Bring first etcd server up before others.

In faster environments (i.e. all local VMs) etcd nodes could come online at
roughly the same time leading to conflicts with self-elections, resulting in a
non-functional cluster.

To solve we configure the first etcd host by itself, then configure the
remaining ones in parallel to keep things as fast as possible.

1 files changed, 26 insertions, 2 deletions

diff --git a/playbooks/common/openshift-etcd/config.yml b/playbooks/common/openshift-etcd/config.yml
index 7d94ced2e..3d7a884e8 100644
--- a/playbooks/common/openshift-etcd/config.yml
+++ b/playbooks/common/openshift-etcd/config.yml
@@ -68,8 +68,32 @@
       validate_checksum: yes
     with_items: etcd_needing_server_certs
 
-- name: Configure etcd hosts
-  hosts: oo_etcd_to_config
+# Configure a first etcd host to avoid conflicts in choosing a leader
+# if other members come online too quickly.
+- name: Configure first etcd host
+  hosts: oo_first_etcd
+  vars:
+    sync_tmpdir: "{{ hostvars.localhost.g_etcd_mktemp.stdout }}"
+    etcd_url_scheme: https
+    etcd_peer_url_scheme: https
+    etcd_peers_group: oo_etcd_to_config
+  pre_tasks:
+  - name: Ensure certificate directory exists
+    file:
+      path: "{{ etcd_cert_config_dir }}"
+      state: directory
+  - name: Unarchive the tarball on the etcd host
+    unarchive:
+      src: "{{ sync_tmpdir }}/{{ etcd_cert_subdir }}.tgz"
+      dest: "{{ etcd_cert_config_dir }}"
+    when: etcd_server_certs_missing
+  roles:
+  - etcd
+  - role: nickhammond.logrotate
+
+# Configure the remaining etcd hosts, skipping the first one we dealt with above.
+- name: Configure remaining etcd hosts
+  hosts: oo_etcd_to_config:!oo_first_etcd
   vars:
     sync_tmpdir: "{{ hostvars.localhost.g_etcd_mktemp.stdout }}"
     etcd_url_scheme: https