> Example: stream.selectKey(...).through(...).countByKey() changes to stream.selectKey().groupByKey().count() </li>
</ul>
<p> TopologyBuilder: </p>
<ul>
<li> methods #sourceTopics(String applicationId) and #topicGroups(String applicationId) got simplified to #sourceTopics() and #topicGroups() </li>
</ul>
<p> DSL: new parameter to specify state store names: </p>
<ul>
<li> The new Interactive Queries feature requires to specify a store name for all source KTables and window aggregation result KTables (previous parameter "operator/window name" is now the storeName) </li>
<li> KStreamBuilder#table(String topic) changes to #topic(String topic, String storeName) </li>
<li> KTable#through(String topic) changes to #through(String topic, String storeName) </li>
<li> Example: stream.countByKey(TimeWindows.of("windowName", 1000)) changes to stream.groupByKey().count(TimeWindows.of(1000), "countStoreName") </li>
</ul>
<p> Windowing: </p>
<ul>
<li> Windows are not named anymore: TimeWindows.of("name", 1000) changes to TimeWindows.of(1000) (cf. DSL: new parameter to specify state store names) </li>
<li> JoinWindows has no default size anymore: JoinWindows.of("name").within(1000) changes to JoinWindows.of(1000) </li>
