kubectl-rebalance

A kubectl plugin to redistribute pods from overloaded Kubernetes nodes by intelligently evicting them for rescheduling on less loaded nodes. Perfect for rebalancing clusters after node failures or replacements.

Features

• 🎯 Targeted eviction - Select pods by label selector
• 📊 Smart node selection - Sort nodes by CPU/memory usage or pod count
• 🔄 Controlled eviction - Configurable intervals between evictions
• 🧪 Safe operation - Dry-run mode to preview changes
• 📈 Metrics-optional - Works without metrics-server (falls back to pod count)
• ⚡ Real-world ready - Built for production scenarios

Installation

From source

# Clone the repository
git clone https://github.com/ssccio/rebalance.git
cd rebalance
# Build the plugin
go build -o kubectl-rebalance cmd/kubectl-rebalance/main.go

# Move to PATH
sudo mv kubectl-rebalance /usr/local/bin/

# Verify installation
kubectl rebalance --help

From releases

Download the latest release from the releases page:

# Linux
curl -L https://github.com/ssccio/rebalance/releases/latest/download/kubectl-rebalance_Linux_x86_64.tar.gz | tar xz
sudo mv kubectl-rebalance /usr/local/bin/

# macOS
curl -L https://github.com/ssccio/rebalance/releases/latest/download/kubectl-rebalance_Darwin_x86_64.tar.gz | tar xz
sudo mv kubectl-rebalance /usr/local/bin/

Using Krew (coming soon)

kubectl krew install rebalance

Usage

Basic usage

# Show help (just run without arguments)
kubectl rebalance

# Evict 10 pods with label 'name=php' from the 3 most loaded nodes
kubectl rebalance --count 10 --selector name=php --from-nodes 3
# Dry-run mode to see what would be evicted
kubectl rebalance --count 10 --selector name=php --dry-run

# Target nodes with over 90% utilization
kubectl rebalance --count 20 --selector app=web --threshold 90

# Ensure at least 5 pods are found before proceeding
kubectl rebalance --count 10 --selector name=php --min-pods 5

# Focus on memory pressure only
kubectl rebalance --count 10 --selector name=php --metric memory

# Target specific namespace
kubectl rebalance --count 10 --selector name=php --namespace production

For clusters without metrics-server

# Explicitly use pod count as the metric
kubectl rebalance --count 10 --selector name=php --metric pod-count

# The plugin automatically falls back to pod-count if metrics-server is unavailable
kubectl rebalance --count 10 --selector name=php --dry-run

Flags

Flag	Short	Description	Default
--selector	-l	Label selector for pods (REQUIRED)	-
--count	-c	Number of pods to evict	10
--from-nodes		Number of most loaded nodes to target	3
--threshold		CPU/Memory threshold percentage	80.0
--min-pods		Minimum pods required to proceed	0
--namespace	-n	Target namespace (empty for all)	""
--dry-run		Show what would be evicted without doing it	false
--metric		Metric type: cpu, memory, both, or pod-count	both
--output	-o	Output format: table, json, yaml	table

Use Cases

Post-node-replacement rebalancing

When nodes rejoin a cluster after maintenance or failure:

# Check node utilization
kubectl top nodes

# Preview the rebalancing plan
kubectl rebalance --count 50 --selector name=php --from-nodes 5 --dry-run

# Execute the rebalancing
kubectl rebalance --count 50 --selector name=php --from-nodes 5

Regular cluster maintenance

Periodically rebalance to prevent hot spots:

# Target the top 10% most loaded nodes
kubectl rebalance --count 20 --selector app=web --threshold 90 --from-nodes 10

Requirements

• kubectl configured with cluster access
• Metrics Server installed in the cluster (optional - falls back to pod count)
• Appropriate RBAC permissions:
  • List nodes and pods
  • Get node metrics (if using metrics-server)
  • Create pod evictions

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.