From ce60327b609f9967d6dab28de1bf17afc01bb0bc Mon Sep 17 00:00:00 2001
From: Mark Goddard <mark@stackhpc.com>
Date: Thu, 12 Oct 2017 11:19:23 +0100
Subject: [PATCH] Fix SSH reconnection for docker role in Ansible 2.3

In ansible 2.3, the path to SSH persistent connection sockets was changed to use a SHA1
hash of the connection parameters, so the workaround to force a reconnection was broken.
This change removes all ControlPath sockets from the ansible control host.
---
 ansible/roles/docker/tasks/main.yml | 33 ++++++++++++++++++++++++++++-
 1 file changed, 32 insertions(+), 1 deletion(-)

diff --git a/ansible/roles/docker/tasks/main.yml b/ansible/roles/docker/tasks/main.yml
index ce611fde9..9b5f3244f 100644
--- a/ansible/roles/docker/tasks/main.yml
+++ b/ansible/roles/docker/tasks/main.yml
@@ -17,6 +17,9 @@
 # After adding the user to the docker group, we need to log out and in again to
 # pick up the group membership. We do this by removing the SSH ControlPersist
 # connection.
+
+# NOTE: This method does not work in Ansible 2.3, which uses a SHA1 hash of the
+# connection parameters to determine the control persist socket path.
 - name: Drop the persistent SSH connection to activate the new group membership
   local_action:
     module: shell ssh -O stop {{ cp_hostname }} -o ControlPath={{ cp_path }}
@@ -26,13 +29,41 @@
     - "'No such file or directory' not in socket_removal.stderr"
   with_items: "{{ play_hosts }}"
   run_once: True
-  when: group_result|changed
+  when:
+    - group_result|changed
+    - ansible_version | version_compare('2.3', 'lt')
   vars:
     cp_hostname: "{{ hostvars[item].ansible_host|default(inventory_hostname) }}"
     cp_username: "{{ hostvars[item].ansible_user }}"
     cp_port: "{{ hostvars[item].ansible_ssh_port|default('22') }}"
     cp_path: "~/.ansible/cp/ansible-ssh-{{ cp_hostname }}-{{ cp_port }}-{{ cp_username }}"
 
+# NOTE: For Ansible 2.3+, ideally we'd use a meta task with the
+# reset_connection option but due to
+# https://github.com/ansible/ansible/issues/27520 this does not work (checked
+# in Ansible 2.3.2.0). Instead, we use the heavy handed method of removing all
+# ansible control sockets. Limitation: if this user is running another ansible
+# process, we will kill its connections.
+- name: Find persistent SSH connection control sockets
+  local_action:
+    module: find
+    file_type: any
+    path: "~/.ansible/cp/"
+    patterns: '[a-f0-9]{10}'
+    use_regex: True
+  register: cp_sockets
+  run_once: True
+  when:
+    - group_result|changed
+    - ansible_version | version_compare('2.3', 'ge')
+
+- name: Drop all persistent SSH connections to activate the new group membership
+  local_action:
+    module: shell ssh -O stop None -o ControlPath={{ item.path }}
+  with_items: "{{ cp_sockets.files }}"
+  run_once: True
+  when: not cp_sockets|skipped
+
 - name: Ensure Docker daemon is started
   service:
     name: docker