[RabbitMQ] Add podManagementPolicy for cluster recovery. (helm#13027)

rabbitmq should be initialized one by one when building cluster for the first time. Therefore, the default value of podManagementPolicy is 'OrderedReady' Once the rabbitmq participates in the cluster, it waits for a response from another rabbitMQ in the same cluster at reboot, except the last rabbitmq of the same cluster. If the cluster exits gracefully, you do not need to change the podManagementPolicy because the first rabbitmq of the statefulset always will be last of the cluster. However if the last rabbitmq of the cluster is not the first rabbitmq due to a failure, you must change podManagementPolicy to 'Parallel'. Otherwide the first rabbitmq waits Infinitely. Signed-off-by: Jungsub Shin <[email protected]>
devnulled · Apr 25, 2019 · 6324ade · 6324ade
1 parent 4b3b885
commit 6324ade
Show file tree

Hide file tree

Showing 5 changed files with 28 additions and 1 deletion.
diff --git a/stable/rabbitmq/Chart.yaml b/stable/rabbitmq/Chart.yaml
@@ -1,6 +1,6 @@
 apiVersion: v1
 name: rabbitmq
-version: 5.4.1
+version: 5.5.0
 appVersion: 3.7.14
 description: Open source message broker software that implements the Advanced Message Queuing Protocol (AMQP)
 keywords:

diff --git a/stable/rabbitmq/README.md b/stable/rabbitmq/README.md
@@ -56,6 +56,7 @@ The following table lists the configurable parameters of the RabbitMQ chart and
 | `image.pullSecrets`                  | Specify docker-registry secret names as an array | `nil`                                                   |
 | `image.debug`                        | Specify if debug values should be set            | `false`                                                 |
 | `rbacEnabled`                        | Specify if rbac is enabled in your cluster       | `true`                                                  |
+| `podManagementPolicy`                | Pod management policy                            | `OrderedReady`                                          |
 | `rabbitmq.username`                  | RabbitMQ application username                    | `user`                                                  |
 | `rabbitmq.password`                  | RabbitMQ application password                    | _random 10 character long alphanumeric string_          |
 | `rabbitmq.existingPasswordSecret`    | Existing secret with RabbitMQ credentials        | nil                                                     |

diff --git a/stable/rabbitmq/templates/statefulset.yaml b/stable/rabbitmq/templates/statefulset.yaml
@@ -9,6 +9,7 @@ metadata:
     heritage: "{{ .Release.Service }}"
 spec:
   serviceName: {{ template "rabbitmq.fullname" . }}-headless
+  podManagementPolicy: {{ .Values.podManagementPolicy }}
   replicas: {{ .Values.replicas }}
   updateStrategy:
     type: {{ .Values.updateStrategy.type }}

diff --git a/stable/rabbitmq/values-production.yaml b/stable/rabbitmq/values-production.yaml
@@ -35,6 +35,19 @@ image:
 ## does your cluster have rbac enabled? assume yes by default
 rbacEnabled: true
 
+
+## RabbitMQ should be initialized one by one when building cluster for the first time.
+## Therefore, the default value of podManagementPolicy is 'OrderedReady'
+## Once the RabbitMQ participates in the cluster, it waits for a response from another
+## RabbitMQ in the same cluster at reboot, except the last RabbitMQ of the same cluster.
+## If the cluster exits gracefully, you do not need to change the podManagementPolicy
+## because the first RabbitMQ of the statefulset always will be last of the cluster.
+## However if the last RabbitMQ of the cluster is not the first RabbitMQ due to a failure,
+## you must change podManagementPolicy to 'Parallel'.
+## ref : https://www.rabbitmq.com/clustering.html#restarting
+##
+podManagementPolicy: OrderedReady
+
 ## section of specific values for rabbitmq
 rabbitmq:
   ## RabbitMQ application username

diff --git a/stable/rabbitmq/values.yaml b/stable/rabbitmq/values.yaml
@@ -35,6 +35,18 @@ image:
 ## does your cluster have rbac enabled? assume yes by default
 rbacEnabled: true
 
+## RabbitMQ should be initialized one by one when building cluster for the first time.
+## Therefore, the default value of podManagementPolicy is 'OrderedReady'
+## Once the RabbitMQ participates in the cluster, it waits for a response from another
+## RabbitMQ in the same cluster at reboot, except the last RabbitMQ of the same cluster.
+## If the cluster exits gracefully, you do not need to change the podManagementPolicy
+## because the first RabbitMQ of the statefulset always will be last of the cluster.
+## However if the last RabbitMQ of the cluster is not the first RabbitMQ due to a failure,
+## you must change podManagementPolicy to 'Parallel'.
+## ref : https://www.rabbitmq.com/clustering.html#restarting
+##
+podManagementPolicy: OrderedReady
+
 ## section of specific values for rabbitmq
 rabbitmq:
   ## RabbitMQ application username