Status

Current state: [ UNDER DISCUSSION | ACCEPTED | REJECTED ]

Discussion thread: <link to mailing list DISCUSS thread>

JIRA: SAMZA-2405

Released: 

Problem

Samza Yarn follows a multi-stage deployment model, where Job Runner, which runs on the submission host, reads configuration, performs planning and persist config in the coordinator stream before submitting the job to Yarn cluster. In Yarn, Application Master (AM) reads config from coordinator stream before spinning up containers to execute. Split of responsibility between job runner and AM is operationally confusing, and makes debugging the pipeline difficult with multiple points of failure. In addition, since planning invokes user code, it usually requires isolation on the runner from security perspective to guard the framework from malicious user code, or a malicious user can gain access to other user jobs running on the same runner. 

Proposed Changes

We will provide a pluggable config retrieval interface on AM, which will simplify the job submission to Yarn, without involving any complex logic. AM on the other hand, will read job config using the provided config loader, performs planning, generate DAG and persist the final config back to coordinator stream.

We will also make changes to start up script, run-app.sh, such that it does not read local config files anymore. All job submission related configs needs to be explicitly provided with --config.

Public Interfaces

We will introduce one job configs to configure the job to load configuration on AM:

• job.config.loader.class

The changes are backward incompatible as we are removing the usage of --config-factory & --config-path. Instead, we will ask for explicit configurations related to job submission such job name, yarn package path using --config.

All the configs provided in the start up script will be passed to AM through environment variable and loaded by the designated config loader to load the complete config. 

Take wikipedia-feed in Hello Samza as an example:

deploy/samza/bin/run-app.sh \
  --config job.name=wikipedia-stats \
  --config job.factory.class=org.apache.samza.job.yarn.YarnJobFactory \
  --config yarn.package.path=file://${basedir}/target/${project.artifactId}-${pom.version}-dist.tar.gz \
  --config job.config.loader.class==org.apache.samza.config.loader.PropertiesConfigLoader \
  --config config.path=/__package/config/wikipedia-feed.properties

Implementation and Test Plan

JobConfig

We will add two new configs in JobConfig to control whether to read AM from ConfigLoader instead of coordinator stream.

// Configuration to a fully qualified class name to load config from.
public static final String CONFIG_LOADER_CLASS = "job.config.loader.class";

ConfigLoader

Interface which AM relies on to read configuration from. It takes in a properties map, which defines the variables it needed in order to get the proper config.

This interface will replace the existing ConfigFactory interface as we no longer need complex configs in runner anymore. Providing minimum Yarn related configs using --config when invoking run-app.sh will be sufficient.

public interface ConfigLoader {
  /**
   * Build a specific Config given job submission config.
   * @param config Config specified during job submission containing information necessary for this ConfigLoader to fetch the complete config.
   * @return Newly constructed Config.
   */
  Config getConfig(Config config);
}

PropertiesConfigLoader

Default implementation of ConfigLoader, which reads "path" from the input properties, which leads to a property file.

public class PropertiesConfigLoader extends ConfigLoader {
  /**
   * Build a specific Config given job submission config.
   * @param config Config specified during job submission containing information necessary for this ConfigLoader to fetch the complete config.
   * @return Newly constructed Config.
   */
  override def getConfig(config: Config): Config = {
    val path = config.get("config.path")

    val props = new Properties()
    val in = new FileInputStream(path)

    props.load(in)
    in.close()

    debug("got config %s from config %s" format (props, path))

    new MapConfig(props.asScala.asJava)
  }
}

RemoteApplicationRunner

RemoteApplicationRunner#run will simplify submit the job to Yarn given the submission configs.

@Override
public void run(ExternalContext externalContext) {
  JobRunner runner = new JobRunner(config);
  runner.getJobFactory().getJob(config).submit();
}

YarnJob

YarnJob#buildEnvironment will build config loader class as well as the submission config env variable.

private[yarn] def buildEnvironment(config: Config, yarnConfig: YarnConfig,
    jobConfig: JobConfig): Map[String, String] = {
    val envMapBuilder = Map.newBuilder[String, String]

    envMapBuilder += ShellCommandConfig.ENV_CONFIG_LOADER_CLASS ->
      Util.envVarEscape(SamzaObjectMapper.getObjectMapper.writeValueAsString(jobConfig.getConfigLoaderClassName))
    envMapBuilder += ShellCommandConfig.ENV_CONFIG ->
      Util.envVarEscape(SamzaObjectMapper.getObjectMapper.writeValueAsString(config))

ClusterBasedJobCoordinator

ClusterBasedJobCoordinator#main will construct the application config through config loader.

Compatibility, Deprecation, and Migration Plan

Backward Incompatible. 

Changes will be announced in Samza 1.3 and takes effect in Samza 1.4

Users need to change job submission script and provide related configs explicitly through --config, instead of using --config-factory and --config-path to load local file;