NOTES.md

github-runner-autoscaler

notes

on new cluster

helm install prometheus prometheus-community/kube-prometheus-stack helm install keda kedacore/keda -n keda --set prometheus.operator.enabled=true --set prometheus.metricServer.enabled=true --create-namespace kubectl create ns runners ./cert.sh kubectl apply -f runners-creds.secret.yaml ( cd operator make install deploy;kubectl delete deployment -n operator-system operator-controller-manager kubectl apply -f ../crs.yaml make run )

Known issues

• The authentication isnt actually authentication
• Runner labels are essentially backwards

This is only required for my home cluster

for i in {0..3}; do echo '{"kind":"PersistentVolume","apiVersion":"v1","metadata":{"name":"hp-pv-000","labels":{"type":"local"}},"spec":{"storageClassName":"manual","capacity":{"storage":"5Gi"},"accessModes":["ReadWriteOnce"],"hostPath":{"path":"/tmp/data00"}}}' | sed "s/00"/0$i"/g" | kubectl apply -f - done

temp - pending fix

kubectl apply -f - <<EOF kind: TriggerAuthentication apiVersion: keda.sh/v1alpha1 metadata: name: test123 namespace: runners spec: secretTargetRef: - key: cert name: cert parameter: cert - key: ca name: cert parameter: ca - key: key name: cert parameter: key EOF

• TODO: Cert auth is just used as a way of fudging TLS
• TODO: Make runners close session on shutdown
• TODO: Have option to not output labels - could be sensitive
• TODO: Watcher issue

• TODO: Get GIT_TOKEN

Cleanup

kubectl delete -f ../crs.yaml sleep 3s make undeploy (echo runners;kubectl get ns | grep -Po 'test-repo\S+') | xargs kubectl delete ns

for node in kubectl get nodes -o yaml | grep -oP '10\.99\.\d+\.\d+' | sort | uniq; do ssh setup@$node "sudo bash -c "docker image ls | grep joeshearn | grep -Po '^\S+\s+\S+' | sed -E 's/\s+/:/g' | xargs docker image rm -f "" done

kubectl create ns runners make deploy IMG=joeshearn/github-runner-autoscaler-operator:master kubectl apply -f ../crs.yaml