Configuring the Number of HetuEngine Worker Nodes

Scenario for Configuring the Number of HetuEngine Worker Nodes

On the HetuEngine web UI, you can adjust the number of worker nodes for a compute instance so that worker nodes can be added when they are insufficient and reduced when they are idle. The number of worker nodes can be adjusted manually or automatically.

When an instance is being scaled in or out, the original services are not affected and the instance can still be used.
Dynamic 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 instances are dynamically scaled in or out, the number of worker nodes displayed in the instance configuration area on the HSConsole page remains the initial value and does not change with dynamic scaling.
The dynamic instance scale-in/out function will be affected if the HSBroker and Yarn services are restarted after the function is enabled. Disable the function before you restart the services.
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.
You can set the timeout period for manual instance scale in/out. For this, 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).
You can set whether to wait for tasks when YARN resources are insufficient during manual scale-out.
On FusionInsight Manager, choose HetuEngine > Configurations > All Configurations, search for application.customized.properties, and add the yarn.hetuserver.engine.worker.scale.out.resource.limit parameter, the options are as follows:
- true (default value): The system checks the YARN resources usage. If the resources are sufficient, they are used for scale-out. If the resources are insufficient, they cannot be used for scale-out.
- false: Scale-out tasks are delivered to YARN without calculating whether there are sufficient Yarn resources. If the resources are sufficient, they are used for scale-out. If the resources are insufficient, the scale-out task waits in a queue.

Procedure for Configuring the Number of HetuEngine Worker Nodes

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.
In the Basic Information area on the Dashboard tab page, click the link next to HSConsole WebUI. The HSConsole page is displayed.

In the Compute Instance page, locate the row that contains the tenant to which the target instance belongs and click Configure in the Operation column.

If manual scaling is required, change the value of Quantity in the Worker Container Resource Configuration area on the configuration page and click OK. The compute instance status changes to SCALING OUT or SCALING IN. After the scaling is complete, the compute instance status changes to RUNNING.

If automatic scaling is required, set Scaling in Advanced Configuration to Yes and configure the following parameters according to Table 1 to enable dynamic scaling.

Compute instances in the Running state are scaled in or out based on the configured auto scaling parameters. For compute instances in other states, only the configuration is saved, and the saved configuration takes effect when the compute instances are restarted.

**Table 1** Dynamic scaling parameters
Parameter	Description	Example Value
Load Collection Period	The interval for collecting instance load information, in seconds.	10
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	The number of worker nodes to be added each time when the instance starts to scale out.	1
Scale-out Decision Period	The interval for determining whether to scale out an instance, in seconds.	200
Scale-out Timeout Period	The timeout period of the scale-out operation, in seconds.	400
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	The number of worker nodes to be reduced each time when the instance starts to scale in.	1
Scale-in Decision Period	The interval for determining whether to scale in an instance, in seconds.	300
Scale-in Timeout Period	The timeout period of the scale-in operation, in seconds.	600