Skip to content

Product Architecture

The management interface of Multi-Cloud Orchestration is primarily responsible for the following functionalities:

  • Lifecycle management (LCM) of multicloud instances (based on Karmada)
  • Serving as a unified entry point for multicloud products (OpenAPI, Kairship UI, internal module GRPC calls)
  • Proxying API requests for multicloud instances (in the native Karmada style)
  • Aggregating cluster information (monitoring, management, control) within multicloud instances
  • Management and monitoring of multicloud workloads and other resources
  • Potential future permission operations

Core Components

Multi-Cloud Orchestration mainly consists of two core components: kairship apiserver and kairship controller-manager .

Kairship Apiserver

The kairship apiserver primarily serves as the entry point for all traffic in Multi-Cloud Orchestration, including OpenAPI and GRPC. It acts as a unified entry point for all APIs. The APIs are defined using protobuf , which generates corresponding frontend and backend code. It supports both HTTP Restful and GRPC through grpc-gateway .

During startup, it retrieves identity information of the operator from the Global Management Module for subsequent security verification during AuthZ.

Kairship Controller-Manager

This is the controller for Multi-Cloud Orchestration, responsible for instance status synchronization, resource collection, Karmada instance registration, and global resource registration.

In a multi-replica deployment, it utilizes the leader mechanism to maintain a single working pod at any given time (similar to Kubernetes' controller-manager election mechanism).

This component handles a series of control logic for Multi-Cloud Orchestration, with each logic represented as a separate controller. It listens for changes in specific objects through the list-watch mechanism and processes corresponding events. The main controllers include:

  • Virtual Cluster Sync Controller:

    • Listens for CRUD events of Multi-Cloud Orchestration instance CRDs. When a Multi-Cloud Orchestration instance is created, it synchronizes the creation of corresponding virtual cluster management resources.
    • Retrieval of all resources for the Multi-Cloud Orchestration instance (Multi-Cloud Workloads, PP, OP) is accomplished through the acceleration mechanism within the Container Management Module (leveraging Clusterpedia), enabling read-write separation and improved performance.
    • When an instance is deleted, it synchronously deletes the virtual cluster registered in the Container Management Module.
  • Resource Statistics Controller:

    • Collects statistical information for all clusters joined by Multi-Cloud Orchestration instances and writes it back to the Multi-Cloud Orchestration instance CRD (e.g., total CPU, memory, and node count in the clusters managed by the instance).
  • Status Sync Controller:

    • Handles status synchronization and statistics for Multi-Cloud Orchestration instances.
  • Instance Registry Controller:

    • Multi-Cloud Orchestration registers all Karmada instances in the platform to the Global Management Module using custom resources. This allows for the binding of roles and Karmada instances in global management. Ultimately, these bindings are synchronized to the Multi-Cloud Orchestration module.
  • Ghippo Webhook Controller:

    • After the binding of roles and Karmada instances is completed in the Global Management Module, the Multi-Cloud Orchestration module is informed via SDK to perform authorization actions.

Data Flow Diagram

Data Flow Diagram

It is important to note that multicloud instances are not aware of each other and are isolated from one another.

Multi-Cloud Orchestration Management:

  • Retrieves distribution policies and application status information related to Karmada.
  • Retrieves cluster and node statistics, monitoring information, and management and control information for multicloud instances.
  • Edits, updates, and deletes information related to multicloud applications in Karmada instances (mainly focused on Karmada workloads and PP, OP).

All request data flows directly to the multicloud instances located in the Global Services Cluster.

Next, all access requests are routed to the corresponding instances after passing through the Multi-Cloud Orchestration. All read requests such as get/list access the Container Management Module, while write requests access Karmada instances, achieving read-write separation and improving response time.

You may wonder how the Container Management Module retrieves resource information for multicloud instances. The solution is to add the instance itself as a virtual cluster to the Container Management Module (not displayed in the Container Management). This allows for complete utilization of the Container Management Module's capabilities (accelerated retrieval of resources, CRDs, etc.). When querying resources (Deployments, deployment policies, differentiation policies, etc.) for a specific multicloud instance in the interface, it can be directly retrieved through the Container Management Module.

Comments