Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?


You need a fahrschein to use the nakadi event bus.


Image Source: Deutsches Bundesarchiv, Photographer: Ralph Hirschberger, Image License: CC BY-SA 3.0 DE

Build Status Release Maven Central


  • Consistent error handling
  • Stream-based parsing
    • Optimized utf-8 decoding by Jackson
    • No unnecessary buffering or line-based processing, causing less garbage
    • Less garbage and higher performance
    • No required base classes for events
  • Support for both high-level (subscription) and low-level APIs
  • Pluggable HTTP client implementations
  • Gzip encoding support for publishing and consuming events
  • ZStandard compression support for publishing events
  • Integrated Spring-Boot-Starter


Fahrschein is available in maven central, so you only have to add the following dependency to your project:


Subscribing to Event Types


The following code example creates an event listener based on Nakadi's subscription API, which gets a call-back for each batch.

String eventName = "sales-order-placed";

// Create a Listener for our event
Listener<SalesOrderPlaced> listener = events -> {
    for (SalesOrderPlaced salesOrderPlaced : events) {"Received sales order [{}]", salesOrderPlaced.getSalesOrder().getOrderNumber());

// Configure client. It is mandatory to choose the RequestFactory implementation (we recommend the JavaNetRequestFactory) and request encoding (we recommend compression via GZIP or ZSTD) 
// PlatformAccessTokenProvider provides a token by reading from the filesystem's CREDENTIALS_DIR 
NakadiClient nakadiClient = NakadiClient.builder(NAKADI_URI, new JavaNetRequestFactory(ContentEncoding.GZIP))
        .withAccessTokenProvider(new PlatformAccessTokenProvider())

// Create subscription using the Subscription API.
// The authorization section is mandatory. We recommend setting your team as admin
// and your application as reader.
Subscription subscription = nakadiClient.subscription(applicationName, eventName)
            .addAdmin("team", "your_team")
            .addReader("service", "stups_your_application").build())

// Define stream parameters, like batch size. For more details and defaults see Nakadi's API definition 
StreamParameters streamParameters = new StreamParameters()

// Start streaming. The listen call will block and automatically reconnect on IOException
        .listen(SalesOrderPlaced.class, listener);

See for an executable version of the above code.

Subscribing to events with an existing subscription ID

One could also use a pre-existing subscription ID to stream events. This will force fahrschein not to create a subscription for the given event and will use the provided subscription ID.

// Create subscription using an existing subscription ID
Subscription subscription = nakadiClient

// Start streaming, the listen call will block and automatically reconnect on IOException
        .listen(SalesOrderPlaced.class, listener);

Initializing partition offsets

By default nakadi will start streaming from the most recent offset. The initial offsets can be changed by requesting data about partitions from Nakadi and using this data to configure the CursorManager.

List<Partition> partitions = nakadiClient.getPartitions(eventName);

// NakadiClient can be configured to start reading from the oldest available offset in each partition;

// Or from the newest available offset, but this is the same as the default;

// Or (for a persistent cursor manager) we can start reading from the last offset that we processed if it's still available, and from the oldest available offset otherwise;

Publishing Events

// event types
String ORDER_CREATED = "order-created";

// Configure client. It is mandatory to choose the RequestFactory implementation (we recommend the JavaNetRequestFactory) and request encoding (we recommend compression via GZIP or ZSTD) 
// PlatformAccessTokenProvider provides a token by reading from the filesystem's CREDENTIALS_DIR 
NakadiClient nakadiClient = NakadiClient.builder(NAKADI_URI, new JavaNetRequestFactory(ContentEncoding.GZIP))
        .withAccessTokenProvider(new PlatformAccessTokenProvider())

// Produce events in batches
while (true) {
    List<OrderCreatedEvent> events = new ArrayList<>();
    for (int i = 0; i < 10; i++) {
        events.add(new OrderCreatedEvent(metadata, "123", Money.of(123, "EUR"), "1234", "paypal"));
    }"publishing {} events", events.size());
    // Publish
    nakadiClient.publish(ORDER_CREATED, events);

OAuth support

Default usage of PlatformAccessTokenProvider

By default fahrschein supports implementation of Zalando platofrm IAM (OAuth 2.0) through PlatformAccessTokenProvider. The implementation expects a mounted directory with the below structure.

└── credentials
    ├── example-token-secret
    └── example-token-type

The resulting token would be "Bearer your-secret-token"

Custom authorization

One can override AuthorizationProvider interface to support custom authorization flow.


//Create custom authorization
class CustomAuthorization implements AuthorizationProvider{

    public String getAuthorizationHeader() throws IOException {
        return "your-secret-token";

NakadiClient nakadiClient = NakadiClient.builder(NAKADI_URI, new SimpleRequestFactory(ContentEncoding.IDENTITY))
        .withAccessTokenProvider(new CustomAuthorization())

Retries and exception handling for event consumption

Exception handling while streaming events follows some simple rules

  • IOException and its subclasses are treated as temporary failures.If an IOException occurs while opening the initial connection, it is considered a temporary failure.Library will automatically recover from this type of failure by retrying processing attempt after time interval specified by the BackoffStrategy.
  • If listener throws RuntimeException streaming of events will be aborted. User is responsible for handling these exceptions. Library code itself will not throw RuntimeExceptions.
  • Exceptions in other client methods are not automatically retried

Backoff Strategies

Fahrschein supports different exponential backoff strategies when streaming events. All backoff values are configurable when passing your own instance in StreamBuilder#withBackoffStrategy(BackoffStrategy).

  • ExponentialBackoffStrategy - Base implementation for exponential backoff without jitter. Initial delay is 500ms, backoff factor 1.5, maximum delay 10min, with no limit on the maximum number of retries.
  • EqualJitterBackoffStrategy (default) - extends ExponentialBackoffStrategy with the same defaults. For each delay it takes half of the delay value and adds the other half multiplied by a random factor [0..1).
  • FullJitterBackoffStrategy - extends ExponentialBackoffStrategy with the same defaults and multiplies each delay by a random factor [0..1).

Retries and exception handling for event publishing

Fahrschein does not have sophisticated mechanisms for retry handling when publishing to Nakadi yet. The recommended way to handle exceptions when publishing is to create a retry-wrapper around the NakadiClient.publish method.

In case of a partial success or also in cases like validation errors, which are complete failures, Fahrschein will throw an EventPublishingException with the BatchItemResponses (as returned from Nakadi) for the failed items in the responses property.

These objects have the eid of the failed event, a publishingStatus (failed/aborted/submitted - but successful itemes are filtered out), the step where it failed and a detail string.

If the application sets the eids itself (i.e. doesn't let Nakadi do it) and keeps track of them, this allows it to resend only the failed items later.

It also allows differentiating between validation errors, which likely don't need to be retried, as they are unlikely to succeed the next time, unless the event type definition is changed, and publishing errors which should be retried with some back-off.

Recommendation: Implement a retry-with-backoff handler for EventPublishingExceptions, which, depending on your ordering consistency requirements, either retries the full batch, or retries the failed events based on the event-ids.

Stopping and resuming streams

The stream implementation gracefully handles thread interruption, so it is possible to stop a running thread and resume consuming events by re-submitting the Runnable:

ExecutorService executorService = Executors.newSingleThreadExecutor();

Runnable runnable =
        .runnable(SalesOrderPlaced.class, listener)

// start consuming events
Future<?> future = executorService.submit(runnable);

// stop consuming events

// resume consuming events
Future<?> future2 = executorService.submit(runnable);

RequestFactory implementations

Fahrschein uses it's own http abstraction which is very similar to spring framework's ClientHttpRequestFactory interface. By default it uses the SimpleRequestFactory which uses a HttpURLConnection internally and has no further dependencies.

There is a version using HttpClient (JDK11+) named JavaNetRequestFactory in the fahrschein-http-jdk11 artifact.


There is also a version using apache http components named HttpComponentsRequestFactory in the fahrschein-http-apache artifact.


The apache version is useful when you want more control about the number of parallel connections in total or per host. The following example shows how to use a customized HttpClient:

RequestConfig config = RequestConfig.custom().setSocketTimeout(readTimeout)

CloseableHttpClient httpClient = HttpClients.custom()
                                                  .setConnectionTimeToLive(readTimeout, TimeUnit.MILLISECONDS)

RequestFactory requestFactory = new HttpComponentsRequestFactory(httpClient, ContentEncoding.GZIP);

NakadiClient nakadiClient = NakadiClient.builder(NAKADI_URI, requestFactory)
        .withAccessTokenProvider(new PlatformAccessTokenProvider())

It is also possible to adapt other client implementations (e.g. OkHttp3) by using the Spring framework abstraction and by wrapping them into SpringRequestFactory, contained in the fahrschein-http-spring artifact. The current version assumes Spring Framework to be a provided dependency. See table below for the range of compatible Spring versions.


Example using OkHttp 3.x:

ClientHttpRequestFactory clientHttpRequestFactory = new OkHttp3ClientHttpRequestFactory();
RequestFactory requestFactory = new SpringRequestFactory(clientHttpRequestFactory, ContentEncoding.GZIP);

NakadiClient nakadiClient = NakadiClient.builder(NAKADI_URI, requestFactory)
        .withAccessTokenProvider(new PlatformAccessTokenProvider())

Note: The implementations from the Spring framework don't handle closing of streams as expected. They will try to consume remaining data, which will usually time out when nakadi does not receive a commit.

Note: Regarding sizing and reuse of HTTP client connection pools, make sure to have a connection pool size bigger than the number of subscriptions you're making, because subscriptions use long-polling to retrieve events, each effectively blocking one connection.


Fahrschein has support for metrics using Dropwizard and Micrometer, at the moment for consuming events only. Each integration has two metrics collector implementations, one for counting, one for last activity timestamps. Multiple metrics integrations can be chained by using the MultiplexingMetricsCollector. In dropwizard, the default metrics prefix is org.zalando.fahrschein, for Micrometer it is fahrschein.listener.

Example using Micrometer:

MeterRegistry registry = new SimpleMeterRegistry();
 .withMetricsCollector(new MicrometerMetricsCollector(registry))

Using Nakadi's Low-level API

Please do not use the Low-level API, as it is deprecated by Nakadi.

The Low-level API requires local persistence of partition offsets. There is one CursorManager implementation left: InMemory. Postgres and Redis cursor managers have been DEPRECATED and removed in version 0.22.0 of Fahrschein.

CursorManager cursorManager = new InMemoryCursorManager();

NakadiClient nakadiClient = NakadiClient.builder(NAKADI_URI, new SimpleRequestFactory(ContentEncoding.IDENTITY))
        .withAccessTokenProvider(new PlatformAccessTokenProvider())
        .listen(SalesOrderPlaced.class, listener);

Using multiple partitions and multiple consumers

With the PartitionManager api it is possible to coordinate between multiple nodes of one application, so that only one node is consuming events from a partition at the same time.

Partitions are locked by one node for a certain time. This requires that every node has a unique name or other identifier.

@Scheduled(fixedDelay = 60*1000L)
public void readSalesOrderPlacedEvents() throws IOException {
    String lockedBy = ... // host name or another unique identifier for this node
    List<Partition> partitions = nakadiClient.getPartitions(eventName);
    Optional<Lock> optionalLock = partitionManager.lockPartitions(eventName, partitions, lockedBy);

    if (optionalLock.isPresent()) {
        Lock lock = optionalLock.get();
        try {
                    .listen(SalesOrderPlaced.class, listener);
        } finally {

Dependency compatibility

Although Fahrschein is using fixed dependency versions, it is integration-tested against the following dependency matrix. We will inform in the release notes in case we bump the compatibility baseline.

Dependency Baseline Latest
Jackson 2.8.0 2.+
Spring Core 6.0.0 6.+
Spring Boot 3.0.0 3.+
okHttp 3.3.0 4.+
Apache HttpClient  4.4 4.+


Fahrschein handles content compression transparently to the API consumer, and mostly independently of the actual HTTP client implementation. Since version 0.20.0 it can be enabled to both compress HTTP POST bodies when event publishing, and requesting payload compression from Nakadi when consuming events.


For event consumption the underlying HTTP client implementations send Accept-Encoding headers, indicating their supported compression algorithms. At the time of writing, all tested client implementations default to gzip compression. If this is undesired, wrap your RequestFactory into a IdentityAcceptEncodingRequestFactory, which sets the Accept-Encoding header to identity.


For event publishing, the Request body can also get compressed by Fahrschein, if enabled when building the RequestFactory. For this, you need to pass either ContentEncoding.GZIP, ContentEncoding.ZSTD, or if compression is undesired, pass ContentEncoding.IDENTITY. Zstandard compression was added in version 0.21.0.

Fahrschein compared to other Nakadi client libraries

Fahrschein nakadi-java
Dependencies Jackson gson, okhttp3, RxJava
Low-level API streaming yes yes
Subscription API streaming yes yes
Compression: consuming gzip (enabled by default) gzip (enabled by default)
Compression: publishing gzip, zstd gzip
Error Handling Automatic retry with exponential backoff Automatic retry with exponential backoff
OpenTracing yes yes
Metrics Collection yes (dropwizard,micrometer) yes (dropwizard,micrometer)
Access to Event Metadata no yes
Event Type manipulation no yes

Getting help

If you have questions, concerns, bug reports, etc, please file an issue in this repository's issue tracker.

Local development

Fahrschein is a gradle-based project. For local development, Fahrschein requires:

  • A local installation of JDK11
  • A local Docker installation for running integration tests

When developing, make sure to run unit and integration tests with ./gradlew check.

Understanding the build

We use Gradle convention plugins to share common build logic across multiple subprojects. These are for common java properties, and fahrschein.maven-publishing-conventions.gradle for subprojects that are released as maven artefacts.

Bumping dependency versions

Most dependencies are defined on a per-subproject level, only the versions for the most-used shared dependencies are controlled centrally, in the file. This also allows you testing your build with a different version by specifying the property on the command-line.

./gradlew check -Pjackson.version=2.9.0

The integration tests include running the build with our supported baseline dependency as well as the latest micro release of Jackson, Apache HttpClient and Spring. Please update the section in the README when bumping dependency baselines, and add this to the release notes.

Unit Tests and Code Coverage

Fahrschein automatically generates test coverage reports via its JaCoCo build integration. For example, you can check code coverage for a single subproject:

./gradlew fahrschein-http-api:check
open fahrschein-http-api/build/reports/jacoco/test/html/index.html

Alternatively, run the full build and check the aggregated coverage report:

./gradlew check -Pe2e.skip
open build/reports/jacoco/jacocoAggregateReport/html/index.html

End-to-End tests

The fahrschein-e2e-test module has end-to-end tests using docker-compose. If you wish to leave the Docker containers running between tests, you can also bring up docker-compose manually using its docker-compose.yaml, before running the tests. In either case, you will need the port 8080 to be available for Nakadi to run.

docker-compose -f fahrschein-e2e-test/src/test/resources/docker-compose.yaml up -d
./gradlew :fahrschein-e2e:check -Pe2e.composeProvided

If you want to skip end-to-end tests completely, run

./gradlew check -Pe2e.skip

CVE scanning

The project integrates CVE scanning to check for vulnerable dependencies. In case of build failure, this can be caused by a high-risk vulnerability in a dependency being identified. You can run the reporting locally:

./gradlew :dependencyCheckAggregate
open build/reports/dependency-check-report.html


Fahrschein uses Github Workflows to build and publish releases. This happens automatically whenever a new release is created in Github. After creating a release, please bump the project.version property in

If needed, you can preview (unsigned) release artifacts in your local maven repository.

./gradlew publishToMavenLocal -xsignMavenPublication

Getting involved

Check the contribution guidelines if you want to get involved in Fahrschein development.