Deploy-Laboratory/ansible/playbooks/longhorn-install.yml

---
# 部署：docs/00-05 §2 步骤 3——Helm 铺栈；验收见 scripts/verify.sh run 03-07。
# Helm 安装 Longhorn（与 docs/03-07 一致）。在控制节点执行，依赖 KUBECONFIG=/etc/rancher/k3s/k3s.yaml
# 变量：group_vars/all.yml 中 longhorn_chart_version、longhorn_install_node_packages、longhorn_apply_local_path_lab

- name: Longhorn node packages (iSCSI, NFS client)
  hosts: k3s_nodes
  become: true
  tasks:
    - name: Install Longhorn OS dependencies
      when: longhorn_install_node_packages | default(true) | bool
      block:
        - name: Install iscsi + nfs (dnf/yum)
          ansible.builtin.package:
            name:
              - iscsi-initiator-utils
              - nfs-utils
            state: present

        - name: Enable iscsid
          ansible.builtin.systemd:
            name: iscsid
            enabled: true
            state: started

    - name: Ensure Longhorn data subdirectory exists on all nodes
      ansible.builtin.file:
        path: "{{ k3s_data_dir }}/longhorn"
        state: directory
        mode: "0700"

    - name: Pre-pull Longhorn images on all nodes (optional, avoid DockerHub EOF/ImagePullBackOff)
      when: longhorn_prepull_images | default(true) | bool
      ansible.builtin.shell: |
        set -e
        CTR="ctr --address /run/k3s/containerd/containerd.sock -n k8s.io"

        imgs=(
          "docker.io/longhornio/longhorn-manager:v{{ longhorn_chart_version }}"
          "docker.io/longhornio/longhorn-ui:v{{ longhorn_chart_version }}"
          "docker.io/longhornio/longhorn-share-manager:v{{ longhorn_chart_version }}"
          "docker.io/longhornio/longhorn-engine:v{{ longhorn_chart_version }}"
          "docker.io/longhornio/longhorn-instance-manager:v{{ longhorn_chart_version }}"
          "docker.io/longhornio/backing-image-manager:v{{ longhorn_chart_version }}"
          "docker.io/longhornio/support-bundle-kit:v0.0.45"
        )

        for img in "${imgs[@]}"; do
          ok=0
          for i in 1 2 3 4 5; do
            echo "[pull] $img (try $i/5)"
            if $CTR images pull "$img"; then
              ok=1
              break
            fi
            sleep $((i * 3))
          done
          if [ "$ok" -ne 1 ]; then
            echo "[ERR] failed pulling $img after retries"
            exit 1
          fi
        done
      args:
        executable: /bin/bash
      changed_when: true

- name: Install Longhorn with Helm on first server
  hosts: k3s_server
  become: true
  run_once: true
  vars:
    longhorn_values_src: "{{ playbook_dir }}/../files/03-07-longhorn/values-lab.yaml"
    longhorn_values_dest: /root/longhorn-values-lab.yaml
    k3s_kubeconfig: /etc/rancher/k3s/k3s.yaml
  tasks:
    - name: Install helm package (Fedora/RHEL family)
      ansible.builtin.package:
        name: helm
        state: present
      ignore_errors: true
      register: helm_pkg

    - name: Hint if helm package install failed (install Helm 3 manually if needed)
      ansible.builtin.debug:
        msg: "dnf/yum 未装上 helm 时，请见 https://helm.sh/docs/intro/install/"
      when: helm_pkg.failed | default(false)

    - name: Fail if helm binary still unavailable
      ansible.builtin.command: which helm
      register: helm_which
      changed_when: false
      failed_when: helm_which.rc != 0

    - name: Copy lab values to server
      ansible.builtin.copy:
        src: "{{ longhorn_values_src }}"
        dest: "{{ longhorn_values_dest }}"
        mode: "0600"

    - name: Ensure longhorn-system namespace is not stuck Terminating (force finalize if needed)
      ansible.builtin.shell: |
        set -e
        export KUBECONFIG={{ k3s_kubeconfig }}
        ns="longhorn-system"
        phase="$(kubectl get ns "$ns" -o jsonpath='{.status.phase}' 2>/dev/null || true)"
        if [ "$phase" = "Terminating" ]; then
          echo "[WARN] namespace $ns is Terminating; force finalize to unblock install"
          kubectl get ns "$ns" -o json > /tmp/ns.json
          python3 -c "import json; obj=json.load(open('/tmp/ns.json')); obj.setdefault('spec',{}); obj['spec']['finalizers']=[]; json.dump(obj, open('/tmp/ns-finalize.json','w'))"
          kubectl replace --raw "/api/v1/namespaces/$ns/finalize" -f /tmp/ns-finalize.json >/dev/null
        fi
      args:
        executable: /bin/bash
      changed_when: true
      failed_when: false

    - name: Ensure longhorn Helm repo
      ansible.builtin.shell: |
        set -e
        if ! helm repo list 2>/dev/null | grep -q '^longhorn'; then
          helm repo add longhorn https://charts.longhorn.io
        fi
        helm repo update
      environment:
        KUBECONFIG: "{{ k3s_kubeconfig }}"
      args:
        executable: /bin/bash
      changed_when: true

    - name: Delete leftover longhorn PriorityClass (cluster-scoped) to avoid Helm ownership conflicts
      ansible.builtin.shell: |
        set -e
        KUBECONFIG={{ k3s_kubeconfig }} kubectl delete priorityclass longhorn-critical --ignore-not-found=true
      args:
        executable: /bin/bash
      changed_when: true
      failed_when: false

    - name: Delete leftover Longhorn CRDs (cluster-scoped) to avoid Helm ownership conflicts
      ansible.builtin.shell: |
        set -e
        export KUBECONFIG={{ k3s_kubeconfig }}
        crd_list="$(kubectl get crd -o name 2>/dev/null | grep 'longhorn.io' || true)"
        if [ -n "$crd_list" ]; then
          echo "$crd_list" | while read -r crd; do
            [ -z "$crd" ] && continue
            timeout 20s kubectl delete "$crd" --ignore-not-found=true || true
          done
        fi
      args:
        executable: /bin/bash
      changed_when: true
      failed_when: false

    - name: Delete leftover Longhorn ClusterRole/ClusterRoleBinding (cluster-scoped)
      ansible.builtin.shell: |
        set -e
        export KUBECONFIG={{ k3s_kubeconfig }}

        role_list="$(kubectl get clusterrole -o name 2>/dev/null | grep 'longhorn' || true)"
        if [ -n "$role_list" ]; then
          echo "$role_list" | while read -r role; do
            [ -z "$role" ] && continue
            timeout 20s kubectl delete "$role" --ignore-not-found=true || true
          done
        fi

        binding_list="$(kubectl get clusterrolebinding -o name 2>/dev/null | grep 'longhorn' || true)"
        if [ -n "$binding_list" ]; then
          echo "$binding_list" | while read -r binding; do
            [ -z "$binding" ] && continue
            timeout 20s kubectl delete "$binding" --ignore-not-found=true || true
          done
        fi
      args:
        executable: /bin/bash
      changed_when: true
      failed_when: false

    - name: Cleanup leftover Helm release records for Longhorn (default + longhorn-system)
      ansible.builtin.shell: |
        set -e
        export KUBECONFIG={{ k3s_kubeconfig }}

        # 有些失败/中断的安装会把 release secret 留在 default 或 longhorn-system，导致后续：
        # - "cannot re-use a name that is still in use"
        # - cluster-scoped 资源的 meta.helm.sh/release-namespace 注解冲突
        for ns in longhorn-system default; do
          if helm -n "$ns" list --all 2>/dev/null | grep -q '^longhorn'; then
            # uninstall 可能卡住（例如 uninstall job / hook），避免阻塞整个自动化流程
            timeout 120s helm -n "$ns" uninstall longhorn --no-hooks || true
          fi

          sec_list="$(kubectl -n "$ns" get secret -o name 2>/dev/null | grep '^secret/sh\\.helm\\.release\\.v1\\.longhorn\\.' || true)"
          if [ -n "$sec_list" ]; then
            echo "$sec_list" | xargs -n1 kubectl -n "$ns" delete --ignore-not-found=true
          fi
        done
      environment:
        KUBECONFIG: "{{ k3s_kubeconfig }}"
      args:
        executable: /bin/bash
      changed_when: true
      failed_when: false

    - name: Helm upgrade/install Longhorn（失败兜底：install --replace）
      ansible.builtin.shell: |
        set -e
        helm upgrade --install longhorn longhorn/longhorn --namespace longhorn-system --create-namespace -f {{ longhorn_values_dest }} --version {{ longhorn_chart_version }} --wait --timeout 15m || helm install --replace longhorn longhorn/longhorn --namespace longhorn-system --create-namespace -f {{ longhorn_values_dest }} --version {{ longhorn_chart_version }} --wait --timeout 15m
      environment:
        KUBECONFIG: "{{ k3s_kubeconfig }}"
      args:
        executable: /bin/bash
      register: helm_longhorn
      changed_when: true

- name: Apply local-path-config lab defaults (optional)
  hosts: k3s_server
  become: true
  run_once: true
  vars:
    k3s_kubeconfig: /etc/rancher/k3s/k3s.yaml
    local_path_json_src: "{{ playbook_dir }}/../files/03-05-local-path-config/local-path-config-lab.json"
    local_path_json_dest: /root/local-path-config-lab.json
  tasks:
    - name: Apply local-path-config lab defaults (optional)
      when: longhorn_apply_local_path_lab | default(false) | bool
      block:
        - name: Copy local-path lab json
          ansible.builtin.copy:
            src: "{{ local_path_json_src }}"
            dest: "{{ local_path_json_dest }}"
            mode: "0644"

        - name: Apply local-path-config ConfigMap
          ansible.builtin.shell: |
            set -e
            KUBECONFIG={{ k3s_kubeconfig }} kubectl -n kube-system create configmap local-path-config \
              --from-file=config.json={{ local_path_json_dest }} \
              --dry-run=client -o yaml | KUBECONFIG={{ k3s_kubeconfig }} kubectl apply -f -
          args:
            executable: /bin/bash
          changed_when: true

        - name: Restart local-path-provisioner if present
          ansible.builtin.shell: |
            KUBECONFIG={{ k3s_kubeconfig }} kubectl -n kube-system rollout restart deploy/local-path-provisioner
          args:
            executable: /bin/bash
          register: lp_restart
          failed_when: false
          changed_when: lp_restart.rc == 0