Updated on 2022-11-18 GMT+08:00

Adjusting the Number of Worker Nodes

Scenarios

On the HetuEngine web UI, you can adjust the number of worker nodes for a compute instance. In this way, resources can be expanded for the compute instance when resources are insufficient and released when the resources are idle. The number of workers can be adjusted manually or automatically.

Prerequisites

You have created a user for accessing the HetuEngine web UI. For details, see Creating a HetuEngine User.

  • When an instance is being scaled in or out, the original services are not affected and the instance can still be used.
  • Instance scale-in/out is delayed to implement smooth adjustment of resource consumption within a long period of time. It cannot respond to the requirements of running SQL tasks for available resources in real time.
  • After the instance scale-in/out function is enabled, restarting the HSBroker and Yarn services affects the scale-in/out function. If you need to restart the services, you are advised to disable the instance scale-in/out function first.
  • Before scaling out a compute instance, ensure that the current queue has sufficient resources. Otherwise, the scale-out cannot reach the expected result and subsequent scale-in operations will be affected.
  • To perform manual scale-in/out, log in to Manager, choose HetuEngine > Configurations > All Configurations, search for application.customized.properties, and add the yarn.hetuserver.engine.flex.timeout.sec parameter. The default value is 300 (in seconds).

Procedure

  1. Log in to FusionInsight Manager as a user who can access the HetuEngine web UI and choose Cluster > Services > HetuEngine. The HetuEngine service page is displayed.
  2. In the Basic Information area on the Dashboard tab page, click the link next to HSConsole WebUI. The HSConsole page is displayed.
  3. Click Compute Instance.
  4. Locate the row that contains the target instance, and click Configure in the Operation column.
  5. If manual scale-in/out is required, change the number of workers on the configuration page and click OK. The compute instance enters the SCALING OUT or SCALING IN state. After the scale-in/out is complete, the compute instance status changes to RUNNING.
  6. If automatic scale-in/out is required, choose Configure Instance > Advanced Configuration and click the Scaling switch.

    • OFF: Disable dynamic scale-in/out.
    • ON: Enable dynamic scale-in/out. For details, see Table 1. Figure 1 shows the configuration page.
      Table 1 Parameters for dynamic scale-in/out

      Parameter

      Description

      Example Value

      Scale-out Threshold

      When the average value of the instance resource usage in the scale-in/out decision-making period exceeds the threshold, the instance starts to scale out.

      0.9

      Scale-out Size

      Number of Workers to be added each time when the instance starts to scale out.

      1

      Scale-out Decision Period

      Interval for determining whether to scale out an instance. Unit: second

      200

      Scale-in Threshold

      When the average value of the instance resource usage in the scale-in/out decision-making period exceeds the threshold, the instance starts to scale in.

      0.1

      Scale-in Size

      Number of Workers to be reduced each time when the instance starts to scale in.

      1

      Scale-in Decision Period

      Interval for determining whether to scale in an instance. Unit: second

      300

      Load Collection Period

      Interval for collecting instance load information. Unit: second

      10

      Scale-out Timeout Period

      Timeout period of the scale-out operation. Unit: second

      400

      Scale-in Timeout Period

      Timeout period of the scale-in operation. Unit: second

      600

      Figure 1 Scaling out/in an instance