From ce0fcb8482f7ea055f1c9c12c741d4b64dd085fb Mon Sep 17 00:00:00 2001
From: noelmcloughlin <noel.mcloughlin@gmail.com>
Date: Sat, 5 Jun 2021 00:14:02 +0100
Subject: [PATCH] feat(cluster): join state with erlang_cookie

---
 docs/README.rst                     | 10 ++++++++--
 pillar.example                      | 18 +++++++++--------
 rabbitmq/config/cluster/install.sls | 31 ++++++++++++++++++++++-------
 rabbitmq/config/init.sls            |  2 +-
 rabbitmq/package/repo/clean.sls     |  1 -
 rabbitmq/service/running.sls        | 11 ++--------
 6 files changed, 45 insertions(+), 28 deletions(-)

diff --git a/docs/README.rst b/docs/README.rst
index 5aaeb517..0a7054ec 100644
--- a/docs/README.rst
+++ b/docs/README.rst
@@ -67,7 +67,7 @@ now ``pre-commit`` will run automatically on each ``git commit``. ::
 Special notes
 -------------
 
-None
+The main state ``rabbitmq`` excludes ``rabbitmq.config.cluster`` (initial cluster setup) state.
 
 Available states
 ----------------
@@ -99,7 +99,13 @@ This state will install the rabbitmqadmin package only.
 ^^^^^^^^^^^^^^^^^^^
 
 This state will configure the rabbitmq service and has a dependency on ``rabbitmq.install``
-via include list. See ``pillar.example``.
+via include list. It excludes ``rabbitmq.config.cluster`` state
+
+``rabbitmq.config.cluster``
+^^^^^^^^^^^^^^^^^^^^^^^^^
+
+For initial setup this state writes the erlang cookie, joins cluster, and restarts service.
+The erlang cookie comes from pillar data and must the identical for all cluster members
 
 ``rabbitmq.service``
 ^^^^^^^^^^^^^^^^^^^^
diff --git a/pillar.example b/pillar.example
index f4d56ad9..f9ab7fa0 100644
--- a/pillar.example
+++ b/pillar.example
@@ -2,13 +2,22 @@
 # vim: ft=yaml
 ---
 rabbitmq:
+  cluster:
+    rabbitmq@locahost:
+      user: rabbitmq
+      host: rabbitmq  # short hostname of node to join to, not fqdn
+      ram_node: None
+      runas: rabbitmq
+      erlang_cookie:
+        name: /var/lib/rabbitmq/.erlang_cookie
+        value: shared-value-for-all-cluster-members
+
   pkg:
     # https://github.com/rabbitmq/rabbitmq-server/releases/tag/v3.8.14
     use_upstream: repo  # if available (i.e. packagecloud)
   service:
     enabled: true
     running: true
-
   config:
     # see rabbitmq.conf.example from rabbitmq-server github docs
     context: {}
@@ -59,13 +68,6 @@ rabbitmq:
           - 'alternate-**exchange': 'amq.fanout'
           - 'test-header': 'testing'
 
-  cluster:
-    rabbit@locahost:
-      - user: rabbit
-      - host: localhost
-      - ram_node: None
-      - runas: root
-
   plugin:
     rabbitmq_management:
       runas: root
diff --git a/rabbitmq/config/cluster/install.sls b/rabbitmq/config/cluster/install.sls
index a084c3cf..701c36d8 100644
--- a/rabbitmq/config/cluster/install.sls
+++ b/rabbitmq/config/cluster/install.sls
@@ -10,15 +10,32 @@ include:
   - {{ sls_service_running }}
   - {{ sls_config_user }}
 
-    {% for name, cluster in salt["pillar.get"]("rabbitmq:cluster", {}).items() %}
+    {%- for name, cluster in salt["pillar.get"]("rabbitmq:cluster", {}).items() %}
+        {%- if 'erlang_cookie' in cluster and cluster.erlang_cookie is mapping %}
 
 rabbitmq-config-cluster-join-{{ name }}:
-  rabbitmq_cluster.join:
-    {% for value in cluster %}
-    - {{ value | json }}
-    {% endfor %}
+  file.managed:
+    - name: {{ cluster.erlang_cookie.name }}
+    - contents: {{ cluster.erlang_cookie.value }}
+    - mode: 400
+    - user: {{ cluster.user }}
+    - group: root
+    - makedirs: True
+    - watch_in:
+      - service: rabbitmq-service-running-service-running
+
+            {%- if 'host' in grains and grains.host not in cluster.host %}
+
+  rabbitmq_cluster.joined:
+    - user: {{ cluster.user }}
+    - host: {{ cluster.host }}
+    - ram_node: {{ cluster.ram_node }}
+    - runas: {{ cluster.runas }}
     - require:
-      - service: {{ rabbitmq.service.name }}
+      - file: rabbitmq-config-cluster-join-{{ name }}
       - sls: {{ sls_config_user }}
+      - service: rabbitmq-service-running-service-running
 
-    {% endfor %}
+            {%- endif %}
+        {%- endif %}
+    {%- endfor %}
diff --git a/rabbitmq/config/init.sls b/rabbitmq/config/init.sls
index 84e1cd59..ab1e7d9c 100644
--- a/rabbitmq/config/init.sls
+++ b/rabbitmq/config/init.sls
@@ -8,4 +8,4 @@ include:
   - .policy
   - .plugin
   - .upstream
-  # .cluster
+  # .cluster should be run once, not always
diff --git a/rabbitmq/package/repo/clean.sls b/rabbitmq/package/repo/clean.sls
index d5ce2970..deb16235 100644
--- a/rabbitmq/package/repo/clean.sls
+++ b/rabbitmq/package/repo/clean.sls
@@ -15,7 +15,6 @@ rabbitmq-package-repo-erlang-clean:
   pkgrepo.absent:
     - names:
       - rabbitmq-rabbitmq-erlang
-      - rabbitmq_rabbitmq-erlang
       - rabbitmq-rabbitmq
       - rabbitmq_rabbitmq-server
       - deb https://dl.bintray.com/rabbitmq-erlang/debian {{ salt['grains.get']('oscodename') }} erlang
diff --git a/rabbitmq/service/running.sls b/rabbitmq/service/running.sls
index 3afcc15b..4d5e597f 100644
--- a/rabbitmq/service/running.sls
+++ b/rabbitmq/service/running.sls
@@ -20,15 +20,8 @@ rabbitmq-service-running-service-running:
     - enable: True
     - watch:
       - sls: {{ sls_config_file }}
+    - onfail_in:
+      - cmd: rabbitmq-service-running-service-running
   cmd.run:
     - names:
-      - localectl || true
       - journalctl -xe -u {{ rabbitmq.service.name }} || true
-      - systemctl status {{ rabbitmq.service.name }} || true
-      - systemd-analyze blame || true
-      - hostname
-      - hostname -s
-      - hostname -A
-      - ip addr
-    - onfail:
-      - service: rabbitmq-service-running-service-running