Cloud Instance Resource Scaling Without Pod Recreation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cloud instance scaling methods, such as Kubernetes' VPA, interrupt services when modifying resource quotas by requiring pod recreation, leading to service disruptions.
Innovation Solution
A cloud instance scaling method where worker nodes analyze status information to determine instances needing scaling, adjusting resource quotas in real time without recreating pods, using cgroup configurations to avoid interruptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the resource quota of a pod is modified by releasing and recreating the pod, then the resource quota can be changed, but the service running on the pod is interrupted
Solution Approach 1:
The patent applies dynamics by enabling the resource quota of running pods to be modified in real-time without requiring pod recreation. The system dynamically adjusts resource allocations by modifying cgroup configurations of running containers, allowing the resource quota to adapt to changing requirements while the pod and its services continue to run uninterrupted.
Solution Approach 2:
The patent changes the parameter of resource quota allocation from a static value set at pod creation to a dynamic value that can be modified during pod runtime. By changing the resource quota parameter through cgroup configuration adjustments rather than pod recreation, the system achieves flexible resource management without service interruption.
2Adaptability or versatility
If the pod is recreated to modify resource quota, then the resource quota can be updated, but the detection and scaling time increases
Solution Approach 1:
The patent applies preliminary action by pre-configuring the cgroup hierarchy and resource management structures during container initialization, enabling rapid resource quota adjustments later without requiring pod recreation. The foundational infrastructure is prepared in advance, allowing quick modification of resource allocations when scaling is needed.
Solution Approach 2:
The system implements dynamic resource quota modification that eliminates the time-consuming pod recreation process. Resource allocations can be adjusted in real-time by modifying cgroup parameters, enabling the system to respond quickly to scaling requirements without the delays associated with releasing and recreating pods.
3Extent of automation
If centralized management is implemented on all dockers by the master node, then orchestration and deployment can be managed, but the complexity of resource management increases
Solution Approach 1:
The patent introduces worker nodes as intermediaries between the master node and individual pods. The worker nodes handle the complex task of monitoring pod resource usage and executing resource quota modifications locally, while the master node maintains centralized orchestration and management. This intermediary layer distributes the management complexity from the master node to the worker nodes.
Solution Approach 2:
The patent segments the resource management function by separating the orchestration role (master node) from the execution role (worker nodes). The master node handles high-level orchestration and deployment, while worker nodes handle specific pod resource management tasks. This segmentation reduces the complexity burden on any single component of the system.
Data Source
AI summary
This application provides a cloud instance scaling method and a related device thereof, to ensure that a service running on a cloud instance is not interrupted when a resource quota of the cloud instance is increased or decreased. The method in this application includes: A first worker node obtains status information of a plurality of cloud instances. The first worker node determines, based on the status information, a to-be-scaled-up cloud instance from the plurality of cloud instances and a quantity of resources required for scale-up. If a quantity of idle resources of the first worker node is greater than or equal to the quantity of resources required for scale-up, the first worker node increases a resource quota of the to-be-scaled-up cloud instance based on the quantity of resources required for scale-up.