AWS re:Invent 2025 - Manage multicloud Kubernetes at scale feat. Adobe (HMC322)

Managing Multicloud Kubernetes at Scale: Insights from AWS and Adobe

Multicloud Strategies and Considerations

• Multicloud refers to running IT solutions or workloads across at least two cloud service providers
• Customers adopt multicloud strategies for various reasons, such as business needs, mergers and acquisitions, or leveraging differentiated cloud capabilities
• AWS partners with customers to help them build technical and non-technical capabilities for successful multicloud implementations

Kubernetes as a Multicloud Abstraction

• Kubernetes has emerged as a fundamental abstraction used by multicloud customers to streamline their experience
• Key considerations for multicloud Kubernetes fleet management include:
  • Cluster lifecycle management
  • Add-on lifecycle management (networking, security, observability)
  • Handling different cluster configurations based on workload requirements
  • Supporting multiple Kubernetes distributions

GitOps-based Multicloud Kubernetes Management

• A successful approach for multicloud Kubernetes fleet management is using a GitOps-based approach
• Customers store cluster and add-on configurations, as well as security, audit, and compliance policies, in Git as the source of truth
• They then leverage GitOps operators like Argo CD and cloud-provider-specific controllers to continuously provision and reconcile the state of their multicloud Kubernetes fleet

Adobe's Multicloud Kubernetes Journey

Ethos: Adobe's Platform Engineering Approach

• Adobe's mission is to help developers write better software faster, even in a complex multicloud environment
• Adobe built "Ethos" - a consistent interface and platform to simplify the developer experience across multiple cloud providers
• Ethos enables developers to:
  • Integrate and manage cloud-native infrastructure using uniform CI/CD pipelines
  • Focus on delivering business value rather than dealing with cloud-specific security or compliance requirements

Ethos at Scale

• Ethos orchestrates around 4 million containers daily, with 90% of Adobe's containerized footprint running on it
• Ethos manages 4,000 distinct services across Adobe's three clouds (Digital Experience, Creative, and Document)
• This footprint is deployed across 450 independent Ethos clusters globally, spanning 30 regions, multiple public cloud providers, and private data centers

Declarative Infrastructure and GitOps Workflows

• Adobe's approach is rooted in declarative infrastructure, with cluster configurations, network policies, versions, and resource definitions managed as code in Git
• They use GitOps tools like Argo CD and Helm to automate the deployment and reconciliation of the desired state
• This enables features like automated rollbacks, continuous pipelines, and automated tests to ensure consistency between production and the source of truth in Git

Modular and Open-Source Ethos Platform

• Ethos is built as a modular codebase, using Helm as a package manager to deploy components and security configurations
• This allows for rapid updates, easier maintenance, and customization for different business needs or technical challenges like multicloud
• Adobe embraces open-source tools like cert-manager, Cluster API, and OpenTelemetry, enabling their own teams to contribute to the platform

Automating Multicloud Kubernetes Lifecycle

• Developers provide input on cluster configurations (name, region, cloud provider) through manifest files
• Argo CD listens to Git signals and uses cloud-provider-specific controllers (e.g., ACK, Cluster API) to provision and manage the desired Kubernetes infrastructure
• This approach enables a fully automated, GitOps-driven workflow for deploying and managing Kubernetes clusters across multiple cloud providers

Multicloud Kubernetes Decommissioning

• Adobe has implemented safeguards to prevent accidental or error-prone infrastructure deletions, including:
  • Running tests and input validation
  • Checking for active resources (e.g., PDBs, ingress controllers) before deletion
• All decommissioning actions are driven through the Git-based workflow, without the need for manual console interactions

Key Takeaways and Lessons Learned

• Invest in building your team's Kubernetes expertise and platform engineering capabilities
• Automate as much as possible using GitOps workflows to reduce manual errors and accelerate delivery
• Leverage the open-source community and ecosystem to build your multicloud Kubernetes platform
• Treat your platform as a product, gathering feedback from developers and continuously improving it
• Embrace a "cookie-cutter" approach to enable repeatable, cloud-agnostic deployments, while still accommodating unique business requirements

Impact and Outcomes

• Adobe's Ethos platform has enabled them to:
  • Deploy changes across their 450-cluster fleet 3 times faster
  • Perform full Kubernetes cluster upgrades 2 times faster
  • Reduce cluster provisioning time by 25%
• The Ethos platform has been well-received by Adobe's developers, leading to happier teams and better business outcomes