kubespray issue fix (#4537)

contrib/kubespray/requirement.sh

@@ -35,6 +35,8 @@ then
   echo "Pass: Cluster meets the requirements"
 else
   echo "Faild: Please check the output, and modify the cluster setting to meet the requirement"
+  rm -rf ${HOME}/pai-pre-check/
+  exit $ret_code_check
 fi
 
 rm -rf ${HOME}/pai-pre-check/

contrib/kubespray/roles/requirement/devbox/tasks/ubuntu.yml

@@ -37,5 +37,39 @@
   when:
     - dev_box_docker_version_replace is version('1.10.0', '<=')
 
+- name: "Dev-box 3.1 Check whether the vm can raw.githubusercontent.com"
+  raw: curl https://raw.githubusercontent.com/microsoft/pai/master/README.md
+  register: devbox_curl_githubusercontent
+  failed_when: false
+  changed_when: false
+  check_mode: false
+  environment: {}
+
+- name: "Dev-box 3.2 Check whether the vm can access to raw.githubusercontent.com"
+  fail:
+    msg: "Unable to access raw.githubusercontent.com"
+  changed_when: false
+  check_mode: false
+  environment: {}
+  when:
+    - devbox_curl_githubusercontent.rc != 0
+
+- name: "Dev-box 4.1 Check whether the vm can access to docker.io"
+  raw: nslookup index.docker.io
+  register: devbox_nslookup_docker_io
+  failed_when: false
+  changed_when: false
+  check_mode: false
+  environment: {}
+
+- name: "Dev-box 4.2 Check whether the vm can access to docker.io"
+  fail:
+    msg: "Unable to access docker.io"
+  changed_when: false
+  check_mode: false
+  environment: {}
+  when:
+    - devbox_nslookup_docker_io.rc != 0
+
 
 

contrib/kubespray/roles/requirement/infra/tasks/resource.yml

@@ -3,7 +3,7 @@
   fail:
     msg: "OpenPAI's infra node should have 40 Gi free memory for service "
   when:
-    - ansible_memfree_mb < 40000
+    - ansible_memory_mb["nocache"]["free"] < 40000
 
 - name: Check cpu resource requirement for openpai service
   fail:

contrib/kubespray/roles/requirement/infra/tasks/ubuntu.yml

@@ -47,6 +47,40 @@
   when:
     - ansible_control_host_address == ansible_default_ipv4.address
 
+- name: "Infra 4.1 Check whether the vm can access to gcr.io"
+  raw: ping gcr.io -c 3
+  register: infra_ping_gcr_io
+  failed_when: false
+  changed_when: false
+  check_mode: false
+  environment: {}
+
+- name: "Infra 4.2 Check whether the vm can access to gcr.io"
+  fail:
+    msg: "Unable to access gcr.io"
+  changed_when: false
+  check_mode: false
+  environment: {}
+  when:
+    - infra_ping_gcr_io.rc != 0
+
+- name: "Infra 5.1 Check whether the vm can access to quay.io"
+  raw: nslookup quay.io
+  register: infra_nslookup_quay_io
+  failed_when: false
+  changed_when: false
+  check_mode: false
+  environment: {}
+
+- name: "Infra 5.2 Check whether the vm can access to quay.io"
+  fail:
+    msg: "Unable to access quay.io"
+  changed_when: false
+  check_mode: false
+  environment: {}
+  when:
+    - infra_nslookup_quay_io.rc != 0
+
 - name: check infra machine's resouce requirement
   include_tasks: resource.yml
   when: resource_check

contrib/kubespray/roles/requirement/worker/tasks/resource.yml

@@ -3,4 +3,4 @@
   fail:
     msg: "OpenPAI's worker node should have 16 Gi free memory for service "
   when:
-    - ansible_memfree_mb < 16000
+    - ansible_memory_mb["nocache"]["free"] < 16000

contrib/kubespray/roles/requirement/worker/tasks/ubuntu.yml

@@ -22,6 +22,40 @@
   when:
     - ansible_control_host_address == ansible_default_ipv4.address
 
+- name: "Worker 3.1 Check whether the vm can access to gcr.io"
+  raw: ping gcr.io -c 3
+  register: worker_ping_gcr_io
+  failed_when: false
+  changed_when: false
+  check_mode: false
+  environment: {}
+
+- name: "Worker 3.2 Check whether the vm can access to gcr.io"
+  fail:
+    msg: "Unable to access gcr.io"
+  changed_when: false
+  check_mode: false
+  environment: {}
+  when:
+    - worker_ping_gcr_io.rc != 0
+
+- name: "Worker 4.1 Check whether the vm can access to quay.io"
+  raw: nslookup quay.io
+  register: worker_nslookup_quay_io
+  failed_when: false
+  changed_when: false
+  check_mode: false
+  environment: {}
+
+- name: "Worker 4.2 Check whether the vm can access to quay.io"
+  fail:
+    msg: "Unable to access quay.io"
+  changed_when: false
+  check_mode: false
+  environment: {}
+  when:
+    - worker_nslookup_quay_io.rc != 0
+
 - name: check worker machines' NVIDIA gpu card, drivers and runtime
   include_tasks: ubuntu-nvidia.yml
   when: gpu_type == 'nvidia'

contrib/kubespray/script/k8s-generator.py

@@ -92,7 +92,7 @@ def main():
     if 'openpai_kube_network_plugin' not in cluster_config or cluster_config['openpai_kube_network_plugin'] != 'weave':
         count_input = 0
         while True:
-            user_input = raw_input("Are your cluster is in Azure cloud or not? (Y/N) (case sensitive)")
+            user_input = input("Are your cluster is in Azure cloud or not? (Y/N) (case sensitive)")
             if user_input == "N":
                 break
             elif user_input == "Y":
@@ -103,7 +103,7 @@ def main():
             if count_input == 3:
                 logger.warning("3 Times.........  Sorry,  we will force stopping your operation.")
                 sys.exit(1)
-        if user_input == "Y" and cluster_config['openpai_kube_network_plugin'] == 'calico':
+        if user_input == "Y" and ('openpai_kube_network_plugin' not in cluster_config or cluster_config['openpai_kube_network_plugin'] == 'calico'):
             logger.warning("Azure does not support calico, please change the openpai_kube_network_plugin to weave")
             logger.warning("https://docs.projectcalico.org/reference/public-cloud/azure#why-doesnt-azure-support-calico-networking")
             sys.exit(1)