Cloud-Native Meets Carbon Intelligence: Multi-Cluster Sustainability with Liqo and Karmada
- 1 day ago
- 4 min read
By Ananta Cloud Engineering Team | September 11, 2025
The cloud was built for scale. But at Ananta Cloud, we believe it should also be built for sustainability.
As enterprises shift toward multi-cluster Kubernetes architectures across hybrid and multi-cloud environments, one crucial factor has often been overlooked in workload scheduling decisions: carbon intensity.
This post explores how to implement carbon-aware Kubernetes scheduling using two powerful open-source tools: Liqo and Karmada—and how Ananta Cloud is pioneering these green ops movement.
The Problem: Compute Without Conscience
Kubernetes is excellent at managing efficiency at scale—optimizing resources, automating failover, and distributing workloads. But by default, Kubernetes does not consider the environmental impact of where workloads run.
Traditional Scheduling Logic:
Node CPU/memory availability
Pod/node affinities
Taints and tolerations
Cost or availability zone
Missing piece: Where is the power coming from? Does that node run on coal, solar, hydro, or wind?
💡 Fun Fact: Running a large ML training workload in a coal-powered data center can emit hundreds of kilograms of CO₂, while the same job in a renewable-powered region could be near-zero emissions.
Solution Overview: Building a Carbon-Aware Multi-Cluster Kubernetes Stack
We combine:
Liqo – to federate clusters and allow transparent offloading
Karmada – to orchestrate workload placement across clusters
Carbon Intensity APIs – to inform real-time, data-driven decisions
Component Deep Dive
1️⃣ Liqo: Dynamic Kubernetes Federation via Virtual Nodes
What it does: Liqo connects Kubernetes clusters via peer-to-peer federation, allowing one cluster to “consume” resources from another. It introduces virtual nodes in the local cluster that map to real nodes in a remote cluster.
How it helps:
Enables offloading workloads to clusters powered by clean energy
Supports hybrid (on-prem + cloud) and multi-cloud architectures
Works transparently with existing K8s workloads
Example: You run a data pipeline in Cluster A (coal-powered) during the day. At night, Cluster B (solar-powered, cheaper) becomes optimal. Liqo offloads the job without code changes.
Key Features
API compatibility with standard Kubernetes
Support for persistent volumes
Cross-cluster network peering and DNS resolution
Fine-grained offloading policies
2️⃣ Karmada: Multi-Cluster Orchestration with Declarative Policies
What it does: Karmada acts as a control plane across multiple Kubernetes clusters, enabling you to deploy workloads once and have them automatically propagated based on customizable rules.
How it helps:
Centralized workload placement
Cluster health and availability monitoring
Intelligent resource propagation via PropagationPolicy and OverridePolicy
Key Features
Failover and rescheduling across clusters
Integration with GitOps pipelines
Extensible scheduling logic (can be carbon-aware!)
3️⃣ Carbon Intensity APIs: External Sustainability Signals
You can integrate real-time and forecasted carbon data using:
ElectricityMap
WattTime
Tomorrow.io
These APIs give access to:
Real-time carbon intensity (gCO2/kWh)
Forecasts for the next 24–48 hours
Emissions by region/grid operator
Architecture Overview
High-Level Workflow
Carbon-Aware Controller (custom) fetches current carbon intensity for each cluster region.
It assigns a carbon score to each cluster.
The controller updates Karmada's PropagationPolicy with cluster preferences based on the lowest score.
If needed, Liqo handles transparent offloading to the selected foreign cluster.
sequenceDiagram
participant Dev as DevOps Engineer
participant Karmada as Karmada Control Plane
participant CarbonAPI as Carbon Intensity API
participant Liqo as Liqo Cluster Peer
participant Workload as Kubernetes Workload
Dev->>Karmada: Define multi-cluster deployment
Karmada->>CarbonAPI: Fetch carbon data
CarbonAPI-->>Karmada: Return carbon intensity scores
Karmada->>Karmada: Evaluate best cluster (lowest CO₂)
Karmada->>Liqo: Offload workload to clean cluster
Liqo->>Workload: Run workload in green region
Real-World Use Case Scenarios
01 - AI/ML Training Jobs
Challenge: High GPU workloads have huge carbon impact.
Solution: Schedule model training in regions with lowest emissions during off-peak times.
02 - Enterprise SaaS
Challenge: Need to meet ESG (Environmental, Social, Governance) goals.
Solution: Report CO₂ savings with audit logs of carbon-aware workload decisions.
03 - Global CDN Services
Challenge: Geo-distributed services have region-based redundancy.
Solution: When latency difference is minimal, prefer green-powered clusters.
Implementation Guide
Setup Liqo Across Clusters
liqoctl install kind --cluster-name cluster-a
liqoctl peer cluster-b --auth-mode token
Deploy Karmada and Join Clusters
# Install Karmada
curl -sSL https://raw.githubusercontent.com/karmada-io/karmada/main/hack/deploy/karmada.sh | bash
# Join clusters
kubectl karmadactl join cluster-a --cluster-kubeconfig=kubeconfig-a
kubectl karmadactl join cluster-b --cluster-kubeconfig=kubeconfig-b
Create a Custom Carbon Score Controller
Use Kubernetes CRDs or an admission controller that:
Queries carbon APIs hourly
Updates PropagationPolicy clusterAffinity weights
Adds carbon score as a label to clusters (e.g., co2-score=low)
Define a PropagationPolicy Example
apiVersion: policy.karmada.io/v1alpha1
kind: PropagationPolicy
metadata:
name: carbon-aware-policy
spec:
resourceSelectors:
- apiVersion: apps/v1
kind: Deployment
name: green-service
placement:
clusterAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 10
preference:
matchExpressions:
- key: co2-score
operator: In
values:
- low
Monitoring and Observability
Use Prometheus + Grafana to visualize:
CO₂ intensity per cluster
Carbon savings over time
Workload migration logs
Add custom Grafana panels using data from:
Carbon APIs
Karmada logs
Liqo virtual node usage
Challenges and Considerations
Challenge | Solution |
Data latency in carbon APIs | Use forecasted data and moving averages |
SLA impact from offloading | Set fallback logic in Karmada policies |
Security concerns in cluster federation | Use Liqo’s RBAC and TLS options |
Multi-cloud cost trade-offs | Integrate with cost optimization tools (e.g., Kubecost) |
Summary: Why This Matters
Implementing carbon-aware Kubernetes scheduling:
Helps enterprises meet net-zero goals
Optimizes for performance, cost, and carbon
Turns cloud-native infrastructure into a climate-positive asset
At Ananta Cloud, we believe that innovation must be sustainable. Leveraging Liqo and Karmada, we’re bringing a new dimension to cloud orchestration: one where code meets conscience.
Ready to take your Kubernetes strategy to the next level and reduce your carbon footprint?
Partner with Ananta Cloud to deploy a carbon-aware CI/CD pipeline tailored to your business.
Email: hello@anantacloud.com | LinkedIn: @anantacloud | Schedule Meeting
Comments