restart-operator

A Kubernetes operator for scheduling recurring restarts of workloads running in a Kubernetes cluster.

Overview

The Restart Operator allows you to define cron-based schedules for automatically restarting Kubernetes workloads. It's perfect for applications that require periodic restarts to clear memory, refresh connections, or apply configuration changes without requiring manual intervention.

Features

• Cron-based scheduling: Use standard cron expressions to define restart schedules
• Multiple workload support: Works with Deployments, StatefulSets, and DaemonSets
• Namespace scoping: Target resources in the same or different namespaces
• Status tracking: Keep track of the last successful restart and the next scheduled restart
• Cross-platform: Works on both ARM64 and AMD64 architectures

Installation

helm repo add archsyscall https://archsyscall.github.io/restart-operator
helm repo update
helm install restart-operator archsyscall/restart-operator

Usage

1. Create a RestartSchedule resource in your cluster:

apiVersion: restart-operator.k8s/v1alpha1
kind: RestartSchedule
metadata:
  name: nightly-app-restart
  namespace: default
spec:
  schedule: "0 3 * * *"
  targetRef:
    kind: Deployment
    name: my-application

2. Check the status of your restart schedule:

kubectl get restartschedule

Example output:

NAME                 TARGET-KIND   TARGET-NAME        SCHEDULE    LAST-RESTART           AGE
nightly-app-restart   Deployment    my-application     0 3 * * *   2025-05-03T03:00:00Z   2d

How It Works

The operator:

1. Watches for RestartSchedule resources
2. Validates the cron schedule and target resource
3. Creates a scheduled job using the cron expression
4. When the schedule triggers, adds a restart annotation to the target resource's pod template
5. Kubernetes sees the template change and initiates a rolling update
6. Updates the status with the last restart time and next scheduled restart

The restart is performed by adding/updating an annotation (restart-operator.k8s/restartedAt) on the pod template spec, which triggers Kubernetes to perform a rolling restart of the workload without modifying any other configuration.

License

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