bittorrent/kube-backup repository overview

⁠kube-backup

Quick 'n dirty kubernetes state backup script, designed to be ran as kubernetes Job. Think of it like RANCID⁠ for kubernetes.

Props to @gianrubio and @pieterlange for coming up with the idea and implementing them.

⁠Setup

Use the deployment example (ssh⁠ or AWS CodeCommit⁠ authentication) and deploy a kubernetes CronJob primitive in your kubernetes (1.5 and up) cluster ensuring backups of kubernetes resource definitions to your private git repo.

Define the following environment parameters:

• GIT_REPO - GIT repo url. Required
• NAMESPACES - List of namespaces to export. Default: all
• GLOBALRESOURCES - List of global resource types to export. Default: namespace
• RESOURCETYPES - List of resource types to export. Default: ingress deployment configmap svc rc ds thirdpartyresource networkpolicy statefulset storageclass cronjob. Notice that Secret objects are intentionally not exported by default.
• GIT_USERNAME - Display name of git user. Default: kube-backup
• GIT_EMAIL - Email address of git user. Default: [email protected]
• GIT_BRANCH - Use a specific git branch . Default: master

Chose one of two authentication mechanisms:

• If using AWS CodeCommit and policy-based access from AWS, modify your cluster configuration to provide GitPull and GitPush access for that CodeCommit repo to your cluster. If using kops, the configuration will look something like this:

  additionalPolicies:
    node: |
      [
        {
          "Effect": "Allow",
          "Action": [
            "codecommit:GitPull",
            "codecommit:GitPush"
          ],
          "Resource": "arn:aws:codecommit:<region>:<account name>:<repo-name>"
        }
      ]

NOTE: in this deployment, the ssh volume and secret are not present.

• If using a different repository (GitHub, BitBucket, etc.), mount a configured ssh directory in /backup/.ssh with the following files:

  • known_hosts - Preloaded with SSH host key of $GIT_REPO host.
  • id_rsa - SSH private key of user allowed to push to $GIT_REPO.

Easiest way of doing this is:

ssh-keygen -f ./id_rsa
ssh-keyscan $YOUR_GIT_HOST > known_hosts

kubectl create secret generic kube-backup-ssh -n kube-system --from-file=id_rsa --from-file=known_hosts

NOTE: If id_rsa isn't found in your ssh directory, the backup script will assume you're using AWS CodeCommit.

Optional:

• Modify the snapshot frequency in spec.schedule using the cron format⁠.
• Modify the number of successful and failed finished jobs to retain in spec.successfulJobsHistoryLimit and spec.failedJobsHistoryLimit.
• If using RBAC (1.6+), use the ClusterRole and ClusterRoleBindings in rbac.yaml.

⁠Result

All configured resources will be exported into a directory tree structure in kind: List YAML format following a $namespace/$resourcetype.yaml file structure.

.
├── default
│   ├── configmap.yaml
│   ├── cronjob.yaml
│   ├── deployment.yaml
│   ├── ds.yaml
│   ├── ingress.yaml
│   ├── networkpolicy.yaml
│   ├── petset.yaml
│   ├── rc.yaml
│   ├── statefulset.yaml
│   ├── storageclass.yaml
│   ├── svc.yaml
│   └── thirdpartyresource.yaml
├── kube-system
│   ├── configmap.yaml
│   ├── cronjob.yaml
│   ├── deployment.yaml
│   ├── ds.yaml
│   ├── ingress.yaml
│   ├── networkpolicy.yaml
│   ├── petset.yaml
│   ├── rc.yaml
│   ├── statefulset.yaml
│   ├── storageclass.yaml
│   ├── svc.yaml
│   └── thirdpartyresource.yaml
├── prd
│   ├── configmap.yaml
│   ├── cronjob.yaml
│   ├── deployment.yaml
│   ├── ds.yaml
│   ├── ingress.yaml
│   ├── networkpolicy.yaml
│   ├── petset.yaml
│   ├── rc.yaml
│   ├── statefulset.yaml
│   ├── storageclass.yaml
│   ├── svc.yaml
│   └── thirdpartyresource.yaml
└── staging
    ├── configmap.yaml
    ├── cronjob.yaml
    ├── deployment.yaml
    ├── ds.yaml
    ├── ingress.yaml
    ├── networkpolicy.yaml
    ├── petset.yaml
    ├── rc.yaml
    ├── statefulset.yaml
    ├── storageclass.yaml
    ├── svc.yaml
    └── thirdpartyresource.yaml

4 directories, 48 files

⁠Caveat

This is using a kubernetes alpha feature (cronjobs⁠) and hasn't been tested for idempotency/concurrent behaviour. See the cronjob documentation⁠ for details.

If your kubernetes cluster runs under version 1.5 or less, spec.successfulJobsHistoryLimit and spec.failedJobsHistoryLimit will be ignored as they've been introduced in version 1.6. In this case, running an export every 10 minutes will quickly run up your Job (and therefor Pod) count, causing a linear increase in master server load. A fix for this is to deploy a blunt instrument⁠ to clean the old kube-backup jobs.

⁠License

This project is MIT licensed.