constellation/.github/actions/e2e_recover/action.yml
2023-08-18 10:43:51 +02:00

91 lines
3.5 KiB
YAML

name: Constellation recover
description: "Recover a Constellation cluster with an unavailable control plane."
inputs:
controlNodesCount:
description: "The amount of control plane nodes in the cluster."
required: true
kubeconfig:
description: "The kubeconfig for the cluster."
required: true
runs:
using: "composite"
steps:
- name: Restart worker node
shell: bash
env:
KUBECONFIG: ${{ inputs.kubeconfig }}
run: |
WORKER_NODE=$(kubectl get nodes --selector='!node-role.kubernetes.io/control-plane' -o json | jq '.items[0].metadata.name' -r)
echo "Disabling the join-service and waiting for the node to be unresponsive"
kubectl patch daemonset -n kube-system join-service -p '{"spec":{"template":{"spec":{"nodeSelector":{"some-tag":""}}}}}'
kubectl debug node/$WORKER_NODE --image=ubuntu -- bash -c "echo reboot > reboot.sh && chroot /host < reboot.sh"
kubectl wait --for=condition=Ready=Unknown --timeout=10m node/$WORKER_NODE
echo "Re-enabling the join-service and waiting for the node to be back up"
kubectl patch daemonset -n kube-system join-service --type=json -p='[{"op": "remove", "path": "/spec/template/spec/nodeSelector/some-tag"}]'
kubectl wait --for=condition=Ready=true --timeout=10m --all nodes
- name: Restart all control plane nodes
shell: bash
env:
KUBECONFIG: ${{ inputs.kubeconfig }}
run: |
CONTROL_PLANE_NODES=$(kubectl get nodes --selector='node-role.kubernetes.io/control-plane' -o json | jq '.items[].metadata.name' -r)
for CONTROL_PLANE_NODE in ${CONTROL_PLANE_NODES}; do
kubectl debug node/$CONTROL_PLANE_NODE --image=ubuntu -- bash -c "echo reboot > reboot.sh && chroot /host < reboot.sh"
done
- name: Constellation recover
shell: bash
run: |
timeout=600
start_time=$(date +%s)
recovered=0
while true; do
output=$(constellation recover --force)
if echo "$output" | grep -q "Pushed recovery key."; then
echo "$output"
i=$(echo "$output" | grep -o "Pushed recovery key." | wc -l | sed 's/ //g')
recovered=$((recovered+i))
if [[ $recovered -gt ${{ inputs.controlNodesCount }}/2 ]]; then
break
fi
fi
current_time=$(date +%s)
if ((current_time - start_time > timeout)); then
echo "Control plane recovery timed out after $timeout seconds."
exit 1
fi
echo "Did not recover a quorum (>${{inputs.controlNodesCount}}/2) of control-plane nodes yet, retrying in 5 seconds [$recovered/${{ inputs.controlNodesCount }}]"
sleep 5
done
- name: Wait for control plane to get back up
shell: bash
env:
KUBECONFIG: ${{ inputs.kubeconfig }}
run: |
timeout=600
start_time=$(date +%s)
while true; do
output=$(kubectl wait --for=condition=Ready --timeout=10m --all nodes || true)
if echo "$output" | grep -q "condition met"; then
echo "$output"
exit 0
fi
current_time=$(date +%s)
if ((current_time - start_time > timeout)); then
echo "Waiting for control plane to get back up timed out after $timeout seconds."
exit 1
fi
echo "Cannot reach control plane, retrying in 10 seconds"
sleep 10
done