Similar to Cloud Foundry’s blue-green deployments, you can perform rolling upgrades on the applications orchestrated by Spring Cloud Data Flow.

Let’s start with the following simple stream definition.

dataflow:>stream create --name foo --definition "time | log" --deploy

List Apps.

→ cf apps
Getting apps in org test-org / space development as test@pivotal.io...
OK

name       requested state   instances   memory   disk   urls
foo-log    started           1/1         1G       1G     foo-log.cfapps.io
foo-time   started           1/1         1G       1G     foo-time.cfapps.io

Let’s assume you’ve to make an enhancement to update the "logger" to append extra text in every log statement.

@SpringBootApplication
@Import(LogSinkConfiguration.class)
public class DemoApplication {

	@Autowired
	private LogSinkProperties properties;

	public static void main(String[] args) {
		SpringApplication.run(DemoApplication.class, args);
	}

	@Bean
	@ServiceActivator(inputChannel = Sink.INPUT)
	public LoggingHandler logSinkHandler() {
		LoggingHandler loggingHandler = new LoggingHandler(this.properties.getLevel().name());
		loggingHandler.setExpression(this.properties.getExpression());
		loggingHandler.setLoggerName("TEST [" + this.properties.getName() + "]");
		return loggingHandler;
	}
}

Let’s deploy the locally built application to Cloud Foundry

→ cf push foo-log-v2 -p demo-0.0.1-SNAPSHOT.jar -n foo-log-v2 --no-start

List Apps.

→ cf apps
Getting apps in org test-org / space development as test@pivotal.io...
OK

name       requested state   instances   memory   disk   urls
foo-log    started           1/1         1G       1G     foo-log.cfapps.io
foo-time   started           1/1         1G       1G     foo-time.cfapps.io
foo-log-v2 stopped           1/1         1G       1G     foo-log-v2.cfapps.io

The stream applications do not communicate via (Go)Router, so they aren’t generating HTTP traffic. Instead, they communicate via the underlying messaging middleware such as Kafka or RabbitMQ. In order to rolling upgrade to route the payload from old to the new version of the application, you’d have to replicate the SPRING_APPLICATION_JSON environment variable from the old application that includes spring.cloud.stream.bindings.input.destination and spring.cloud.stream.bindings.input.group credentials.

	Note
	You can find the SPRING_APPLICATION_JSON of the old application via: "cf env foo-log".

cf set-env foo-log-v2 SPRING_APPLICATION_JSON '{"spring.cloud.stream.bindings.input.destination":"foo.time","spring.cloud.stream.bindings.input.group":"foo"}'

Let’s start foo-log-v2 application.

cf start foo-log-v2

As soon as the application bootstraps, you’d now notice the payload being load balanced between two log application instances running on Cloud Foundry. Since they both share the same "destination" and "consumer group", they are now acting as competing consumers.

Old App Logs:

2016-08-08T17:11:08.94-0700 [APP/0]      OUT 2016-08-09 00:11:08.942  INFO 19 --- [ foo.time.foo-1] log.sink                                 : 08/09/16 00:11:08
2016-08-08T17:11:10.95-0700 [APP/0]      OUT 2016-08-09 00:11:10.954  INFO 19 --- [ foo.time.foo-1] log.sink                                 : 08/09/16 00:11:10
2016-08-08T17:11:12.94-0700 [APP/0]      OUT 2016-08-09 00:11:12.944  INFO 19 --- [ foo.time.foo-1] log.sink                                 : 08/09/16 00:11:12

New App Logs:

2016-08-08T17:11:07.94-0700 [APP/0]      OUT 2016-08-09 00:11:07.945  INFO 26 --- [ foo.time.foo-1] TEST [log.sink                       : 08/09/16 00:11:07]
2016-08-08T17:11:09.92-0700 [APP/0]      OUT 2016-08-09 00:11:09.925  INFO 26 --- [ foo.time.foo-1] TEST [log.sink                       : 08/09/16 00:11:09]
2016-08-08T17:11:11.94-0700 [APP/0]      OUT 2016-08-09 00:11:11.941  INFO 26 --- [ foo.time.foo-1] TEST [log.sink                       : 08/09/16 00:11:11]

Deleting the old version foo-log from the CF CLI would make all the payload consumed by the foo-log-v2 application. Now, you’ve successfully upgraded an application in the streaming pipeline without bringing it down in entirety to do an adjustment in it.

List Apps.

→ cf apps
Getting apps in org test-org / space development as test@pivotal.io...
OK

name       requested state   instances   memory   disk   urls
foo-time   started           1/1         1G       1G     foo-time.cfapps.io
foo-log-v2 started           1/1         1G       1G     foo-log-v2.cfapps.io

	Note
	A comprehensive canary analysis along with rolling upgrades will be supported via Spinnaker in future releases.