Merge pull request #1333 from wanyaoqi/fix/no-reboot-on-controller-node

fix: no reboot on controller node

lib/install.py

@@ -93,7 +93,10 @@ def start(config_file):
     except AttributeError:
         pass
     vars = config.ansible_global_vars()
-    vars['no_reboot'] = 'false' if need_reboot(ip, inside=True) else 'true'
+    # 240722 wanyaoqi: no_reboot has been removed before this change
+    # vars['no_reboot'] = 'false' if need_reboot(ip, inside=True) else 'true'
+    vars['is_controller_node'] = 'true' if config.is_controller_node() else 'false'
+    print("vars: ", vars)
     return_code = cmd.run_ansible_playbook(
         inventory_f,
         './onecloud/install-cluster.yml',

lib/ocboot.py

@@ -101,6 +101,13 @@ def get_primary_master_ssh_port(self):
             return
         return self.primary_master_config.node.port
 
+    def is_controller_node(self):
+        if self.primary_master_config:
+            return True
+        if self.master_config:
+            return True
+        return False
+
     def _fetch_conf(self, config_cls):
         group = config_cls.get_group()
         group_config = self.config.get(group, None)

onecloud/roles/common/tasks/utils/unified_cgroup_hierarchy.yml

@@ -21,6 +21,11 @@
   - name: Update GRUB
     command: update-grub
 
+  - name: Check is need manual reboot to disable unified cgroup hierarchy
+    fail:
+      msg: "GRUB updated, Please reboot to disable cgroup hierarchy and run install script again"
+    when: is_controller_node|default(false)|bool == true
+
   - name: "Reboot the system because GRUB updated. You can rerun this script when system running"
     ansible.builtin.reboot:
 

onecloud/roles/utils/gpu-init/files/gpu_setup.sh

@@ -285,7 +285,7 @@ env_check() {
         error_exit "You need sudo or root to run this script."
     fi
 
-    local supported_distros=("centos" "debian" "openeuler", "ubuntu")
+    local supported_distros=("centos" "debian" "openeuler" "ubuntu")
     local distros=($(get_distro))
 
     local found_supported_distro=false

onecloud/roles/utils/kernel-check/files/hugetlb_setup.sh

@@ -139,7 +139,7 @@ env_check() {
         error_exit "You need sudo or root to run this script."
     fi
 
-    local supported_distros=("centos" "debian" "openeuler", "ubuntu")
+    local supported_distros=("centos" "debian" "openeuler" "ubuntu")
     local distros=($(get_distro))
 
     local found_supported_distro=false

onecloud/roles/utils/kernel-check/tasks/main.yml

@@ -89,6 +89,7 @@
   when:
     - is_centos_x86 | default(false) | bool  == true
     - is_yunion_kernel_running.rc != 0
+    - is_controller_node is not defined or is_controller_node|default(false)|bool == false
     - ansible_connection == "ssh"
 
 - name: Local Reboot system if not cloud kernel, it should take a few minutes...
@@ -97,6 +98,7 @@
   when:
     - is_centos_x86 | default(false) | bool  == true
     - is_yunion_kernel_running.rc != 0
+    - is_controller_node is not defined or is_controller_node|default(false)|bool == false
     - ansible_connection == "local"
 
 - name: SSH Reboot system if hugetlb conf is updated
@@ -106,11 +108,13 @@
   become: yes
   when:
   - hugetlb_is_ready == false
+  - is_controller_node is not defined or is_controller_node|default(false)|bool == false
   - ansible_connection == "ssh"
 
 - name: Local Reboot system if hugetlb conf is updated
   command: reboot
   become: yes
   when:
   - hugetlb_is_ready == false
+  - is_controller_node is not defined or is_controller_node|default(false)|bool == false
   - ansible_connection == "local"