KQL/ADX下基于表名批量重命名列的高性能方案咨询

• 原有方案存在三个行级高开销操作：全量数据逐行打包为动态对象、动态对象转字符串执行正则替换、替换后反序列化再拆包，数据量越大性能损耗越明显。

方案1：轻量替换（代码改动最小，性能提升3-5倍） #

使用Kusto原生函数bag_rename_keys直接操作动态包的键名，完全避免字符串序列化、正则匹配开销：

let T1 = datatable (Key:string , Col2:string , Col3:string )
[
  "1", "b", "c",
  "2", "e", "f",
  "3", "h", "i"
] 
| extend PackedRecord = bag_rename_keys(pack_all(), dynamic([]), (k) => strcat("T1_", k))
| evaluate bag_unpack(PackedRecord);
let T2 = datatable (Key:string , Col2:string , Col3:string )
[
  "1", "B", "C",
  "2", "E", "F",
  "4", "H", "I"
] 
| extend PackedRecord = bag_rename_keys(pack_all(), dynamic([]), (k) => strcat("T2_", k))
| evaluate bag_unpack(PackedRecord);
let JoinTable = T1 | join kind=inner T2 on $left.T1_Key == $right.T2_Key;
JoinTable

方案2：最优性能（和手动写列重命名效率完全一致，适合TB级以上大数据集） #

提前获取表的列名生成静态投影语句，完全避免逐行动态对象操作，性能无额外损耗：

// 定义原始表
let T1 = datatable (Key:string , Col2:string , Col3:string )
[
  "1", "b", "c",
  "2", "e", "f",
  "3", "h", "i"
];
let T2 = datatable (Key:string , Col2:string , Col3:string )
[
  "1", "B", "C",
  "2", "E", "F",
  "4", "H", "I"
];
// 生成T1重命名投影语句
let T1Cols = toscalar(T1 | getschema | summarize make_list(ColumnName));
let T1RenameExpr = strcat_array(array_map((col) => strcat("T1_", col, " = ", col), T1Cols), ", ");
let T1Renamed = dynamic_to_scalar(execute_query(strcat("T1 | project ", T1RenameExpr)));
// 生成T2重命名投影语句
let T2Cols = toscalar(T2 | getschema | summarize make_list(ColumnName));
let T2RenameExpr = strcat_array(array_map((col) => strcat("T2_", col, " = ", col), T2Cols), ", ");
let T2Renamed = dynamic_to_scalar(execute_query(strcat("T2 | project ", T2RenameExpr)));
// 关联查询
T1Renamed | join kind=inner T2Renamed on $left.T1_Key == $right.T2_Key

如果需要简化关联语法，可以在生成投影语句时单独处理Key列，不用加前缀，关联时直接写on Key即可。

内容的提问来源于stack exchange，提问作者PorgtheEaten