Nakadi is a distributed event bus that implements a RESTful API abstraction instead of Kafka-like queues.
The goal of this Spring Boot starter is to simplify the reliable integration between event producer and Nakadi. When we send events from a transactional application, a few recurring challenges appear:
- we have to make sure that events from a transaction get sent, when the transaction has been committed,
- we have to make sure that events from a transaction do not get sent, when the transaction has been rolled back,
- we have to make sure that events get sent, even if an error occurred while sending the event,
- we want to give the event consumer a way to infer the order in which the events occurred and
- it is very comfortable for initial data loads and error recovery to be able to generate snapshots of the current db state as events.
There are already multiple clients for the Nakadi REST API, but none of them solves the mentioned issues.
We solved them by persisting new events in a log table as part of the producing JDBC transaction. They will then be sent asynchronously to Nakadi after the transaction completed. If the transaction is rolled back, the events will vanish too. As a result, events will always be sent if and only if the transaction succeeded.
The Transmitter generates a strictly monotonically increasing event id that can be used for ordering the events during retrieval. It is not guaranteed, that events will be sent to Nakadi in the order they have been produced. If an event could not be sent to Nakadi, the library will periodically retry the transmission.
This project is mature, used in production in some services at Zalando, and in active development.
Be aware that this library does neither guarantee that events are sent exactly once, nor that they are sent in the order they have been persisted. This is not a bug but a design decision that allows us to skip and retry sending events later in case of temporary failures. So make sure that your events are designed to be processed out of order (See Rule 203 in Zalando's API guidelines). To help you in this matter, the library generates a strictly monotonically increasing event id (field metadata/eid in Nakadi's event object) that can be used to reconstruct the message order.
Unfortunately this approach is fundamentally incompatible with Nakadi's compacted event types – it can happen that the last event submitted (and thus the one which will stay after compaction) is not the last event which was actually been fired. We still provide means to set the compaction key, see compacted event types below.
This library follows the semantic versioning schema. A major version change means either an incompatible change in the API, or some incompatible behavioral change (e.g. the database usage), minor versions mean new features without breaking compatibility, and patch versions are backwards compatible bug fixes.
Please also be aware that, when udating between major releases of this lib, you must not jump over a major release (1.0 → 3.0). Please always deploy the intermediate major releases at least once – otherwise you might lose events. You will find migration instructions between major release in the release notes.
You may of course always setup a fresh system with the newest version.
This library was tested with Java 11 (starting with 21.0.0), Spring Boot 2.5.6.RELEASE and Flyway 7. It relies on an existing configured PostgreSQL DataSource. If you are still using Spring Boot 1.x, please use versions < 20.0.0, they are still actively maintained (Documentation).
This library also uses:
- flyway-core
- Spring JDBC
- fahrschein Nakadi client library
- jackson >= 2.7.0
- (optional) Zalando's tracer-spring-boot-starter
- (optional) Zalando's tokens library >= 0.14.0
- Please note that tokens-spring-boot-starter 0.10.0 comes with tokens 0.9.9, which is not enough. You can manually add tokens 0.10.0 with that starter, though. To be used in zalando's k8s environment, you must at least use 0.11.0.
If you are using Maven, include the library in your pom.xml:
<dependency>
<groupId>org.zalando</groupId>
<artifactId>nakadi-producer-spring-boot-starter</artifactId>
<version>${nakadi-producer.version}</version>
</dependency>The latest available version is visible in the Maven central badge at the top of this README.
The library uses Flyway migrations to set up its own database schema nakadi_events.
By default, Nakadi-producer-spring-boot starter uses the Fahrschein library to submit its events. It needs some configuration to know how to do this – we support two ways:
- Using existing Fahrschein setup
- Letting this library set things up
If you are already using the Fahrschein library directly (e.g. for event consumption) and have already a configured org.zalando.fahrschein.NakadiClient object, just make sure it is available as a Spring bean. Nakadi-Producer-Spring-Boot-Starter will pick it up and use it directly.
The configuration in the next section is then not needed at all.
You must tell the library, where it can reach your Nakadi instance:
nakadi-producer:
nakadi-base-uri: https://nakadi.example.orgSince the communication between your application and Nakadi is secured using OAuth2, you must also provide an OAuth2 token. The easiest way to do so is to include the Zalando Tokens library into your classpath:
<dependency>
<groupId>org.zalando.stups</groupId>
<artifactId>tokens</artifactId>
<version>${tokens.version}</version>
</dependency>This starter will detect and autoconfigure it.
If your application is running in Zalando's Kubernetes environment, you have to configure the credential rotation:
apiVersion: "zalando.org/v1"
kind: PlatformCredentialsSet
metadata:
name: {{{APPLICATION}}}-credentials
spec:
application: {{{KIO_NAME}}}
token_version: v2
tokens:
nakadi:
privileges: []If your application is running in Zalando's STUPS environment (or you provide tokens via your own oAuth server) it needs to know the address of your oAuth2 server:
nakadi-producer:
access-token-uri: https://token.auth.example.org/oauth2/access_tokenBy default, the initialized Fahrschein Nakadi client uses the SimpleRequestFactory with ContentEncoding.GZIP.
If you want to override this setting (e.g. to enable zstd compression), you can create an overwriting configuration property:
nakadi-producer:
content-encoding: ZSTDIf you want to disable compression completely:
nakadi-producer:
content-encoding: IDENTITYAs we gain experience, the default value might change in future versions of this library, so if you need to stay on gzip compression, use content-encoding: GZIP.
Alternatively, you can define your own bean of type RequestFactory, which will then be used instead of ours.
Please consult the manual of Zalando's tokens library for more configuration options (like CREDENTIALS_DIR or via environment variables.
Since July 2017, Nakadi (at least in the version operated at Zalando) doesn't require any scopes other than the pseudo-scope uid for writing events, the authorization is instead based on event-type configuration using the service's uid.
If your Nakadi installation needs real scopes for submitting events, you can provide them via configuration, too (as a comma-separated list):
nakadi-producer:
access-token-scopes: my.scope.name,other.scope.nameIf you do not use the STUPS Tokens library, you can implement token retrieval yourself by defining a Spring bean of
type AccessTokenProvider.
The starter will detect it and call it once for each request to retrieve the token.
You can disable the whole Nakadi integration completely with this property:
nakadi-producer:
submission-enabled: falseIn this case you don't need to configure anything related to the Nakadi communication ↑ (and this library won't set up any beans related to it).
A use case for this might be that you have several components of your application connected to the same database,
and want the submission of the events centralized in one of these components. Then for all other components you'd
set nakadi-producer.submission-enabled: false (true is the default), but still can use the EventLogWriter to
create events.
The typical use case for this library is to publish events like creating or updating of some objects.
In order to store events you can autowire the EventLogWriter
service and use its methods: fireCreateEvent, fireUpdateEvent, fireDeleteEvent, fireSnapshotEvent or fireBusinessEvent.
To store several events of the same type in bulk, the methods fireCreateEvents, fireUpdateEvents, fireDeleteEvents, fireSnapshotEvents or fireBusinessEvents can be used.
You normally don't need to call fireSnapshotEvent directly, see below for snapshot creation.
Example of using fireCreateEvent:
@Service
public class SomeYourService {
@Autowired
private EventLogWriter eventLogWriter;
@Autowired
private WarehouseRepository repository;
@Transactional
public void createObject(Warehouse data) {
// here we store an object in a database table
repository.save(data);
// and then in the same transaction we save the event about this object creation
eventLogWriter.fireCreateEvent("wholesale.warehouse-change-event", "wholesale:warehouse", data);
}
}Note: The parameters to the fire*Event methods (except for business events) are the following:
-
eventType - event type name string that determines to which channel/topic the event will get sent. That event type name needs to exist at Nakadi.
-
dataType - data type name string that will end up as the
data_typeproperty of the data change event. (It is not really clear what this property is used for, but it is required.) -
data - event data payload itself, which will end up in the
dataproperty of the data change event. This should be an object representing the resource which was created/updated/deleted. This doesn't necessarily have to be the same object as you store in your DB, it can be a different class which is optimized for JSON serialization.It will be serialized using the default Jackson ObjectMapper bean – so make sure all properties which should be in the event are accessible (usually via public getters), or annotated with the usual Jackson annotations. The JSON serialization should confirm to the JSON schema defined at the event type definition in Nakadi.
The choice of the method (Create/Update/Delete/Snapshot) event will determine the data_op field of the event.
It makes sense to use these methods in one transaction with corresponding object creation or mutation. This way we get rid of distributed-transaction problem as mentioned earlier.
For business events, you have just two parameters, the eventType and the event payload object. You usually should fire those also in the same transaction as you are storing the results of the process step the event is reporting.
Nakadi offers a "log-compaction" feature, where each event (on an event type) has a
partition_compaction_key, and
Nakadi will (after delivering to live subscribers) clean up events, but leave the latest event for each
compaction key available long-term.
This library (by design) doesn't guarantee the submission order of events – especially when there are problems on Nakadi side and some events fail (and are retried later), earlier produced events (for the same entity) can be submitted after later events. For log-compacted event types this means that an outdated event will remain in the topic for future subscribers to read. It is therefore generally not recommended to use this library (or any solution which doesn't guarantee the order) for sending events to a compacted event type.
In some cases, like when there usually are large time gaps between producing events for the same compaction key,
the risk of getting events for the same key out-of-order is small.
For these cases, you just can define a bean of type CompactionKeyExtractor,
and then all events of that event type will be sent with a compaction key.
@Configuration
public class NakadiProducerConfiguration {
@Bean
public CompactionKeyExtractor extractorForWarehouseEvents() {
return CompactionKeyExtractor.of("wholesale.warehouse-change-event",
Warehouse.class, Warehouse::getCode);
}
}The service class sending the event looks exactly the same as above.
For corner cases: You can have multiple such extractors for the same event type, any one where the class object matches the payload object (in undefined order) will be used. There are also some more factory methods with different signatures for more special cases, and you can also write your own implementation (but for the usual cases, the one shown here should be enough).
A Snapshot event is a special type of data change event (data operation) defined by Nakadi. It does not represent a change of the state of a resource, but a current snapshot of its state. It can be useful to bootstrap a new consumer or to recover from inconsistencies between sender and consumer after an incident.
You can create snapshot events programmatically (using EventLogWriter.fireSnapshotEvent), but usually snapshot event creation is an irregular, manually triggered maintenance task.
This library provides a Spring Boot Actuator endpoint named snapshot_event_creation that can be used to trigger a Snapshot for a given event type. Assuming your management port is set to 7979,
GET localhost:7979/actuator/snapshot-event-creation
will return a list of all event types available for snapshot creation and
POST localhost:7979/actuator/snapshot-event-creation/my.event-type
will trigger a snapshot for the event type my.event-type. You can change the port, the authentication scheme and the
path prefix as part of your Spring Boot Actuator configuration.
You can provide an optional filter specifier that will be passed to your application to implement any application
specific event/entity filtering logic. It can be provided either as a query parameter called filter, or as a
request body
{"filter":"myFilter"}
This endpoint will only work if your application includes spring-boot-actuator,
<dependency>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-actuator</artifactId>
</dependency>your application.properties includes
management.endpoints.web.exposure.include=snapshot-event-creation,your-other-endpoints,...`
and if one or more Spring Beans implement the org.zalando.nakadiproducer.snapshots.SnapshotEventGenerator interface.
(Note that this will automatically work together with the compaction key feature mentioned above,
if you have registered a compaction key extractor matching the type of the data objects in your snapshots.)
The optional filter specifier of the trigger request will be passed as a string parameter to the
SnapshotEventGenerator's generateSnapshots method and may be null, if none is given.
The SnapshotEventGenerator has static of methods to ease bean creation using a more functional style:
@Bean
public SnapshotEventGenerator snapshotEventGenerator(MyService service) {
return SnapshotEventGenerator.of("event type", service::createSnapshotEvents);
}This library supports tracer-spring-boot-starter (another library from Zalando) that provides a support of X-Flow-ID header.
In order to use it you should provide the library as a dependency in your project and configure it:
<dependency>
<groupId>org.zalando</groupId>
<artifactId>tracer-spring-boot-starter</artifactId>
<version>${tracer.version}</version>
</dependency>tracer:
traces:
X-Flow-ID: flow-idThen the current flow-ID of the fireXXX-calls will be added to the created events, and thus can be used by the event consumers.
By default, the library will pick up your flyway data source (or the primary data source if no flyway data source is configured), create its own schema and start setting up its tables in there. You can customize this process in two ways:
If you want to use a different data source for schema maintenance (for example to use a different username) and
configuring the Spring Flyway datasource is not enough, your can define a spring bean of type DataSource and annotate
it with @NakadiProducerDataSource.
You may also define a spring bean of type NakadiProducerFlywayCallback. The interface provides several hooks into the
schema management lifecycle that may, for example, be used to SET ROLE migrator before and RESET ROLE after each
migration.
This library provides a mock implementation of its Nakadi client that can be used in integration testing:
public class MyIT {
@Autowired
private EventTransmitter eventTransmitter;
@Autowired
// Just define it in your tests spring config. It will automatically be picked up by the auto configuration.
private MockNakadiPublishingClient nakadiClient;
@Before
@After
public void clearNakadiEvents() {
eventTransmitter.sendEvents();
nakadiClient.clearSentEvents();
}
@Test
public void businessEventsShouldBeSubmittedToNakadi() throws IOException {
myTransactionalService.doSomethingAndFireEvent();
eventTransmitter.sendEvents();
List<String> jsonStrings = nakadiClient.getSentEvents("my_event_type");
assertThat(jsonStrings.size(), is(1));
assertThat(read(jsonStrings.get(0), "$.data_op"), is("C"));
assertThat(read(jsonStrings.get(0), "$.data_type"), is(PUBLISHER_DATA_TYPE));
assertThat(read(jsonStrings.get(0), "$.data.id"), is(123));
assertThat(read(jsonStrings.get(0), "$.data.items.length()"), is(3));
assertThat(read(jsonStrings.get(0), "$.items[0].detail"), is(payload.getItems().get(0).getDetail()));
}
}The example above uses com.jayway.jsonpath:json-path:jar:2.2.0 to parse and test the json results.
Note that you should disable the scheduled event transmission for the test (e.g. by setting nakadi-producer.scheduled-transmission-enabled:false), as that might interfere with the manual transmission and the clearing in the test setup, leading to events from one test showing up in the next test, depending on timing issues.
This property can of course also be used in production, if you need to disable sending out events for some reason.
When transmitting events, a scheduled transmitter locks a bunch of events (so other instances of the app are not trying to send out the same events), then sends them out to Nakadi (in one batch per event-type). There are three properties for customizing this event locking on transmission:
-
lock-duration (seconds): The time the selected events are locked before transmission. If the transmission fails, the events stay locked until the lock expires. The default is currently 600 seconds but might change in future releases.
-
lock-duration-buffer (seconds): Since clocks never work exactly synchronous and sending events also takes some time, a safety buffer is included. During the last x seconds before the expiration of the lock, the events are not considered for transmission. The default is currently 60 seconds but might change in future releases.
-
lock-size (events): Defines the maximum amount of events which are loaded into memory and published in one run (in one submission per event type). By default, all events are loaded into memory. In some future release, this property will become mandatory. This should be set to a value which is not too high so out-of-memory situations are avoided.
Example:
nakadi-producer:
lock-duration: 600
lock-duration-buffer: 60
lock-size: 5000This is a list of all the documented spring properties (in alphabetical order), with the link to the corresponding section.
| Property | Summary |
|---|---|
management.endpoints.web.exposure.include: snapshot-event-creation |
Enable snapshot event production endpoint. |
nakadi-producer.access-token-uri |
The URI of an OAuth2 server where an access token can be acquired. (Only for the legacy STUPS environment.) |
nakadi-producer.access-token-scopes |
The Scopes needed on a token. (Not needed for Zalando's Nakadi setup.) |
nakadi-producer.content-encoding |
Compression setting for Nakadi submission, one of GZIP, ZSTD or IDENTITY. |
nakadi-producer.lock-duration |
The number of seconds events are locked before transmission. |
nakadi-producer.lock-duration-buffer |
Number of seconds before the expiry of a lock an event is not used. |
nakadi-producer.lock-size |
Number of events to lock (and then load into memory) at once. |
nakadi-producer.nakadi-base-uri |
The Nakadi base URI used for submitting events. |
nakadi-producer.scheduled-transmission-enabled: false |
Disable event transmission scheduler (but still set up Nakadi connection beans, so it can be used manually). |
nakadi-producer.submission-enabled: false |
Disable event submission completely (including all beans for this). |
tracer.traces.X-Flow-ID: flow-id |
Enable flow-ID support |
We welcome contributions. Please have a look at our contribution guidelines.
If you have an idea of what the library should do, please have a look into our Issues to see whether it was already proposed before, and otherwise open an issue. We also welcome pull requests (for your issues or even for issues from others).
If you want to support us, we collected a few open issues that should be easy to contribute.
In the interest of fostering an open and welcoming environment, we follow and enforce our Code of Conduct.
Build with unit tests and integration tests:
./mvnw clean installThis will sign the created artifact, which is needed for publication to Maven Central. If the GPG integration causes headaches (and you do not plan to publish the created artifact to maven central anyway), you can skip gpg signing:
./mvnw -Dgpg.skip=true clean installWe (the maintainers) want to thank our main contributors:
- Alexander Libin (@qlibin), who created a similar predecessor library (tarbela-producer-spring-boot-starter, now not public anymore), from which this one was forked.
- Lucas Medeiros de Azevedo (@wormangel), who added support for business events.
- Benjamin Gehrels (@BGehrels), who co-maintained this project from its inception in 2017 until 2019, contributing many features and ideas.
For all questions, bug reports, proposals, etc., please create an issue. We try to react to new issues usually by the next working day.
If you need to contact the maintainers confidentially, please use the email addresses in the MAINTAINERS file. In case of a security issue, please also send a copy to tech-security@zalando.de.
The MIT License (MIT) Copyright © 2016 Zalando SE, https://tech.zalando.com
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.