Flink Streaming Spring Boot

The application submits a Flink job via Flink's remote environment. The job is submitted once the spring boot server starts up.

How to run

See com.tenble.flink.streaming.springboot.FlinkStreamingSpringBootTest as an example.

1. Import this project as a dependency for another project.
2. Add @EnableAutoConfigure to your spring boot app.
3. During spring's context loading phase, autowire the bean StreamExecutionEnvironment flinkEnvironment in another bean and add operations to the environment.
4. Copy the shade plugin step in pom.xml of this project to create an executable spring boot jar. See <transformer implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer" /> to specify a mainClass.
5. Run mvn clean package on your project, which will create an executable spring boot application, with all dependencies put onto the classpath. Can be found in the target directory.
6. Run this jar using java -jar ${JAR_FILE}. Specify any spring property overrides as jvm or program arguments.

Properties

flink-properties:
  job-name: "FlinkStreamingSpringBoot"                              # flink job name
  job-manager-url: "localhost"                                      # hostname for the flink job manager
  job-manager-port: 8081                                            # REST port of the flink job manager
  remote-env-jar-files:                                             # any jars to upload to the flink job manager alongside your job
    - "target/flink-streaming-spring-boot-0.0.1-SNAPSHOT.jar"
  max-client-rest-request-size-bytes: 2000000000                    # maximum job size for the REST request to restrict on the client in bytes
  terminate: true                                                   # terminate application after the allocated time has passed
  termination-grace-period-ms: 120000                               # amount of time to wait before this spring boot application terminates

Kubernetes files

The files in the folder k8s/ are slightly modified files from Flink's Kubernetes Setup page. They have been modified to increase the maximum allowable rest request size.

Caveats

1. Java API for Flink does not have an option to use detached mode, which is why the termination-grace-period-ms is necessary.