[Cherry-pick] fix installation compatibility of ubuntu 20.04 (#5257)

microsoft · Jan 18, 2021 · 1c290c9 · 1c290c9
1 parent 6fab928
commit 1c290c9
Show file tree

Hide file tree

Showing 13 changed files with 80 additions and 7 deletions.
diff --git a/contrib/kubespray/environment-check.yml b/contrib/kubespray/environment-check.yml
@@ -1,4 +1,9 @@
 ---
+- hosts: all,localhost
+  gather_facts: false
+  roles:
+    - { role: set-python3 }
+
 - hosts: all
   gather_facts: false
   tasks:

diff --git a/contrib/kubespray/pre-installation.yml b/contrib/kubespray/pre-installation.yml
@@ -0,0 +1,11 @@
+---
+- hosts: all,localhost
+  gather_facts: false
+  roles:
+    - { role: set-python3 }
+
+- hosts: all
+  become: true
+  become_user: root
+  roles:
+    - { role: pre-installation }
diff --git a/contrib/kubespray/quick-start-kubespray.sh b/contrib/kubespray/quick-start-kubespray.sh
@@ -35,5 +35,8 @@ ansible all -i ${HOME}/pai-deploy/cluster-cfg/hosts.yml -m ping || exit $?
 echo "Performing pre-check..."
 ansible-playbook -i ${HOME}/pai-pre-check/pre-check.yml set-host-daemon-port-range.yml -e "@${CLUSTER_CONFIG}" || exit $?
 
+echo "Performing pre-installation..."
+ansible-playbook -i ${HOME}/pai-deploy/cluster-cfg/hosts.yml pre-installation.yml || exit $?
+
 echo "Starting kubernetes..."
 /bin/bash script/kubernetes-boot.sh || exit $?
diff --git a/contrib/kubespray/roles/pre-installation/defaults/main.yml b/contrib/kubespray/roles/pre-installation/defaults/main.yml
diff --git a/contrib/kubespray/roles/pre-installation/tasks/main.yml b/contrib/kubespray/roles/pre-installation/tasks/main.yml
@@ -0,0 +1,21 @@
+---
+# If sshpass is not installed on target machine, 
+# kubespray 2.11 may have these issues:
+# https://github.com/kubernetes-sigs/kubespray/issues/6906
+# https://github.com/kubernetes-sigs/kubespray/issues/5693
+# This is resolved in PR https://github.com/kubernetes-sigs/kubespray/pull/6907
+- name: "Install sshpass"
+  apt:
+    name: sshpass
+    state: present
+
+
+# workaround for ubuntu 20.04
+# This is resolved in https://github.com/kubernetes-sigs/kubespray/pull/6157
+- name: "Install Python"
+  apt:
+    name: python
+    state: present
+  when:
+  - ansible_distribution_version is version('20.04', '=')
+
diff --git a/contrib/kubespray/roles/set-python3/defaults/main.yml b/contrib/kubespray/roles/set-python3/defaults/main.yml
diff --git a/contrib/kubespray/roles/set-python3/tasks/main.yml b/contrib/kubespray/roles/set-python3/tasks/main.yml
@@ -0,0 +1,13 @@
+---
+# Ubuntu 20.04 doesn't have python2 pre-installed,
+# but Ubuntu 16.04, 18.04, and 20.04 all have python3 pre-installed.
+# So we use python3 as the default interpreter.
+# Also, we use `sudo python3` to execute python on dev box machine.
+# It is better to align with it here to prevent `ansible-playbook` from failing into another python.
+- name: set ansible_python_interpreter to python3
+  set_fact:
+    ansible_python_interpreter: "/usr/bin/python3"
+
+- name: output ansible_python_interpreter
+  debug:
+    var: ansible_python_interpreter
diff --git a/contrib/kubespray/script/environment.sh b/contrib/kubespray/script/environment.sh
@@ -40,3 +40,10 @@ sudo apt-get -y install sshpass
 
 echo "Install kubespray's requirements and ansible is included"
 sudo python3 -m pip install -r ${HOME}/pai-deploy/kubespray/requirements.txt
+
+# ansible 2.7 doesn't support distribution info collection on Ubuntu 20.04
+# Use ansible 2.9.7 as a workaround
+# Reference: https://stackoverflow.com/questions/61460151/ansible-not-reporting-distribution-info-on-ubuntu-20-04
+# We can upgrade kubespray version to avoid this issue in the future.
+sudo python3 -m pip install ansible==2.9.7
+
diff --git a/contrib/kubespray/script/kubernetes-boot.sh b/contrib/kubespray/script/kubernetes-boot.sh
@@ -2,7 +2,10 @@
 
 echo "setup k8s cluster"
 cd ${HOME}/pai-deploy/kubespray
-ansible-playbook -i inventory/pai/hosts.yml cluster.yml --become --become-user=root -e "@inventory/pai/openpai.yml" || exit $?
+# Ubuntu 20.04 doesn't have python2 pre-installed,
+# but Ubuntu 16.04, 18.04, and 20.04 all have python3 pre-installed.
+# So we use python3 as the default interpreter.
+ansible-playbook -i inventory/pai/hosts.yml -e "ansible_python_interpreter=/usr/bin/python3" cluster.yml --become --become-user=root -e "@inventory/pai/openpai.yml" || exit $?
 
 sudo mkdir -p ${HOME}/pai-deploy/kube || exit $?
 sudo cp -rf ${HOME}/pai-deploy/kubespray/inventory/pai/artifacts/admin.conf ${HOME}/pai-deploy/kube/config || exit $?

diff --git a/contrib/kubespray/set-host-daemon-port-range.yml b/contrib/kubespray/set-host-daemon-port-range.yml
@@ -1,4 +1,9 @@
 ---
+- hosts: all,localhost
+  gather_facts: false
+  roles:
+    - { role: set-python3 }
+
 - hosts: gpu
   become: true
   become_user: root

diff --git a/contrib/kubespray/set-kubectl.yml b/contrib/kubespray/set-kubectl.yml
@@ -1,4 +1,9 @@
 ---
+- hosts: all,localhost
+  gather_facts: false
+  roles:
+    - { role: set-python3 }
+
 - hosts: localhost
   gather_facts: false
   tasks:

diff --git a/docs/manual/cluster-admin/installation-guide.md b/docs/manual/cluster-admin/installation-guide.md
@@ -34,7 +34,7 @@ We recommend you to use CPU-only machines for dev box and master. The detailed r
     </td>
     <td>
       <ul>
-        <li>Ubuntu 16.04 (18.04 should work, but not fully tested)</li>
+        <li>Ubuntu 16.04 (18.04, 20.04 should work, but not fully tested)</li>
         <li>SSH service is enabled.</li>
         <li>Passwordless ssh to all other machines (master and worker machines).</li>
         <li>Docker is installed.</li>
@@ -52,7 +52,7 @@ We recommend you to use CPU-only machines for dev box and master. The detailed r
     </td>
     <td>
       <ul>
-        <li>Ubuntu 16.04 (18.04 should work, but not fully tested)</li>
+        <li>Ubuntu 16.04 (18.04, 20.04 should work, but not fully tested)</li>
         <li>SSH service is enabled. </li>
         <li>It shares the same username/password with worker machines, and have sudo privilege.</li>
         <li>Docker is installed.</li>
@@ -100,7 +100,7 @@ Please check the following requirements for different types of worker machines:
     </td>
     <td>
       <ul>
-        <li>Ubuntu 16.04 (18.04 should work, but not fully tested)</li>
+        <li>Ubuntu 16.04 (18.04, 20.04 should work, but not fully tested)</li>
         <li>SSH service is enabled. </li>
         <li>It shares the same username/password with all other machines, and have sudo privilege.</li>
         <li>Docker is installed.</li>

diff --git a/docs_zh_CN/manual/cluster-admin/installation-guide.md b/docs_zh_CN/manual/cluster-admin/installation-guide.md
@@ -34,7 +34,7 @@ master机器用于运行核心Kubernetes组件和核心OpenPAI服务。目前，
     </td>
     <td>
       <ul>
-        <li>Ubuntu 16.04 (18.04应该可用，但没有经过完整测试)</li>
+        <li>Ubuntu 16.04 (18.04、20.04应该可用，但没有经过完整测试)</li>
         <li>SSH服务已开启。</li>
         <li>可以免密登录所有master和worker机器。</li>
         <li>Docker已被正确安装。</li>
@@ -52,7 +52,7 @@ master机器用于运行核心Kubernetes组件和核心OpenPAI服务。目前，
     </td>
     <td>
       <ul>
-        <li>Ubuntu 16.04 (18.04应该可用，但没有经过完整测试)</li>
+        <li>Ubuntu 16.04 (18.04、20.04应该可用，但没有经过完整测试)</li>
         <li>SSH服务已开启。</li>
         <li>和所有worker机器有同样的SSH用户名和密码，且该SSH用户有sudo权限。</li>
         <li>Docker已被正确安装。</li>
@@ -100,7 +100,7 @@ hivedscheduler是OpenPAI的默认调度器，它支持虚拟集群划分，拓
     </td>
     <td>
       <ul>
-        <li>Ubuntu 16.04 (18.04应该可用，但没有经过完整测试)</li>
+        <li>Ubuntu 16.04 (18.04、20.04应该可用，但没有经过完整测试)</li>
         <li>SSH服务已开启。 </li>
         <li>所有master和worker机器有同样的SSH用户名和密码，且该SSH用户有sudo权限。</li>
         <li>Docker已被正确安装。</li>