Describe your problem

Although we have the spill/checkpoint mechanism to handle the scenarios out of memory locally, the existing mechanism is not suitable on the kubernetes, so we hope to introduce the Horizontal Pod Autoscaling mechanism to implement the autoscaling of the vineyard cluster on the kubernetes.
The issue mainly includes the following tasks.

• Find a proper watermark of memory usage to scale out or scale in.
• Expose the metric of memory usage to the vineyard operator.
• Create a HorizontalPodAutoscaler for controlling the scaling of the vineyard cluster.