Void: Cloud-Native Node Hibernation Framework

Void is a cloud-native framework designed to hibernate and manage bare-metal worker nodes in Kubernetes clusters. It enables efficient resource utilization, cost savings, and environmental sustainability by automatically identifying and hibernating underutilized nodes while ensuring seamless application continuity.

Warning

Void project is still in an early stage which means that any feature changes are very welcome and the features aren't 100% feasible.

Key Features

Automated Hibernation: Void monitors node utilization and triggers hibernation for idle nodes based on user-defined policies.
Seamless Wake-up: Nodes wake up based on workload demand, ensuring minimal disruption.
Resource Optimization: Dynamic node state adjustments optimize resource allocation.
Kubernetes Integration: Fully integrates with Kubernetes APIs for centralized management.
Scalable and Reliable: Designed for scalability and reliability in production environments.

Initial Workflow (Further research needed)

Monitoring Node Utilization: Continuously monitor CPU, memory, and custom metrics to determine node activity.
Define Hibernation Policies: Create policies specifying conditions for hibernating nodes based on resource thresholds.
Trigger Hibernation: Execute commands or API calls to put nodes into low-power states (e.g., using server power management features or out-of-band tools like IPMI).
Preserve Data and State: Gracefully shut down applications, move services to other nodes, and preserve data/configurations.
Manage Wake-up Criteria: Monitor workload demand and define criteria for waking up hibernated nodes based on increased demand or specific service needs.
Integrate with Kubernetes: Use Kubernetes APIs to update node states, coordinate with the scheduler, and manage resource distribution.
Monitor Health: Continuously check hibernated node health, perform periodic checks, and take corrective actions if issues arise.

Challenges

Important

The challenge with running and managing bare metal K8s clusters today is that servers and operating systems are managed separately from the K8s nodes lifecycle.

Resource Fluctuations: Handle rapid changes in resource utilization to avoid early hibernation.

Workload Variability: Manage varying workloads to prevent unnecessary node state changes.

Critical Services: Prioritize critical services to prevent hibernation disruptions.

Scheduled Jobs: Adjust hibernation criteria for periodic workload changes.

High Availability: Implement failover mechanisms to maintain service continuity.

Graceful Shutdown: Ensure applications gracefully shut down before hibernation.

Health Checks: Perform health checks to verify node and application readiness.

Pending Tasks: Handle pending tasks or operations before hibernation.

Emergency Wake-up: Implement rapid response procedures for critical situations.

Fault Tolerance: Handle failures and provide error recovery mechanisms.

Development Tech Stack

Primary programming language for development

Go Language:

Go Language Docs: https://go.dev/

Rust/C (Optional): Utilized for interacting with the operating system and triggering hibernation.

Prometheus Integration

Prometheus Client Library: Instrument metrics for monitoring.

Prometheus Client Library Docs: https://prometheus.io/docs/instrumenting/clientlibs/

Prometheus Node Exporter: Collect system metrics for monitoring.

Prometheus Node Exporter Docs: https://prometheus.io/docs/guides/node-exporter/

Kubernetes Tools

Operator SDK: Build Kubernetes operators for automation.

Operator SDK Docs: https://sdk.operatorframework.io/

Kubebuilder: SDK for building Kubernetes APIs using CRDs.

Kubebuilder Docs: https://book.kubebuilder.io/

ClusterAPI Cluster API is a project that brings K8s-style APIs and an “end-state” approach to cluster lifecycle management (creation, configuration, upgrade, destruction).

ClusterAPI Docs: https://cluster-api.sigs.k8s.io/