Transform a Single‑Cluster CD Pipeline into a Multi‑Cluster System with ACK One

Overview

In cloud‑native environments, workloads may exceed the capacity of a single Kubernetes cluster, encounter regional resource shortages (e.g., GPU), or require disaster‑recovery and compliance across multiple regions. ACK One’s multi‑cluster application distribution enables users to spread resources across several clusters using strategies such as static weight, dynamic scheduling, and Gang scheduling, providing cross‑region intelligent scheduling, rescheduling, and fault migration.

Key Capabilities

Intelligent cross‑region, cross‑cluster resource scheduling : The Global Scheduler searches both existing node resources and elastic ECS/Serverless resources in all registered clusters to maximize utilization and avoid resource shortages.

Multi‑cluster Gang scheduling : Pre‑allocation or dynamic detection ensures that a workload can be scheduled simultaneously on multiple clusters.

Multiple distribution strategies : Supports copy, static weight, dynamic weight, etc.

Rescheduling and fault migration : If a workload fails in a sub‑cluster, the Global Scheduler recovers the job and redeploys it to another suitable cluster.

Workflow

Submit application manifests together with a PropagationPolicy that defines the distribution strategy to the ACK One fleet.

The Global Scheduler selects target clusters based on the policy and propagates the resources to those clusters.

If execution fails, the scheduler recovers the job and performs rescheduling or fault migration to a cluster with sufficient resources.

Building a Multi‑Cluster CD with ACK One and Cloud Effect

No changes are required to the original application YAML (e.g., Deployment).

Add a PropagationPolicy and, if needed, an OverridePolicy to the application directory in a Git/Codeup repository.

Update the Cloud Effect pipeline to deploy the application and the policies to the ACK One fleet instead of a single cluster.

Replace the single‑cluster kubeconfig with the fleet kubeconfig, inserting the fleet ID as the cluster identifier.

After the pipeline runs, the resources and policies are deployed to the fleet, which then distributes the application to the target clusters according to the defined policies. The original workload continues running without interruption or pod restarts, provided the workload YAML itself is unchanged.

Practical Steps

Create an ACK One fleet and register multiple clusters (public‑cloud or IDC) to the fleet.

Commit PropagationPolicy and optional OverridePolicy files to a Git repository. Example repository: https://github.com/AliyunContainerService/gitops-demo/tree/main/manifests/directory/mc-app

Modify the Cloud Effect pipeline to use the fleet kubeconfig and deploy the policies.

Execute the pipeline; the console shows deployment status and the fleet UI can be used to verify distribution across sub‑clusters.

Summary

ACK One’s multi‑cluster application distribution provides unified management of applications across clusters, reduces operational cost, and maximizes resource utilization through intelligent scheduling. By adding only two artifacts—a distribution policy ( PropagationPolicy) and, if needed, an override policy ( OverridePolicy)—to the Git repository and switching to the fleet kubeconfig, a single‑cluster continuous‑delivery pipeline can be upgraded to a multi‑cluster pipeline without modifying existing application manifests or causing pod restarts.