This project allows to leverage Zeebe (the orchestration engine that comes as part of Camunda Platform 8) within your Spring or Spring Boot environment easily. It is basically a wrapper around the Zeebe Java Client.
Spring Zeebe version | JDK | Camunda Platform version | Bundled Spring Boot version | Compatible Spring Boot versions |
---|---|---|---|---|
>= 8.1.15 | > 11 | 8.1.x | 2.7.7 | >= 2.7.6, 3.0.x |
< 8.1.14 | > 11 | 8.1.x | 2.7.5 | = 2.7.x (no 3.0.x) |
There are full examples, including test cases, are available here: Twitter Review example, Process Solution Template. Further, you might want to have a look into the example/ folder.
Create a new Spring Boot project (e.g. using Spring initializr), or open a pre-existing one you already have, or simply fork our Camunda Platform 8 Process Solution Template.
Add the following Maven dependency to your Spring Boot Starter project:
<dependency>
<groupId>io.camunda</groupId>
<artifactId>spring-zeebe-starter</artifactId>
<version>8.2.1</version>
</dependency>
Although Spring Zeebe has a transitive dependency to the Zeebe Java Client, you could also add a direct dependency if you need to specify the concrete version in your pom.xml
(even this is rarely necessary):
<dependency>
<groupId>io.camunda</groupId>
<artifactId>zeebe-client-java</artifactId>
<version>8.1.9</version>
</dependency>
Connections to the Camunda SaaS can be easily configured, create the following entries in your src/main/resources/application.properties
:
zeebe.client.cloud.cluster-id=xxx
zeebe.client.cloud.client-id=xxx
zeebe.client.cloud.client-secret=xxx
zeebe.client.cloud.region=bru-2
You can also configure the connection to a self-managed Zeebe broker:
zeebe.client.broker.gateway-address=127.0.0.1:26500
zeebe.client.security.plaintext=true
You can enforce the right connection mode, for example if multiple contradicting properties are set:
zeebe.client.connection-mode=CLOUD
zeebe.client.connection-mode=ADDRESS
You can inject the ZeebeClient and work with it, e.g. to create new workflow instances:
@Autowired
private ZeebeClient client;
Use the @Deployment
annotation:
@SpringBootApplication
@EnableZeebeClient
@Deployment(resources = "classpath:demoProcess.bpmn")
public class MySpringBootApplication {
This annotation uses (which internally uses [https://docs.spring.io/spring-framework/docs/current/reference/html/core.html#resources-resourceloader] (the Spring resource loader) mechanism which is pretty powerful and can for example also deploy multiple files at once:
@Deployment(resources = {"classpath:demoProcess.bpmn" , "classpath:demoProcess2.bpmn"})
or define wildcard patterns:
@Deployment(resources = "classpath*:/bpmn/**/*.bpmn")
@JobWorker(type = "foo")
public void handleJobFoo(final ActivatedJob job) {
// do whatever you need to do
}
See documentation below for a more in-depth discussion on parameters and configuration options of JobWorkers.
You can startup an in-memory test engine and do assertions by adding this Maven dependency:
<dependency>
<groupId>io.camunda</groupId>
<artifactId>spring-zeebe-test</artifactId>
<version>${spring-zeebe.version}</version>
<scope>test</scope>
</dependency>
Note that the test engines requires Java version >= 17. If you cannot run on this Java version, you can use Testcontainers instead. Testcontainers require that you have a docker installation locally available on the developer machine. Use this dependency:
<!--
Alternative dependency if you cannot run Java 17, so you will leverage Testcontainer
Make sure NOT to have spring-zeebe-test on the classpath in parallel!
-->
<dependency>
<groupId>io.camunda</groupId>
<artifactId>spring-zeebe-test-testcontainer</artifactId>
<version>${spring-zeebe.version}</version>
<scope>test</scope>
</dependency>
Using Maven profiles you can also switch the test dependencies based on the available Java version.
Then you need to startup the test engine in your test case by adding @ZeebeSpringTest
@SpringBootTest
@ZeebeSpringTest
public class TestMyProcess {
// ...
An example test case is available here.
Spring Zeebe project previously included the Runtime for Camunda 8 Connectors. It has been moved to a separate Connectors project. To run Connectors, use can now use the following dependency in your project:
<dependency>
<groupId>io.camunda.connector</groupId>
<artifactId>spring-boot-starter-connectors/artifactId>
<version>${connectors.version}</version>
</dependency>
If you have previously used the pure Spring Zeebe project to run Connectors, you should migrate to the new dependency.
You can find the latest version of Connectors on this page. Consult the Connector SDK for details on Connectors in general.
You can configure the job type via the JobWorker
annotation:
@JobWorker(type = "foo")
public void handleJobFoo() {
// handles jobs of type 'foo'
}
If you don't specify the type
the method name is used as default:
@JobWorker
public void foo() {
// handles jobs of type 'foo'
}
As a third possibility, you can set a default job type:
zeebe.client.worker.default-type=foo
This is used for all workers that do not set a task type via the annoation.
You can specify that you only want to fetch some variables (instead of all) when executing a job, which can decrease load and improve performance:
@JobWorker(type = "foo", fetchVariables={"variable1", "variable2"})
public void handleJobFoo(final JobClient client, final ActivatedJob job) {
String variable1 = (String)job.getVariablesAsMap().get("variable1");
System.out.println(variable1);
// ...
}
By using the @Variable
annotation there is a shortcut to make variable retrieval simpler, including the type cast:
@JobWorker(type = "foo")
public void handleJobFoo(final JobClient client, final ActivatedJob job, @Variable String variable1) {
System.out.println(variable1);
// ...
}
With @Variable
or fetchVariables
you limit which variables are loaded from the workflow engine. You can also override this and force that all variables are loaded anyway:
@JobWorker(type = "foo", fetchAllVariables = true)
public void handleJobFoo(final JobClient client, final ActivatedJob job, @Variable String variable1) {
}
You can also use your own class into which the process variables are mapped to (comparable to getVariablesAsType()
in the Java Client API). Therefore use the @VariablesAsType
annotation. In the below example, MyProcessVariables
refers to your own class:
@JobWorker(type = "foo")
public ProcessVariables handleFoo(@VariablesAsType MyProcessVariables variables){
// do whatever you need to do
variables.getMyAttributeX();
variables.setMyAttributeY(42);
// return variables object if something has changed, so the changes are submitted to Zeebe
return variables;
}
You can access variables of a process via the ActivatedJob object, which is passed into the method if it is a parameter:
@JobWorker(type = "foo")
public void handleJobFoo(final ActivatedJob job) {
String variable1 = (String)job.getVariablesAsMap().get("variable1");
sysout(variable1);
// ...
}
By default, the autoComplete
attribute is set to true
for any job worker.
Note that the described default behavior of auto-completion was introduced with 8.1 and was different before, see camunda-community-hub#239 for details.
In this case, the Spring integration will take care about job completion for you:
@JobWorker(type = "foo")
public void handleJobFoo(final ActivatedJob job) {
// do whatever you need to do
// no need to call client.newCompleteCommand()...
}
Which is the same as:
@JobWorker(type = "foo", autoComplete = true)
public void handleJobFoo(final ActivatedJob job) {
// ...
}
Note that the code within the handler method needs to be synchronously executed, as the completion will be triggered right after the method has finished.
When using autoComplete
you can:
- Return a
Map
,String
,InputStream
, orObject
, which then will be added to the process variables - Throw a
ZeebeBpmnError
which results in a BPMN error being sent to Zeebe - Throw any other
Exception
that leads in an failure handed over to Zeebe
@JobWorker(type = "foo")
public Map<String, Object> handleJobFoo(final ActivatedJob job) {
// some work
if (successful) {
// some data is returned to be stored as process variable
return variablesMap;
} else {
// problem shall be indicated to the process:
throw new ZeebeBpmnError("DOESNT_WORK", "This does not work because...");
}
}
Your job worker code can also complete the job itself. This gives you more control about when exactly you want to complete the job (e.g. allowing the completion to be moved to reactive callbacks):
@JobWorker(type = "foo", autoComplete = false)
public void handleJobFoo(final JobClient client, final ActivatedJob job) {
// do whatever you need to do
client.newCompleteCommand(job.getKey())
.send()
.exceptionally( throwable -> { throw new RuntimeException("Could not complete job " + job, throwable); });
}
Ideally, you don't use blocking behavior like send().join()
, as this is a blocking call to wait for the issues command to be executed on the workflow engine. While this is very straightforward to use and produces easy-to-read code, blocking code is limited in terms of scalability.
That's why the worker above showed a different pattern (using exceptionally
), often you might also want to use the whenComplete
callback:
send().whenComplete((result, exception) -> {})
This registers a callback to be executed if the command on the workflow engine was executed or resulted in an exception. This allows for parallelism. This is discussed in more detail in this blog post about writing good workers for Camunda Cloud.
Note that when completing jobs programmatically, you must specify autoComplete = false
. Otherwise, there is a race condition between your programmatic job completion and the Spring integration job completion, this can lead to unpredictable results.
You can use the @CustomHeaders
annotation for a parameter to retrieve custom headers for a job:
@JobWorker(type = "foo")
public void handleFoo(@CustomHeaders Map<String, String> headers){
// do whatever you need to do
}
Of course you can combine annotations, for example @VariablesAsType
and @CustomHeaders
@JobWorker
public ProcessVariables foo(@VariablesAsType ProcessVariables variables, @CustomHeaders Map<String, String> headers){
// do whatever you need to do
return variables;
}
Whenever your code hits a problem that should lead to a BPMN error being raised, you can simply throw a ZeebeBpmnError providing the error code used in BPMN:
@JobWorker(type = "foo")
public void handleJobFoo() {
// some work
if (!successful) {
// problem shall be indicated to the process:
throw new ZeebeBpmnError("DOESNT_WORK", "This does not work because...");
}
}
If you don't want to use a ZeebeClient for certain scenarios, you can switch it off by setting:
zeebe.client.enabled=false
zeebe.client.broker.gateway-address=127.0.0.1:26500
zeebe.client.security.plaintext=true
If you don't connect to the Camunda SaaS production environment you might have to also adjust these properties:
zeebe.client.cloud.base-url=zeebe.camunda.io
zeebe.client.cloud.port=443
zeebe.client.cloud.auth-url=https://login.cloud.camunda.io/oauth/token
As an alternative you can use the Zeebe Client environment variables.
If you build a worker that only serves one thing, it might also be handy to define the worker job type globally - and not in the annotation:
zeebe.client.worker.defaultType=foo
Number of jobs that are polled from the broker to be worked on in this client and thread pool size to handle the jobs:
zeebe.client.worker.max-jobs-active=32
zeebe.client.worker.threads=1
For a full set of configuration options please see ZeebeClientConfigurationProperties.java
Note that we generally do not advise to use a thread pool for workers, but rather implement asynchronous code, see Writing Good Workers.
If you need to customize the ObjectMapper that the Zeebe client uses to work with variables, you can declare a bean with type io.camunda.zeebe.client.api.JsonMapper
like this:
@Configuration
class MyConfiguration {
@Bean
public JsonMapper jsonMapper() {
ObjectMapper objectMapper = new ObjectMapper()
.configure(DeserializationFeature.ACCEPT_EMPTY_ARRAY_AS_NULL_OBJECT, true);
new ZeebeObjectMapper(objectMapper);
}
}
You can disable workers via the enabled
parameter of the @JobWorker
annotation :
class SomeClass {
@JobWorker(type = "foo", enabled = false)
public void handleJobFoo() {
// worker's code - now disabled
}
}
You can also override this setting via your application.properties
file:
zeebe.client.worker.override.foo.enabled=false
This is especially useful, if you have a bigger code base including many workers, but want to start only some of them. Typical use cases are
- Testing: You only want one specific worker to run at a time
- Load Balancing: You want to control which workers run on which instance of cluster nodes
- Migration: There are two applications, and you want to migrate a worker from one to another. With this switch, you can simply disable workers via configuration in the old application once they are available within the new.
You can override the JobWorker
annotation's values, as you could see in the example above where the enabled
property is overridden:
zeebe.client.worker.override.foo.enabled=false
In this case, foo
is the type of the worker that we want to customize.
You can override all supported configuration options for a worker, e.g.:
zeebe.client.worker.override.foo.timeout=10000
You could also provide a custom class that can customize the JobWorker
configuration values by implementing the io.camunda.zeebe.spring.client.annotation.customizer.ZeebeWorkerValueCustomizer
interface.
Spring-zeebe-starter will provide some out-of-the-box metrics, that can be leveraged via Spring Actuator. Whenever actuator is on the classpath, you can access the following metrics:
camunda.job.invocations
: Number of invocations of job workers (tagging the job type)camunda.connector.inbound.invocations
: Number of invocations of any inbound connectors (tagging the connector type)camunda.connector.outbound.invocations
: Number of invocations of any outbound connectors (tagging the connector type)
For all of those metrics, the following actions are recorded:
activated
: The job/connector was activated and started to process an itemcompleted
: The processing was completed successfullyfailed
: The processing failed with some exceptionbpmn-error
: The processing completed by throwing an BpmnError (which means there was no technical problem)
In a default setup, you can can enable metrics to be served via http:
management.endpoints.web.exposure.include=metrics
And then access them via http://localhost:8080/actuator/metrics/.
This project adheres to the Contributor Covenant Code of Conduct. By participating, you are expected to uphold this code. Please report unacceptable behavior to code-of-conduct@zeebe.io.