49. Creating Your Own Auto-configuration

If you work in a company that develops shared libraries, or if you work on an open-source or commercial library, you might want to develop your own auto-configuration. Auto-configuration classes can be bundled in external jars and still be picked-up by Spring Boot.

Auto-configuration can be associated to a “starter” that provides the auto-configuration code as well as the typical libraries that you would use with it. We first cover what you need to know to build your own auto-configuration and then we move on to the typical steps required to create a custom starter.

[Tip]Tip

A demo project is available to showcase how you can create a starter step-by-step.

49.1 Understanding Auto-configured Beans

Under the hood, auto-configuration is implemented with standard @Configuration classes. Additional @Conditional annotations are used to constrain when the auto-configuration should apply. Usually, auto-configuration classes use @ConditionalOnClass and @ConditionalOnMissingBean annotations. This ensures that auto-configuration applies only when relevant classes are found and when you have not declared your own @Configuration.

You can browse the source code of spring-boot-autoconfigure to see the @Configuration classes that Spring provides (see the META-INF/spring.factories file).

49.2 Locating Auto-configuration Candidates

Spring Boot checks for the presence of a META-INF/spring.factories file within your published jar. The file should list your configuration classes under the EnableAutoConfiguration key, as shown in the following example:

org.springframework.boot.autoconfigure.EnableAutoConfiguration=\
com.mycorp.libx.autoconfigure.LibXAutoConfiguration,\
com.mycorp.libx.autoconfigure.LibXWebAutoConfiguration
[Note]Note

Auto-configurations must be loaded that way only. Make sure that they are defined in a specific package space and that they are never the target of component scanning. Furthermore, auto-configuration classes should not enable component scanning to find additional components. Specific @Imports should be used instead.

You can use the @AutoConfigureAfter or @AutoConfigureBefore annotations if your configuration needs to be applied in a specific order. For example, if you provide web-specific configuration, your class may need to be applied after WebMvcAutoConfiguration.

If you want to order certain auto-configurations that should not have any direct knowledge of each other, you can also use @AutoConfigureOrder. That annotation has the same semantic as the regular @Order annotation but provides a dedicated order for auto-configuration classes.

49.3 Condition Annotations

You almost always want to include one or more @Conditional annotations on your auto-configuration class. The @ConditionalOnMissingBean annotation is one common example that is used to allow developers to override auto-configuration if they are not happy with your defaults.

Spring Boot includes a number of @Conditional annotations that you can reuse in your own code by annotating @Configuration classes or individual @Bean methods. These annotations include:

49.3.1 Class Conditions

The @ConditionalOnClass and @ConditionalOnMissingClass annotations let @Configuration classes be included based on the presence or absence of specific classes. Due to the fact that annotation metadata is parsed by using ASM, you can use the value attribute to refer to the real class, even though that class might not actually appear on the running application classpath. You can also use the name attribute if you prefer to specify the class name by using a String value.

This mechanism does not apply the same way to @Bean methods where typically the return type is the target of the condition: before the condition on the method applies, the JVM will have loaded the class and potentially processed method references which will fail if the class is not present.

To handle this scenario, a separate @Configuration class can be used to isolate the condition, as shown in the following example:

@Configuration
// Some conditions
public class MyAutoConfiguration {

	// Auto-configured beans

	@Configuration
	@ConditionalOnClass(EmbeddedAcmeService.class)
	static class EmbeddedConfiguration {

		@Bean
		@ConditionalOnMissingBean
		public EmbeddedAcmeService embeddedAcmeService() { ... }

	}

}
[Tip]Tip

If you use @ConditionalOnClass or @ConditionalOnMissingClass as a part of a meta-annotation to compose your own composed annotations, you must use name as referring to the class in such a case is not handled.

49.3.2 Bean Conditions

The @ConditionalOnBean and @ConditionalOnMissingBean annotations let a bean be included based on the presence or absence of specific beans. You can use the value attribute to specify beans by type or name to specify beans by name. The search attribute lets you limit the ApplicationContext hierarchy that should be considered when searching for beans.

When placed on a @Bean method, the target type defaults to the return type of the method, as shown in the following example:

@Configuration
public class MyAutoConfiguration {

	@Bean
	@ConditionalOnMissingBean
	public MyService myService() { ... }

}

In the preceding example, the myService bean is going to be created if no bean of type MyService is already contained in the ApplicationContext.

[Tip]Tip

You need to be very careful about the order in which bean definitions are added, as these conditions are evaluated based on what has been processed so far. For this reason, we recommend using only @ConditionalOnBean and @ConditionalOnMissingBean annotations on auto-configuration classes (since these are guaranteed to load after any user-defined bean definitions have been added).

[Note]Note

@ConditionalOnBean and @ConditionalOnMissingBean do not prevent @Configuration classes from being created. The only difference between using these conditions at the class level and marking each contained @Bean method with the annotation is that the former prevents registration of the @Configuration class as a bean if the condition does not match.

49.3.3 Property Conditions

The @ConditionalOnProperty annotation lets configuration be included based on a Spring Environment property. Use the prefix and name attributes to specify the property that should be checked. By default, any property that exists and is not equal to false is matched. You can also create more advanced checks by using the havingValue and matchIfMissing attributes.

49.3.4 Resource Conditions

The @ConditionalOnResource annotation lets configuration be included only when a specific resource is present. Resources can be specified by using the usual Spring conventions, as shown in the following example: file:/home/user/test.dat.

49.3.5 Web Application Conditions

The @ConditionalOnWebApplication and @ConditionalOnNotWebApplication annotations let configuration be included depending on whether the application is a “web application”. A servlet-based web application is any application that uses a Spring WebApplicationContext, defines a session scope, or has a ConfigurableWebEnvironment. A reactive web application is any application that uses a ReactiveWebApplicationContext, or has a ConfigurableReactiveWebEnvironment.

49.3.6 SpEL Expression Conditions

The @ConditionalOnExpression annotation lets configuration be included based on the result of a SpEL expression.

49.4 Testing your Auto-configuration

An auto-configuration can be affected by many factors: user configuration (@Bean definition and Environment customization), condition evaluation (presence of a particular library), and others. Concretely, each test should create a well defined ApplicationContext that represents a combination of those customizations. ApplicationContextRunner provides a great way to achieve that.

ApplicationContextRunner is usually defined as a field of the test class to gather the base, common configuration. The following example makes sure that UserServiceAutoConfiguration is always invoked:

private final ApplicationContextRunner contextRunner = new ApplicationContextRunner()
		.withConfiguration(AutoConfigurations.of(UserServiceAutoConfiguration.class));
[Tip]Tip

If multiple auto-configurations have to be defined, there is no need to order their declarations as they are invoked in the exact same order as when running the application.

Each test can use the runner to represent a particular use case. For instance, the sample below invokes a user configuration (UserConfiguration) and checks that the auto-configuration backs off properly. Invoking run provides a callback context that can be used with Assert4J.

@Test
public void defaultServiceBacksOff() {
	this.contextRunner.withUserConfiguration(UserConfiguration.class).run((context) -> {
		assertThat(context).hasSingleBean(UserService.class);
		assertThat(context.getBean(UserService.class))
				.isSameAs(context.getBean(UserConfiguration.class).myUserService());
	});
}

@Configuration
static class UserConfiguration {

	@Bean
	public UserService myUserService() {
		return new UserService("mine");
	}

}

It is also possible to easily customize the Environment, as shown in the following example:

@Test
public void serviceNameCanBeConfigured() {
	this.contextRunner.withPropertyValues("user.name=test123").run((context) -> {
		assertThat(context).hasSingleBean(UserService.class);
		assertThat(context.getBean(UserService.class).getName()).isEqualTo("test123");
	});
}

The runner can also be used to display the ConditionEvaluationReport. The report can be printed at INFO or DEBUG level. The following example shows how to use the ConditionEvaluationReportLoggingListener to print the report in auto-configuration tests.

@Test
public void autoConfigTest {
	ConditionEvaluationReportLoggingListener initializer = new ConditionEvaluationReportLoggingListener(
			LogLevel.INFO);
	ApplicationContextRunner contextRunner = new ApplicationContextRunner()
			.withInitializer(initializer).run((context) -> {
					// Do something...
			});
}

49.4.1 Simulating a Web Context

If you need to test an auto-configuration that only operates in a Servlet or Reactive web application context, use the WebApplicationContextRunner or ReactiveWebApplicationContextRunner respectively.

49.4.2 Overriding the Classpath

It is also possible to test what happens when a particular class and/or package is not present at runtime. Spring Boot ships with a FilteredClassLoader that can easily be used by the runner. In the following example, we assert that if UserService is not present, the auto-configuration is properly disabled:

@Test
public void serviceIsIgnoredIfLibraryIsNotPresent() {
	this.contextRunner.withClassLoader(new FilteredClassLoader(UserService.class))
			.run((context) -> assertThat(context).doesNotHaveBean("userService"));
}

49.5 Creating Your Own Starter

A typical Spring Boot starter contains code to auto-configure and customize the infrastructure of a given technology, let’s call that "acme". To make it easily extensible, a number of configuration keys in a dedicated namespace can be exposed to the environment. Finally, a single "starter" dependency is provided to help users get started as easily as possible.

Concretely, a custom starter can contain the following:

  • The autoconfigure module that contains the auto-configuration code for "acme".
  • The starter module that provides a dependency to the autoconfigure module as well as "acme" and any additional dependencies that are typically useful. In a nutshell, adding the starter should provide everything needed to start using that library.

This separation in two modules is in no way necessary. If "acme" has several flavours, options or optional features, then it is better to separate the auto-configuration as you can clearly express the fact some features are optional. Besides, you have the ability to craft a starter that provides an opinion about those optional dependencies. At the same time, others can rely only on the autoconfigure module and craft their own starter with different opinions.

If the auto-configuration is relatively straightforward and does not have optional feature, merging the two modules in the starter is definitely an option.

49.5.1 Naming

You should make sure to provide a proper namespace for your starter. Do not start your module names with spring-boot, even if you use a different Maven groupId. We may offer official support for the thing you auto-configure in the future.

As a rule of thumb, you should name a combined module after the starter. For example, assume that you are creating a starter for "acme" and that you name the auto-configure module acme-spring-boot and the starter acme-spring-boot-starter. If you only have one module that combines the two, name it acme-spring-boot-starter.

49.5.2 Configuration keys

If your starter provides configuration keys, use a unique namespace for them. In particular, do not include your keys in the namespaces that Spring Boot uses (such as server, management, spring, and so on). If you use the same namespace, we may modify these namespaces in the future in ways that break your modules. As a rule of thumb, prefix all your keys with a namespace that you own (e.g. acme).

Make sure that configuration keys are documented by adding field javadoc for each property, as shown in the following example:

@ConfigurationProperties("acme")
public class AcmeProperties {

	/**
	 * Whether to check the location of acme resources.
	 */
	private boolean checkLocation = true;

	/**
	 * Timeout for establishing a connection to the acme server.
	 */
	private Duration loginTimeout = Duration.ofSeconds(3);

	// getters & setters

}
[Note]Note

You should only use simple text with @ConfigurationProperties field Javadoc, since they are not processed before being added to the JSON.

Here are some rules we follow internally to make sure descriptions are consistent:

  • Do not start the description by "The" or "A".
  • For boolean types, start the description with "Whether" or "Enable".
  • For collection-based types, start the description with "Comma-separated list"
  • Use java.time.Duration rather than long and describe the default unit if it differs from milliseconds, e.g. "If a duration suffix is not specified, seconds will be used".
  • Do not provide the default value in the description unless it has to be determined at runtime.

Make sure to trigger meta-data generation so that IDE assistance is available for your keys as well. You may want to review the generated metadata (META-INF/spring-configuration-metadata.json) to make sure your keys are properly documented. Using your own starter in a compatible IDE is also a good idea to validate that quality of the metadata.

49.5.3 autoconfigure Module

The autoconfigure module contains everything that is necessary to get started with the library. It may also contain configuration key definitions (such as @ConfigurationProperties) and any callback interface that can be used to further customize how the components are initialized.

[Tip]Tip

You should mark the dependencies to the library as optional so that you can include the autoconfigure module in your projects more easily. If you do it that way, the library is not provided and, by default, Spring Boot backs off.

Spring Boot uses an annotation processor to collect the conditions on auto-configurations in a metadata file (META-INF/spring-autoconfigure-metadata.properties). If that file is present, it is used to eagerly filter auto-configurations that do not match, which will improve startup time. It is recommended to add the following dependency in a module that contains auto-configurations:

<dependency>
	<groupId>org.springframework.boot</groupId>
	<artifactId>spring-boot-autoconfigure-processor</artifactId>
	<optional>true</optional>
</dependency>

With Gradle 4.5 and earlier, the dependency should be declared in the compileOnly configuration, as shown in the following example:

dependencies {
	compileOnly "org.springframework.boot:spring-boot-autoconfigure-processor"
}

With Gradle 4.6 and later, the dependency should be declared in the annotationProcessor configuration, as shown in the following example:

dependencies {
	annotationProcessor "org.springframework.boot:spring-boot-autoconfigure-processor"
}

49.5.4 Starter Module

The starter is really an empty jar. Its only purpose is to provide the necessary dependencies to work with the library. You can think of it as an opinionated view of what is required to get started.

Do not make assumptions about the project in which your starter is added. If the library you are auto-configuring typically requires other starters, mention them as well. Providing a proper set of default dependencies may be hard if the number of optional dependencies is high, as you should avoid including dependencies that are unnecessary for a typical usage of the library. In other words, you should not include optional dependencies.

[Note]Note

Either way, your starter must reference the core Spring Boot starter (spring-boot-starter) directly or indirectly (i.e. no need to add it if your starter relies on another starter). If a project is created with only your custom starter, Spring Boot’s core features will be honoured by the presence of the core starter.