Managing Optimizers

The optimizer is the execution unit for performing self-optimizing tasks on a table. To isolate optimizing tasks on different tables and support the deployment of optimizers in different environments, Amoro has proposed the concepts of optimizer containers and optimizer groups:

• Optimizer container: Encapsulate the deployment method of optimizers, there are three implementations for now: flink container based on Flink streaming job, local container based on Java Application, and external container based on manually started by users.
• Optimizer group: A collection of optimizers, where each table must select an optimizer group to perform optimizing tasks on it. Tables under the same optimizer group contribute resources to each other, and tables under different optimizer groups can be isolated in terms of optimizer resources.
• Optimizer: The specific unit that performs optimizing tasks, usually with multiple concurrent units.

Optimizer container

Before using self-optimizing, you need to configure the container information in the configuration file. Optimizer container represents a specific set of runtime environment configuration, and the scheduling scheme of optimizer in that runtime environment. The container includes three types: flink, local, and external.

Local container

Local container is a way to start Optimizer by local process and supports multi-threaded execution of Optimizer tasks. It is recommended to be used only in demo or local deployment scenarios. If the environment variable for jdk is not configured, the user can configure java_home to point to the jdk root directory. If already configured, this configuration item can be ignored.

containers:
  - name: localContainer
    container-impl: com.netease.arctic.server.manager.LocalOptimizerContainer
    properties:
      export.JAVA_HOME: "/opt/java"   # JDK environment

Flink container

Flink container is a way to start Optimizer through Flink jobs. With Flink, you can easily deploy Optimizer on yarn clusters or kubernetes clusters to support large-scale data scenarios. To use flink container, you need to add a new container configuration. with container-impl as com.netease.arctic.server.manager.FlinkOptimizerContainer

FlinkOptimizerContainer support the following properties:

An example for yarn-per-job mode:

containers:
  - name: flinkContainer
    container-impl: com.netease.arctic.server.manager.FlinkOptimizerContainer
    properties:
      flink-home: /opt/flink/                              #flink install home
      export.HADOOP_CONF_DIR: /etc/hadoop/conf/            #hadoop config dir
      export.HADOOP_USER_NAME: hadoop                      #hadoop user submit on yarn
      export.JVM_ARGS: -Djava.security.krb5.conf=/opt/krb5.conf       #flink launch jvm args, like kerberos config when ues kerberos
      export.FLINK_CONF_DIR: /etc/hadoop/conf/              #flink config dir

An example for kubernetes-application mode:

containers:
  - name: flinkContainer
    container-impl: com.netease.arctic.server.manager.FlinkOptimizerContainer
    properties:
      flink-home: /opt/flink/                                                        # Flink install home
      target: kubernetes-application                                                 # Flink run as native kubernetes
      job-uri: "local:///opt/flink/usrlib/optimizer-job.jar"                         # Optimizer job main jar for kubernetes application
      ams-optimizing-uri: thrift://ams.amoro.service.local:1261                      # AMS optimizing uri 
      export.FLINK_CONF_DIR: /opt/flink/conf/                                        # Flink config dir
      flink-conf.kubernetes.container.image: "arctic163/optimizer-flink:{version}"   # Optimizer image ref
      flink-conf.kubernetes.service-account: flink                                   # Service account that is used within kubernetes cluster.

External container

External container refers to the way in which the user manually starts the optimizer. The system has a built-in external container called external, so you don’t need to configure it manually.

Optimizer group

Optimizer group (optimizer resource group) is a concept introduced to divide Optimizer resources. An Optimizer Group can contain several optimizers with the same container implementation to facilitate the expansion and contraction of the resource group.

Add optimizer group

You can add an optimizer group on the Amoro dashboard by following these steps:

1.Click the “Add Group” button in the top left corner of the Optimizer Groups page.

2.Configure the newly added Optimizer group.

The following configuration needs to be filled in:

• name: the name of the optimizer group, which can be seen in the list of optimizer groups on the front-end page.
• container: the name of a container configured in containers.
• properties: the default configuration under this group, is used as a configuration parameter for tasks when the optimize page is scaled out.Supports native parameters for flink on yarn, and users can set parameters using the flink-conf.<property>=<value> or use flink-conf.yaml to configure parameters.

The optimizer group supports the following properties:

Edit optimizer group

You can click the edit button on the Optimizer Groups page to modify the configuration of the Optimizer group.

Remove optimizer group

You can click the remove button on the Optimizer Groups page to delete the optimizer group, but only if the group is not referenced by any catalog or table and no optimizer belonging to this group is running.

Optimizer Scale-out and Release

Scale out optimizer

You can click the Scale-Out button on the Optimizer Groups page to scale out the optimizer for the corresponding optimizer group, and then click OK to start the optimizer for this optimizer group according to the parallelism configuration. If the optimizer runs normally, you will see an optimizer with the status RUNNING on the Optimizers page.

Release optimizer

You can click the Release button on the Optimizer page to release the optimizer.

Deploy external optimizer

You can submit optimizer in your own Flink task development platform or local Flink environment with the following configuration. The main parameters include:

./bin/flink run-application -t yarn-application \
 -Djobmanager.memory.process.size=1024mb \
 -Dtaskmanager.memory.process.size=2048mb \
 -Dtaskmanager.memory.managed.size=32mb \
 -Dtaskmanager.memory.network.max=32mb \
 -Dtaskmanager.memory.network.min=32mb \
 -c com.netease.arctic.optimizer.flink.FlinkOptimizer \
 ${AMORO_HOME}/plugin/optimizer/flink/optimizer-job.jar \
 -a 127.0.0.1:1261 \
 -g flinkGroup \
 -p 1 \
 -eds \
 -dsp /tmp \
 -msz 512

The description of the relevant parameters is shown in the following table: