Backup Kubernetes Cluster State

README.md

Run this in order to backup all you k8s cluster data. It will be saved in a folder bkp. To restore the cluster, you can run kubectl apply -f bkp.

Please note: this recovers all resources correctly, including dynamically generated PV's. However, it will not recover ELB endpoints. You will need to update any DNS entries manually, and manually remove the old ELB's.

Please note: This has not been tested with all resource types. Supported resource types include:

• services
• replicationcontrollers
• secrets
• deployments
• horizontal pod autoscalers
• persistent volumes
• persistent volume claims
• quotas
• limits
• namespaces
• storageclasses
• daemonsets

backup.sh

#!/bin/bash -e

# Adapted from:
# https://github.com/colhom/coreos-docs/blob/cluster-dump-restore/kubernetes/cluster-dump-restore.md

ABSOLUTE_PATH="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/../bkp"
mkdir -p $ABSOLUTE_PATH

kubectl get --export -o=json ns | \
jq '.items[] |
    del(.status,
        .metadata.uid,
        .metadata.selfLink,
        .metadata.resourceVersion,
        .metadata.creationTimestamp,
        .metadata.generation
    )' > $ABSOLUTE_PATH/ns.json

echo "" > $ABSOLUTE_PATH/cluster-dump.json
for ns in $(jq -r '.metadata.name' < $ABSOLUTE_PATH/ns.json);do
    echo "Namespace: $ns"

    kubectl --namespace="${ns}" get --export -o=json svc,rc,secrets,ds,cm,deploy,hpa,pv,pvc,quota,limits,ns,storageclass | \
    jq '.items[] |
        select(.type!="kubernetes.io/service-account-token") |
        del(
            .spec.clusterIP, # clusterIP is dynamically assigned
            .spec.claimRef,  # Throw this out so we can rebind
            .metadata.uid,
            .metadata.selfLink,
            .metadata.resourceVersion,
            .metadata.creationTimestamp,
            .metadata.generation,
            .spec.template.spec.securityContext,
            .spec.template.spec.terminationGracePeriodSeconds,
            .spec.template.spec.restartPolicy,
            .spec?.ports[]?.nodePort? # Delete nodePort from service since this is dynamic
        ) |

        # Set reclaim policy to retain so we can recover volumes
        if .kind == "PersistentVolume" then 
            .spec.persistentVolumeReclaimPolicy = "Retain" 
        else
            . 
        end' >> $ABSOLUTE_PATH/cluster-dump.json
done