selfhosted-services

Services running on hubbe.club, in local k8s cluster

SOPS setup for config/secret .yaml decryption

• Install sops
  • See https://github.com/mozilla/sops
• Get age keys.txt from backup storage
  • Set SOPS_AGE_RECIPIENTS env var to public key
• To decrypt all secrets in repo:
  • Run find . -type f -name '*.yaml' -exec sops --decrypt --in-place '{}' \;

k8s Cluster Setup

1. Set up machines with latest version of Debian
2. Adjust settings in init-scripts/INSTALL_SETTINGS
   • The other scripts pull variables from this file:
     • KUBE_VERSION sets the kubernetes version for the cluster
     • POD_NETWORK_CIDR sets the pod networking subnet, this should be set to avoid conflicts with existing networking
     • K8S_SERVICE_CIDR sets the service subnet, this should also avoid conflicting
     • CONTROL_PLANE_ENDPOINT sets the kube-apiserver loadbalancer endpoint, which allows for stacked HA setups
       • This causes init and join-controlplane scripts to deploy kube-vip as a static pod
         • see https://kube-vip.io/docs/installation/static/ for details
       • See https://kubernetes.io/docs/setup/production-environment/tools/kubeadm/high-availability/
       • See https://github.com/kubernetes/kubeadm/blob/master/docs/ha-considerations.md
3. (optional) Adjust kubeadm configuration in init-scripts/kubeadm-configs
   • single.yaml is used if CONTROL_PLANE_ENDPOINT is not set
   • stacked-ha.yaml is used if CONTROL_PLANE_ENDPOINT is set
   • This file is needed in order to set the cgroupDriver for the kubelet on kubeadm init
   • For info on what the configurations in this file control, see:
     • https://pkg.go.dev/k8s.io/kubelet/config/v1beta1?utm_source=godoc#KubeletConfiguration
     • https://pkg.go.dev/k8s.io/kube-proxy/config/v1alpha1#KubeProxyConfiguration
     • https://pkg.go.dev/k8s.io/kubernetes/cmd/kubeadm/app/apis/kubeadm/v1beta2?utm_source=godoc#InitConfiguration
     • https://pkg.go.dev/k8s.io/kubernetes/cmd/kubeadm/app/apis/kubeadm/v1beta2?utm_source=godoc#ClusterConfiguration
4. Set up the first node of the control-plane with init-scripts/kubeadm-init.sh
   • Required packages are installed - containerd.io, kubeadm, kubelet, and kubectl
   • Project Calico is installed as a CNI plugin, see core/calico.yaml
   • Kernel source address verification is enabled by the init-scripts/install-prereqs.sh script
   • Note that the control plane nodes are not un-tainted, and thus no user pods are scheduled on them by default
   • Note that control plane nodes also have the node.kubernetes.io/exclude-from-external-load-balancers label, which must be removed for MetalLB to work with user pods on them.
5. (optional) Add additional control-plane nodes with init-scripts/kubeadm-join-controlplane.sh <node_user>@<node_address>
   • <node_user> must be able to SSH to the node, and have sudo access
   • Same inital setup is performed as on first node
   • Nodes are then joined as control-plane nodes using kubeadm token create --print-join-command and kubeadm join
6. Add in worker nodes by running init-scripts/kubeadm-join-worker.sh <node_user>@<node_address>
   • <node_user> must be able to SSH to the node, and have sudo access
   • Same initial setup is performed as on first node
   • Nodes are then joined as worker nodes using kubeadm token create --print-join-command and kubeadm join

Storage setup

• NFS shares for volumes PVCs need to be created
  • Must be accessible from the IPs of the nodes in the k8s cluster
  • Deploy volumes (PV/PVC) with kubectl apply -f volumes
• Longhorn needs no additional setup - simply deploy operators/longhorn with kubectl apply -f operators/longhorn
  • The storageclass settings can be tweaked if needed for volume HA stuff - see operators/longhorn/storageclass.yaml
• At this stage, restore volumes from Longhorn backups using the Longhorn UI.
  • Make sure to also create PVs/PVCs via the Longhorn UI.
• Longhorn handles backups - see Longhorn UI and/or default settings in operators/longhorn/deployment.yaml

Ingress setup

• Set up cert-manager for automated cert renew
  • Run kubectl apply -f core/cert-manager.yml
  • Check core/cert-issuer/*.yaml.example files
    • Alternatively, run sops --decrypt --in-place on existing files
    • Run kubectl apply -f core/cert-issuer
• Deploy metalLB for nginx LoadBalancer service
  • kubectl apply -f core/metallb && sops -d core/metallb/memberlist-secret.yaml | kubectl apply -f -
• Deploy ingress-nginx for reverse proxying to deployed pods
  • Run kubectl apply -f core/nginx/
  • Current configuration assumes a single wildcard cert, ingress-nginx/tls for all sites
  • Issued by LetsEncrypt, solved by CloudFlare DNS verification
  • See nginx/certificate.yaml for certificate request fulfilled by cert-manager

Apps setup

• Some apps use operators, install these from operators folder.
  • dragonfly-operator for DragonflyDB/Redis, Cloudnative-PG for postgres
• Some apps need GPU acceleration (jellyfin)
  • This uses node-feature-discovery to target the gpu plugin deployment
    • kubectl apply -f core/nfd.yaml
  • See https://github.com/intel/intel-device-plugins-for-kubernetes/tree/master/cmd/gpu_plugin
  • See core/gpu-plugins
• Create configs from *.yaml.example files
  • Alternatively, run sops --decrypt --in-place on existing files
• Set PUID, PGID, and TZ variables in apps/0-linuxserver-envs.yaml
• Deploy apps
  • All apps can be deployed simply with kubectl apply -R -f apps/ once SOPS decryption is done
  • If deploying single apps, remember to also deploy related configs
    • Most things need the apps/0-linuxserver-envs.yaml ConfigMap

Monitoring setup

• See monitoring/README.md
  • mainly monitoring/build.sh and monitoring/apply.sh
• After deployment of monitoring stack, deploy extra rules from extras folder