From 6fdcaa1a634aa00dd22766d36290f7dcce4115b2 Mon Sep 17 00:00:00 2001
From: Bogdan Dobrelya <bdobrelia@mirantis.com>
Date: Tue, 13 Sep 2016 14:36:01 +0200
Subject: [PATCH 1/2] Add retries for copying binaries from containers

Closes issue: https://github.com/kubespray/kargo/issues/479

Signed-off-by: Bogdan Dobrelya <bdobrelia@mirantis.com>
---
 roles/etcd/tasks/install.yml               | 4 ++++
 roles/kubernetes/master/tasks/main.yml     | 5 ++++-
 roles/network_plugin/calico/tasks/main.yml | 4 ++++
 3 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/roles/etcd/tasks/install.yml b/roles/etcd/tasks/install.yml
index 26271943a..959133c29 100644
--- a/roles/etcd/tasks/install.yml
+++ b/roles/etcd/tasks/install.yml
@@ -17,6 +17,10 @@
            /usr/bin/docker cp etcdctl-binarycopy:{{ etcd_container_bin_dir }}etcdctl {{ bin_dir }}/etcdctl &&
            /usr/bin/docker rm -f etcdctl-binarycopy"
   when: etcd_deployment_type == "docker"
+  register: etcd_task_result
+  until: etcd_task_result.rc == 0
+  retries: 4
+  delay: "{{ 20 | random + 3 }}"
   changed_when: false
 
 #Plan B: looks nicer, but requires docker-py on all hosts:
diff --git a/roles/kubernetes/master/tasks/main.yml b/roles/kubernetes/master/tasks/main.yml
index 4d0933550..e8dfe08fc 100644
--- a/roles/kubernetes/master/tasks/main.yml
+++ b/roles/kubernetes/master/tasks/main.yml
@@ -9,8 +9,11 @@
 
 - name: Copy kubectl from hyperkube container
   command: "/usr/bin/docker run --rm -v {{ bin_dir }}:/systembindir {{ hyperkube_image_repo }}:{{ hyperkube_image_tag }} /bin/cp /hyperkube /systembindir/kubectl"
+  register: kube_task_result
+  until: kube_task_result.rc == 0
+  retries: 4
+  delay: "{{ 20 | random + 3 }}"
   changed_when: false
-  register: kubectl_launcher
 
 - name: Write kube-apiserver manifest
   template:
diff --git a/roles/network_plugin/calico/tasks/main.yml b/roles/network_plugin/calico/tasks/main.yml
index abe0f690d..09c3b47d5 100644
--- a/roles/network_plugin/calico/tasks/main.yml
+++ b/roles/network_plugin/calico/tasks/main.yml
@@ -34,6 +34,10 @@
 
 - name: Calico | Copy cni plugins from hyperkube
   command: "/usr/bin/docker run --rm -v /opt/cni/bin:/cnibindir {{ hyperkube_image_repo }}:{{ hyperkube_image_tag }} /bin/cp -r /opt/cni/bin/. /cnibindir/"
+  register: cni_task_result
+  until: cni_task_result.rc == 0
+  retries: 4
+  delay: "{{ 20 | random + 3 }}"
   changed_when: false
   when: use_hyperkube_cni
 

From 783871a2539392a5cd0cb0deb93c05ca8a502252 Mon Sep 17 00:00:00 2001
From: Bogdan Dobrelya <bdobrelia@mirantis.com>
Date: Tue, 13 Sep 2016 15:29:22 +0200
Subject: [PATCH 2/2] Add retries for packages installation

Signed-off-by: Bogdan Dobrelya <bdobrelia@mirantis.com>
---
 roles/docker/tasks/main.yml                | 8 ++++++++
 roles/kubernetes/preinstall/tasks/main.yml | 4 ++++
 2 files changed, 12 insertions(+)

diff --git a/roles/docker/tasks/main.yml b/roles/docker/tasks/main.yml
index 6d71c4980..826e16978 100644
--- a/roles/docker/tasks/main.yml
+++ b/roles/docker/tasks/main.yml
@@ -27,6 +27,10 @@
     id: "{{item}}"
     keyserver: "{{docker_repo_key_info.keyserver}}"
     state: present
+  register: keyserver_task_result
+  until: keyserver_task_result|success
+  retries: 4
+  delay: "{{ 20 | random + 3 }}"
   with_items: "{{ docker_repo_key_info.repo_keys }}"
   when: ansible_os_family != "CoreOS"
 
@@ -51,6 +55,10 @@
     pkg: "{{item.name}}"
     force: "{{item.force|default(omit)}}"
     state: present
+  register: docker_task_result
+  until: docker_task_result|success
+  retries: 4
+  delay: "{{ 20 | random + 3 }}"
   with_items: "{{ docker_package_info.pkgs }}"
   when: (ansible_os_family != "CoreOS") and (docker_package_info.pkgs|length > 0)
 
diff --git a/roles/kubernetes/preinstall/tasks/main.yml b/roles/kubernetes/preinstall/tasks/main.yml
index ec1bd64c0..8c22b73bf 100644
--- a/roles/kubernetes/preinstall/tasks/main.yml
+++ b/roles/kubernetes/preinstall/tasks/main.yml
@@ -101,6 +101,10 @@
     module: "{{ ansible_pkg_mgr }}"
     name: "{{ item }}"
     state: latest
+  register: pkgs_task_result
+  until: pkgs_task_result|success
+  retries: 4
+  delay: "{{ 20 | random + 3 }}"
   with_items: "{{required_pkgs | default([]) | union(common_required_pkgs|default([]))}}"
   when: ansible_os_family != "CoreOS"