diff --git a/apps/emqx_conf/src/emqx_conf.erl b/apps/emqx_conf/src/emqx_conf.erl
index c57d2c97b..eed6b0bb9 100644
--- a/apps/emqx_conf/src/emqx_conf.erl
+++ b/apps/emqx_conf/src/emqx_conf.erl
@@ -217,7 +217,7 @@ gen_doc(File, SchemaModule, I18nFile, Lang) ->
     Title =
         "# " ++ emqx_release:description() ++ " Configuration\n\n" ++
             "<!--" ++ Version ++ "-->",
-    BodyFile = filename:join([rel, "emqx_conf.template.en.md"]),
+    BodyFile = filename:join([rel, "emqx_conf.template." ++ Lang ++ ".md"]),
     {ok, Body} = file:read_file(BodyFile),
     Opts = #{title => Title, body => Body, desc_file => I18nFile, lang => Lang},
     Doc = hocon_schema_md:gen(SchemaModule, Opts),
diff --git a/rel/emqx_conf.template.en.md b/rel/emqx_conf.template.en.md
index e79e79862..48fa01b61 100644
--- a/rel/emqx_conf.template.en.md
+++ b/rel/emqx_conf.template.en.md
@@ -15,11 +15,15 @@ From bottom up:
 When environment variable `$EMQX_NODE__DATA_DIR` is not set, config `node.data_dir`
 is used.
 
-The `*-override.conf` files are overwritten at runtime when changes
-are made from dashboard UI, management HTTP API, or CLI.
+The `cluster-override.conf` file is overwritten at runtime when changes
+are made from dashboard UI, management HTTP API, or CLI. When clustered,
+after EMQX restarts, it copies the file from the node which has the greatest `uptime`.
 
-**NOTE** Config values from `*-override.conf` are **not** mapped to boot configs for
+:::tip Tip
+Some of the configs (such as `node.name`) are boot-only configs and not overridable.
+Config values from `*-override.conf` are **not** mapped to boot configs for
 the config fields attributed with `mapping: path.to.boot.config.key`
+:::
 
 For detailed override rules, see [Config Overlay Rules](#config-overlay-rules).
 
@@ -43,7 +47,7 @@ node.cookie = "mysecret"
 This flat format is almost backward compatible with EMQX's config file format
 in 4.x series (the so called 'cuttlefish' format).
 
-It is 'almost' compatible because the often HOCON requires strings to be quoted,
+It is not fully compatible because the often HOCON requires strings to be quoted,
 while cuttlefish treats all characters to the right of the `=` mark as the value.
 
 e.g. cuttlefish: `node.name = emqx@127.0.0.1`, HOCON: `node.name = "emqx@127.0.0.1"`.
@@ -57,21 +61,33 @@ For more HOCON syntax, please refer to the [specification](https://github.com/li
 
 To make the HOCON objects type-safe, EMQX introduced a schema for it.
 The schema defines data types, and data fields' names and metadata for config value validation
-and more. In fact, this config document itself is generated from schema metadata.
+and more.
+
+::: tip Tip
+The configuration document you are reading now is generated from schema metadata.
+:::
 
 ### Complex Data Types
 
 There are 4 complex data types in EMQX's HOCON config:
 
-1. Struct: Named using an unquoted string, followed by a predefined list of fields,
-   fields can not start with a number, and are only allowed to use
-   lowercase letters and underscores as word separator.
+1. Struct: Named using an unquoted string, followed by a predefined list of fields.
+   Only lowercase letters and digits are allowed in struct and field names.
+   Alos, only underscore can be used as word separator.
 1. Map: Map is like Struct, however the fields are not predefined.
-   1-based index number can also be used as map keys for an alternative
-   representation of an Array.
 1. Union: `MemberType1 | MemberType2 | ...`
 1. Array: `[ElementType]`
 
+::: tip Tip
+If map filed name is a positive integer number, it is interpreted as an alternative representation of an `Array`.
+For example:
+```
+myarray.1 = 74
+myarray.2 = 75
+```
+will be interpreated as `myarray = [74, 75]`, which is handy when trying to override array elements.
+:::
+
 ### Primitive Data Types
 
 Complex types define data 'boxes' which may contain other complex data
@@ -88,8 +104,10 @@ There are quite some different primitive types, to name a few:
 * `emqx_schema:duration()` # time duration, another format of integer()
 * ...
 
-The primitive types are mostly self-describing, some are built-in, such
-as `atom()`, some are defined in EMQX modules, such as `emqx_schema:duration()`.
+::: tip Tip
+The primitive types are mostly self-describing, so there is usually not a lot to document.
+For types that are not so clear by their names, the field description is to be used to find the details.
+:::
 
 ### Config Paths
 
@@ -128,15 +146,19 @@ For example, this environment variable sets an array value.
 export EMQX_LISTENERS__SSL__L1__AUTHENTICATION__SSL__CIPHERS="[\"TLS_AES_256_GCM_SHA384\"]"
 ```
 
-Unknown environment variables are logged as a `warning` level log, for example:
+::: tip Tip
+Unknown root paths are silently discarded by EMQX, for example `EMQX_UNKNOWN_ROOT__FOOBAR` is
+silently discarded because `unknown_root` is not a predefined root path.
+
+Unknown field names in environment variables are logged as a `warning` level log, for example:
 
 ```
 [warning] unknown_env_vars: ["EMQX_AUTHENTICATION__ENABLED"]
 ```
 
 because the field name is `enable`, not `enabled`.
+:::
 
-<strong>NOTE:</strong> Unknown root keys are however silently discarded.
 
 ### Config Overlay Rules
 
@@ -193,14 +215,16 @@ Arrays in EMQX config have two different representations
 Dot-separated paths with number in it are parsed to indexed-maps
 e.g. `authentication.1={...}` is parsed as `authentication={"1": {...}}`
 
-Indexed-map arrays can be used to override list arrays:
+This feature makes it easy to override array elment values. For example:
 
 ```
 authentication=[{enable=true, backend="built_in_database", mechanism="password_based"}]
 # we can disable this authentication provider with:
 authentication.1.enable=false
 ```
-However, list arrays do not get recursively merged into indexed-map arrays.
+
+::: warning Warning
+List arrays is a full-array override, but not a recursive merge, into indexed-map arrays.
 e.g.
 
 ```
@@ -208,3 +232,4 @@ authentication=[{enable=true, backend="built_in_database", mechanism="password_b
 ## below value will replace the whole array, but not to override just one field.
 authentication=[{enable=true}]
 ```
+:::
diff --git a/rel/emqx_conf.template.zh.md b/rel/emqx_conf.template.zh.md
new file mode 100644
index 000000000..b4a56b28f
--- /dev/null
+++ b/rel/emqx_conf.template.zh.md
@@ -0,0 +1,218 @@
+EMQX的配置文件格式是 [HOCON](https://github.com/emqx/hocon) .
+HOCON（Human-Optimized Config Object Notation）是一个JSON的超集，非常适用于易于人类读写的配置数据存储。
+
+## 分层结构
+
+EMQX的配置文件可分为三层，自底向上依次是：
+
+1. 不可变的基础层 `emqx.conf` 加上 `EMQX_` 前缀的环境变量.</br>
+   修改这一层的配置之后，需要重启节点来使之生效。
+1. 集群范围重载层：`$EMQX_NODE__DATA_DIR/configs/cluster-override.conf`
+1. 节点本地重载层：`$EMQX_NODE__DATA_DIR/configs/local-override.conf`
+
+如果环境变量 `$EMQX_NODE__DATA_DIR` 没有设置，那么该目录会从 emqx.conf 的 `node.data_dir`配置中读取。
+
+配置文件 `cluster-override.conf` 的内容会在运行时被EMQX重写。
+这些重写发生在 dashboard UI，管理HTTP API，或者CLI对集群配置进行修改时。
+当EMQX运行在集群中时，一个EMQX节点重启之后，会从集群中其他节点复制该文件内容到本地。
+
+:::tip Tip
+有些配置项是不能被重载的（例如 `node.name`）.
+配置项如果有 `mapping: path.to.boot.config.key` 这个属性，
+则不能被添加到重载文件中 `*-override.conf` 中。
+:::
+
+更多的重载规则，请参考下文 [配置重载规则](#配置重载规则).
+
+## 配置文件语法
+
+在配置文件中，值可以被记为类似JSON的对象，例如
+
+```
+node {
+    name = "emqx@127.0.0.1"
+    cookie = "mysecret"
+}
+```
+
+另一种等价的表示方法是扁平的，例如
+
+```
+node.name = "127.0.0.1"
+node.cookie = "mysecret"
+```
+
+这种扁平格式几乎与EMQX的配置文件格式向后兼容
+在4.x系列中（所谓的'cuttlefish'格式）。
+
+它并不是完全兼容，因为HOCON经常要求字符串两端加上引号。
+而cuttlefish把`=`符右边的所有字符都视为值。
+
+例如，cuttlefish：`node.name = emqx@127.0.0.1`，HOCON：`node.name = "emqx@127.0.0.1"`。
+
+没有特殊字符的字符串在HOCON中也可以不加引号。
+例如：`foo`，`foo_bar`和`foo_bar_1`。
+
+关于更多的HOCON语法，请参考[规范](https://github.com/lightbend/config/blob/main/HOCON.md)
+
+## Schema
+
+为了使HOCON对象类型安全，EMQX为它引入了一个schema。
+该schema定义了数据类型，以及数据字段的名称和元数据，用于配置值的类型检查等等。
+
+::: tip Tip
+当前阅读到配置文件的文档本身就是由模式元数据生成的。
+:::
+
+### 复杂数据类型
+
+EMQX的配置文件中，有4中复杂数据结构类型，它们分别是：
+
+1. Struct：结构体都是有类型名称的，结构体中可以有任意多个字段。
+   结构体和字段的名称由不带特殊字符的全小些字母组成，名称中可以带数字，但不得以数字开头，多个单词可用下划线分隔。
+1. Map: Map与Struct（结构体）类似，但是内部的字段不是预先定义好的.
+1. Union: 联合 `MemberType1 | MemberType2 | ...`，可以理解为：“不是这个，就是那个”
+1. Array: 数组 `[ElementType]`
+
+::: tip Tip
+如果Map的字段名称是纯数字，它会被解释成一个数组。
+例如
+```
+myarray.1 = 74
+myarray.2 = 75
+```
+会被解析成 `myarray = [74, 75]`。这个用法在重载数组元素的值时候非常有用。
+:::
+
+### 原始数据类型
+
+复杂类型定义了数据 "盒子"，其中可能包含其他复杂数据或原始值。
+有很多不同的原始类型，仅举几个例子。
+
+* 原子 `atom()`
+* 布尔 `boolean()`.
+* 字符串 `string()'。
+* 整形 `integer()'。
+* 浮点数 `float()'.
+* 数值 `number()'。
+* 二进制编码的字符串 `binary()` # 是 `string()` 的另一种格式
+* 时间间隔 `emqx_schema:duration()` # 时间间隔，是 `integer()` 的另一种格式
+* ...
+
+::: tip Tip
+原始类型的名称大多是自我描述的，所以不需要过多的注释。
+但是有一些不是那么直观的数据类型，则需要配合字段的描述文档进行理解
+:::
+
+
+### 配置路径
+
+如果我们把EMQX的配置值理解成一个类似目录树的结构，那么类似于文件系统中使用斜杠或反斜杠进行层级分割，
+EMQX使用的配置路径的层级分割符是 `'.'`
+
+被`'.'`号分割的每一段，则是Struct（结构体）的字段，或Map的key.
+
+下面有几个例子：
+
+```
+node.name = "emqx.127.0.0.1"
+zone.zone1.max_packet_size = "10M"
+authentication.1.enable = true
+```
+
+### 环境变量重载
+
+因为`'.'` 分隔符不能使用于环境变量，所以我们需要使用另一个分割符。EMQX选用的是双下划线`__`。
+为了与其他的环境变量有所区分，EMQX还增加了一个前缀 `EMQX_` 来用作环境变量命名空间。
+
+例如 `node.name` 的重载变量名是 `EMQX_NODE__NAME`。
+
+环境变量的值，是解析成HOCON值的。所以这也使得环境变量可以用来传递复杂数据类型的值。
+
+例如，下面这个环境变量传入一个数组类型的值。
+
+```
+export EMQX_LISTENERS__SSL__L1__AUTHENTICATION__SSL__CIPHERS="[\"TLS_AES_256_GCM_SHA384\"]"
+```
+
+::: tip Tip
+未定义的根路径会被EMQX忽略，例如 `EMQX_UNKNOWN_ROOT__FOOBAR` 这个环境变量会被EMQX忽略，
+因为 `UNKNOWN_ROOT` 不是预先定义好的根路径。
+对于已知的根路径，未知的字段名称将被记录为warning日志，比如下面这个例子。
+
+```
+[warning] unknown_env_vars: ["EMQX_AUTHENTICATION__ENABLED"]
+```
+
+这是因为正确的字段名称是 `enable`，而不是 `enabled`.
+:::
+
+### 配置重载规则
+
+HOCON的值是分层覆盖的，普遍规则如下：
+
+- 在同一个文件中，后（在文件底部）定义的值，覆盖前（在文件顶部）到值。
+- 当按层级覆盖时，高层级的值覆盖低层级的值。
+
+结下来的文档将解释更详细的规则。
+
+#### 结构体
+
+合并覆盖规则。在如下配置中，最后一行的 `debug` 值会覆盖覆盖原先`level`字段的 `error` 值
+但是`enable` 字段保持不变。
+```
+log {
+    console_handler{
+        enable=true,
+        level=error
+    }
+}
+
+## 控制台日志打印先定义为`error`级，后被覆写成`debug`级
+
+log.console_handler.level=debug
+```
+
+#### Map
+
+Map与结构体类似，也是合并覆盖规则。
+如下例子中，`zone1` 的 `max_packet_size` 可以在文件后面覆写.
+
+```
+zone {
+    zone1 {
+        mqtt.max_packet_size = 1M
+    }
+}
+
+## 报文大小限制最先被设置成1MB，后被覆写为10MB
+
+zone.zone1.mqtt.max_packet_size = 10M
+```
+
+#### 数组元素
+
+如上面介绍过，EMQX配置中的数组有两种表达方式。
+
+* 列表格式，例如： `[1, 2, 3]`
+* 带下标的Map格式，例如： `{"1"=1, "2"=2, "3"=3}`
+
+点好（`'.'`）分隔到路径中的纯数字会被解析成数组下标。
+例如，`authentication.1={...}`  会被解析成 `authentication={"1": {...}}`，进而进一步解析成 `authentication=[{...}]`
+有了这个特性，我们就可以轻松覆写数组某个元素的值，例如：
+
+```
+authentication=[{enable=true, backend="built_in_database", mechanism="password_based"}]
+# 可以用下面的方式将第一个元素的 `enable` 字段覆写
+authentication.1.enable=false
+```
+
+::: warning Warning
+使用列表格式是的数组将全量覆写原值，如下例：
+
+```
+authentication=[{enable=true, backend="built_in_database", mechanism="password_based"}]
+## 下面这中方式会导致数组第一个元素的除了 `enable` 以外的其他字段全部丢失
+authentication=[{enable=true}]
+```
+:::