Update 3.8.12 release notes

site/release-notes/3.8.12.md

@@ -39,6 +39,61 @@ and [RabbitMQ community Slack](https://rabbitmq-slack.herokuapp.com/).
 
 ## Changes Worth Mentioning
 
+### Core Server
+
+#### Bug Fixes
+
+ * [Maintenance mode](https://www.rabbitmq.com/upgrade.html#maintenance-mode) no longer transfers leaders of classic mirrored queue.
+
+   The original transfer implementation was not guaranteed to be safe when only a
+   subset of cluster nodes hosted replicas. In addition, the process could take
+   a lot of time in environments with a lot of classic mirrored queues, more
+   than in case of node shutdown.
+
+   Quorum queue leadership transfer is still performed as it is dramatically more
+   efficient and never suffered from any downsides of leadership transfer of CMQs.
+
+   GitHub issue: [#2749](https://github.com/rabbitmq/rabbitmq-server/issues/2749)
+
+ * Default replica count for quorum queues was using an incorrect configuration key name.
+   The default value is now `3`, so in a cluster of five or seven nodes, only three
+   will host replicas for newly declared quorum queues. One replica is the node to which
+   the declaring client is connected to, two others are selected randomly.
+
+   Declaring connections can control the number of replicas using the `x-quorum-initial-group-size`
+   optional queue argument at declaration time.
+
+   GitHub issue: [#2759](https://github.com/rabbitmq/rabbitmq-server/pull/2759)
+
+ * Memory monitor could run into an exception if queried very early on node boot
+   before it was fully initialized.
+
+   Contribute @tomyouyou.
+
+   GitHub issue: [#2733](https://github.com/rabbitmq/rabbitmq-server/pull/2733)
+
+#### Enhancements
+
+ * `rabbitmq.conf` validation of TLS certificate and private key paths is now more thorough:
+   it checks if the files have read or read/write permissions instead of checking if it can read
+   the file's metadata.
+
+   GitHub issue: [#2685](https://github.com/rabbitmq/rabbitmq-server/issues/2685)
+
+### CLI Tools
+
+#### Enhancements
+
+ * `rabbitmqctl close_all_user_connections` is a new command that closes all connections that
+   authenticated using a specific username:
+
+   ```
+   rabbitmqctl close_all_user_connections "user-99fc15d3f" "closed to force reconnection"
+   ```
+
+   GitHub issue: [#2715](https://github.com/rabbitmq/rabbitmq-server/issues/2715)
+
+
 ### Prometheus Plugin
 
 #### Bug Fixes
@@ -54,9 +109,19 @@ and [RabbitMQ community Slack](https://rabbitmq-slack.herokuapp.com/).
    GitHub issue: [rabbitmq/rabbitmq-server#2740](https://github.com/rabbitmq/rabbitmq-server/issues/2740)
 
 
+## RabbitMQ Erlang Client
+
+#### Bug Fixes
+
+ * The client performed hostname resolution twice even if IP version preference was specified by the user.
+
+   Contributed by Pawel @haljin Antemijczuk.
+
+   GitHub issue: [#2748](https://github.com/rabbitmq/rabbitmq-server/issues/2748)
+
 ## Dependency Upgrades
 
-No dependency changes.
+ * jose has been upgraded [from `1.10.1` to `1.11.1`](https://github.com/potatosalad/erlang-jose/blob/master/CHANGELOG.md)
 
 
 ## Source Code Archives

site/upgrade.md

@@ -310,7 +310,6 @@ Currently this involves the following steps:
 
 * Suspend all client connection listeners (no new client connections will be accepted)
 * Close all existing client connections: applications are expected to reconnect to other nodes and recover
-* Transfer primary replicas of all classic mirrored queues hosted on the target node
 * Transfer primary replicas of all quorum queues hosted on the target node, and prevent them from participating
     in the subsequently triggered Raft elections
 * Mark the node as down for maintenance