apache · zwy991114 · Aug 5, 2026
diff --git a/docs/sql-manual/sql-statements/table-and-view/table/CREATE-TABLE.md b/docs/sql-manual/sql-statements/table-and-view/table/CREATE-TABLE.md
@@ -64,6 +64,7 @@ columns_definition
       [ DEFAULT <col_default_value> ]
       [ ON UPDATE CURRENT_TIMESTAMP (<col_on_update_precision>) ]
       [ COMMENT '<col_comment>' ]
+      [ COMPRESSION '<algorithm>[:<level>]' ]
     -- Additional column definitions
     [ , <col_name> <col_type> [ ... ] ]
 ```
@@ -258,6 +259,10 @@ CREATE TABLE <new_table_name> LIKE <existing_table_name>
 
 > When data is updated, if no value is specified for this column, the current timestamp is used to update the data in this column. Can only be used on tables with a UNIQUE (primary key model).
 
+**COMPRESSION '\<algorithm\>[:\<level\>]'**
+
+> Overrides the table-level compression algorithm for this column. This clause is supported only for OLAP tables in non-cloud deployments and can be specified only in `CREATE TABLE`. Compression levels are optional and are supported only for ZSTD (`1` to `22`) and LZ4HC (`1` to `12`). For supported algorithms, restrictions, and examples, see [Data Compression](../../../../table-design/column-compression.md).
+
 ### Index Related Parameters
 
 **<index_name>**

diff --git a/docs/table-design/column-compression.md b/docs/table-design/column-compression.md
@@ -2,14 +2,16 @@
 {
     "title": "Data Compression",
     "language": "en",
-    "description": "Doris supports columnar storage compression algorithms such as LZ4, ZSTD, Snappy, and Zlib, allowing flexible trade-offs between storage cost and query performance based on workload.",
+    "description": "Doris supports table-level and per-column compression with algorithms such as LZ4, ZSTD, Snappy, and Zlib, allowing flexible trade-offs between storage cost and query performance.",
     "keywords": [
         "Doris data compression",
         "columnar storage compression",
         "LZ4",
         "ZSTD",
         "Snappy",
         "Zlib",
+        "per-column compression",
+        "compression level",
         "compression algorithm selection",
         "storage efficiency optimization"
     ]
@@ -105,9 +107,9 @@ Compression effectiveness depends not only on the chosen algorithm but also on t
 | **Column length**       | Shorter columns are usually easier to compress than longer ones, because compression algorithms can find repeating patterns more efficiently in shorter data blocks. |
 | **NULL value ratio**    | When the proportion of NULL values is high, the compression algorithm can encode them as a special pattern, further reducing storage space. |
 
-## Configuring Compression in Doris
+## Configure Table-Level Compression
 
-When creating a table, specify the compression algorithm via the `compression` parameter in `PROPERTIES`:
+When creating a table, use the `compression` property to set the default compression algorithm for all columns:
 
 ```sql
 CREATE TABLE example_table (
@@ -123,3 +125,91 @@ PROPERTIES (
 ```
 
 Supported values for `compression` are: `none`, `lz4`, `lz4f`, `lz4hc`, `zstd`, `snappy`, and `zlib`. If not specified explicitly, Doris uses the default compression algorithm `ZSTD` (controlled by the FE configuration `default_compression_type`).
+
+## Configure Per-Column Compression
+
+In a non-cloud deployment, you can override the table-level compression algorithm for individual columns of an OLAP table. This is useful when a table contains columns with different data characteristics. For example, you can use fast LZ4F compression for most columns and a higher ZSTD compression level for a large text column.
+
+### Syntax
+
+Add the `COMPRESSION` clause to a column definition in `CREATE TABLE`:
+
+```sql
+<column_name> <data_type> [column_attributes]
+    COMPRESSION '<algorithm>[:<level>]'
+```
+
+- `<algorithm>` is case-insensitive.
+- `<level>` is optional. If it is omitted, Doris uses the default level of the selected algorithm.
+- A column without a `COMPRESSION` clause inherits the table-level `compression` property.
+
+### Parameters
+
+| Parameter | Required | Default | Description |
+| --------- | -------- | ------- | ----------- |
+| `<algorithm>` | Yes | None | Compression algorithm for this column. The value is case-insensitive and must be one of the algorithms listed below. |
+| `<level>` | No | Codec default | Compression level. It can be specified only for `zstd` and `lz4hc`. |
+
+### Supported Algorithms and Levels
+
+| Algorithm | Compression level | Description |
+| --------- | ----------------- | ----------- |
+| `no_compression` | Not supported | Disables compression for the column. |
+| `lz4` | Not supported | Uses the LZ4 codec. |
+| `lz4f` | Not supported | Uses the LZ4 frame codec. |
+| `lz4hc` | Optional, `1` to `12` | Uses LZ4 high-compression mode. A higher level generally increases compression time and compression ratio. |
+| `zstd` | Optional, `1` to `22` | Uses Zstandard. A higher level generally increases compression time and compression ratio. |
+| `snappy` | Not supported | Uses the Snappy codec. |
+| `zlib` | Not supported | Uses the Zlib codec. |
+| `default_compression` | Not supported | Uses the algorithm configured by `default_compression_type`; if that configuration also specifies `default_compression`, Doris uses LZ4F. |
+
+Specifying a level for any algorithm other than `zstd` or `lz4hc` causes the DDL statement to fail.
+
+### Example
+
+The following example uses LZ4F as the table default and overrides the `event_body` column with ZSTD level 9:
+
+```sql
+CREATE TABLE compression_example (
+    event_id BIGINT,
+    event_type VARCHAR(32),
+    event_body STRING COMPRESSION 'zstd:9'
+)
+DUPLICATE KEY(event_id)
+DISTRIBUTED BY HASH(event_id) BUCKETS 1
+PROPERTIES (
+    "replication_num" = "1",
+    "compression" = "lz4f"
+);
+
+INSERT INTO compression_example VALUES
+    (1, 'login', '{"user":"alice","result":"success"}'),
+    (2, 'query', '{"sql":"SELECT COUNT(*) FROM orders"}');
+
+SELECT event_id, event_type FROM compression_example ORDER BY event_id;
+```
+
+```text
++----------+------------+
+| event_id | event_type |
++----------+------------+
+|        1 | login      |
+|        2 | query      |
++----------+------------+
+```
+
+In this table, `event_id` and `event_type` use the table-level LZ4F codec, while `event_body` uses ZSTD level 9.
+
+### Limitations
+
+- Per-column compression is supported only for OLAP tables in non-cloud deployments.
+- You can specify per-column compression only when creating a table. `ALTER TABLE ADD COLUMN` and `ALTER TABLE MODIFY COLUMN` do not support the `COMPRESSION` clause.
+- The `COMPRESSION` clause does not support `ARRAY`, `MAP`, `STRUCT`, `VARIANT`, or `AGG_STATE` columns.
+- An empty specification, an unknown algorithm, a non-integer level, or a level outside the supported range causes the DDL statement to fail.
+
+### Best Practices
+
+- Start with one table-level algorithm, and add per-column overrides only for columns whose size or access pattern justifies a different trade-off.
+- Use ZSTD or LZ4HC levels selectively on large, compressible columns. Higher levels increase write-time CPU consumption and do not always produce a meaningful storage reduction.
+- Benchmark representative load and query workloads before applying different codecs broadly. Compression results depend on value distribution, repetition, encoding, and page size.
+- Because per-column compression cannot be added or changed through schema change, choose the policy during table design. To change it later, create a new table with the desired definitions and migrate the data.
diff --git a/...ent-docs/current/sql-manual/sql-statements/table-and-view/table/CREATE-TABLE.md b/...ent-docs/current/sql-manual/sql-statements/table-and-view/table/CREATE-TABLE.md
@@ -64,6 +64,7 @@ columns_definition
       [ DEFAULT <col_default_value> ]
       [ ON UPDATE CURRENT_TIMESTAMP (<col_on_update_precision>) ]
       [ COMMENT '<col_comment>' ]
+      [ COMPRESSION '<algorithm>[:<level>]' ]
     -- 其他列定义
     [ , <col_name> <col_type> [ ... ] ]
 ```
@@ -260,6 +261,10 @@ CREATE TABLE <table_name> LIKE <source_table>
 
 > 当数据更新时，如果没有指定此列的值，则使用当前时间戳更新此列数据。只能在 UNIQUE（主键模型）的表上使用。
 
+**COMPRESSION '\<algorithm\>[:\<level\>]'**
+
+> 覆盖该列的表级默认压缩算法。该子句仅支持非 Cloud 部署中的 OLAP 表，且只能在 `CREATE TABLE` 中指定。压缩级别为可选参数，仅 ZSTD（`1` 到 `22`）和 LZ4HC（`1` 到 `12`）支持。有关支持的算法、使用限制和示例，请参阅[数据压缩](../../../../table-design/column-compression.md)。
+
 ### 索引相关参数
 
 **<index_name>**

diff --git a/...zh-CN/docusaurus-plugin-content-docs/current/table-design/column-compression.md b/...zh-CN/docusaurus-plugin-content-docs/current/table-design/column-compression.md
@@ -2,14 +2,16 @@
 {
     "title": "数据压缩",
     "language": "zh_CN",
-    "description": "Doris 支持 LZ4、ZSTD、Snappy、Zlib 等列式存储压缩算法，可根据工作负载在存储成本与查询性能间灵活权衡。",
+    "description": "Doris 支持使用 LZ4、ZSTD、Snappy、Zlib 等算法配置表级和列级压缩，可在存储成本与查询性能间灵活权衡。",
     "keywords": [
         "Doris 数据压缩",
         "列式存储压缩",
         "LZ4",
         "ZSTD",
         "Snappy",
         "Zlib",
+        "列级压缩",
+        "压缩级别",
         "压缩算法选择",
         "存储效率优化"
     ]
@@ -105,9 +107,9 @@ Doris 采用**页（Page）级别**的压缩策略：
 | **列长度**         | 较短的列通常比长列更容易压缩，因为压缩算法在较短数据块上能更高效地找到重复模式。      |
 | **空值（NULL）比例** | 空值比例较高时，压缩算法可将其作为特殊模式编码，进一步减少存储空间。                  |
 
-## 在 Doris 中配置压缩
+## 配置表级压缩
 
-建表时通过 `PROPERTIES` 中的 `compression` 参数指定压缩算法：
+建表时通过 `PROPERTIES` 中的 `compression` 参数，为所有列设置默认压缩算法：
 
 ```sql
 CREATE TABLE example_table (
@@ -123,3 +125,91 @@ PROPERTIES (
 ```
 
 `compression` 支持的取值：`none`、`lz4`、`lz4f`、`lz4hc`、`zstd`、`snappy`、`zlib`。如未显式指定，Doris 将使用默认压缩算法 `ZSTD`（由 FE 配置项 `default_compression_type` 控制）。
+
+## 配置列级压缩
+
+在非 Cloud 部署中，可以为 OLAP 表的单个列覆盖表级压缩算法。此功能适用于同一张表中不同列的数据特征差异较大的场景。例如，大多数列使用解压速度较快的 LZ4F，而对大文本列使用更高级别的 ZSTD。
+
+### 语法
+
+在 `CREATE TABLE` 的列定义中添加 `COMPRESSION` 子句：
+
+```sql
+<column_name> <data_type> [column_attributes]
+    COMPRESSION '<algorithm>[:<level>]'
+```
+
+- `<algorithm>` 不区分大小写。
+- `<level>` 可选。省略时，Doris 使用对应算法的默认压缩级别。
+- 未声明 `COMPRESSION` 的列继承表级 `compression` 属性。
+
+### 参数
+
+| 参数 | 是否必需 | 默认值 | 说明 |
+| ---- | -------- | ------ | ---- |
+| `<algorithm>` | 是 | 无 | 该列使用的压缩算法。取值不区分大小写，且必须为下表列出的算法之一。 |
+| `<level>` | 否 | 编解码器默认值 | 压缩级别。仅 `zstd` 和 `lz4hc` 支持指定该参数。 |
+
+### 支持的算法与压缩级别
+
+| 算法 | 压缩级别 | 说明 |
+| ---- | -------- | ---- |
+| `no_compression` | 不支持 | 禁用该列的压缩。 |
+| `lz4` | 不支持 | 使用 LZ4 编解码器。 |
+| `lz4f` | 不支持 | 使用 LZ4 Frame 编解码器。 |
+| `lz4hc` | 可选，取值范围为 `1` 到 `12` | 使用 LZ4 高压缩模式。级别越高，通常压缩耗时和压缩率越高。 |
+| `zstd` | 可选，取值范围为 `1` 到 `22` | 使用 Zstandard。级别越高，通常压缩耗时和压缩率越高。 |
+| `snappy` | 不支持 | 使用 Snappy 编解码器。 |
+| `zlib` | 不支持 | 使用 Zlib 编解码器。 |
+| `default_compression` | 不支持 | 使用 `default_compression_type` 配置指定的算法；如果该配置仍为 `default_compression`，则使用 LZ4F。 |
+
+如果为 `zstd` 和 `lz4hc` 以外的算法指定压缩级别，DDL 语句将执行失败。
+
+### 示例
+
+以下示例使用 LZ4F 作为表级默认算法，并将 `event_body` 列覆盖为 ZSTD 9 级：
+
+```sql
+CREATE TABLE compression_example (
+    event_id BIGINT,
+    event_type VARCHAR(32),
+    event_body STRING COMPRESSION 'zstd:9'
+)
+DUPLICATE KEY(event_id)
+DISTRIBUTED BY HASH(event_id) BUCKETS 1
+PROPERTIES (
+    "replication_num" = "1",
+    "compression" = "lz4f"
+);
+
+INSERT INTO compression_example VALUES
+    (1, 'login', '{"user":"alice","result":"success"}'),
+    (2, 'query', '{"sql":"SELECT COUNT(*) FROM orders"}');
+
+SELECT event_id, event_type FROM compression_example ORDER BY event_id;
+```
+
+```text
++----------+------------+
+| event_id | event_type |
++----------+------------+
+|        1 | login      |
+|        2 | query      |
++----------+------------+
+```
+
+在该表中，`event_id` 和 `event_type` 使用表级 LZ4F，`event_body` 使用 ZSTD 9 级。
+
+### 使用限制
+
+- 列级压缩仅支持非 Cloud 部署中的 OLAP 表。
+- 只能在建表时指定列级压缩。`ALTER TABLE ADD COLUMN` 和 `ALTER TABLE MODIFY COLUMN` 不支持 `COMPRESSION` 子句。
+- `COMPRESSION` 子句不支持 `ARRAY`、`MAP`、`STRUCT`、`VARIANT` 或 `AGG_STATE` 列。
+- 压缩配置为空、算法未知、压缩级别不是整数或超出取值范围时，DDL 语句将执行失败。
+
+### 最佳实践
+
+- 优先使用统一的表级算法，仅对数据量或访问模式确实需要不同权衡的列设置覆盖算法。
+- 仅对体积较大且易压缩的列选择 ZSTD 或 LZ4HC 压缩级别。更高级别会增加写入时的 CPU 开销，并不一定能显著降低存储空间。
+- 在大范围使用不同编解码器前，使用有代表性的导入与查询负载进行测试。实际压缩效果取决于数据分布、重复度、编码方式和页大小。
+- 由于无法通过 Schema Change 新增或修改列级压缩策略，应在表设计阶段确定该策略。如需后续调整，请使用目标列定义创建新表并迁移数据。
diff --git a/...docs/version-4.x/sql-manual/sql-statements/table-and-view/table/CREATE-TABLE.md b/...docs/version-4.x/sql-manual/sql-statements/table-and-view/table/CREATE-TABLE.md
@@ -64,6 +64,7 @@ columns_definition
       [ DEFAULT <col_default_value> ]
       [ ON UPDATE CURRENT_TIMESTAMP (<col_on_update_precision>) ]
       [ COMMENT '<col_comment>' ]
+      [ COMPRESSION '<algorithm>[:<level>]' ]
     -- 其他列定义
     [ , <col_name> <col_type> [ ... ] ]
 ```
@@ -260,6 +261,10 @@ CREATE TABLE <table_name> LIKE <source_table>
 
 > 当数据更新时，如果没有指定此列的值，则使用当前时间戳更新此列数据。只能在 UNIQUE（主键模型）的表上使用。
 
+**COMPRESSION '\<algorithm\>[:\<level\>]'**
+
+> 覆盖该列的表级默认压缩算法。该子句仅支持非 Cloud 部署中的 OLAP 表，且只能在 `CREATE TABLE` 中指定。压缩级别为可选参数，仅 ZSTD（`1` 到 `22`）和 LZ4HC（`1` 到 `12`）支持。有关支持的算法、使用限制和示例，请参阅[数据压缩](../../../../table-design/column-compression.md)。
+
 ### 索引相关参数
 
 **<index_name>**