GreatSQL 并行Load Data加快数据导入

关注

原创最新推荐文章于 2026-08-18 08:35:03 发布 508 阅读 AI 写同款· GEO 优化 ›

如何使用MCP2518FD外部CAN FD控制器实现速速CAN通信 1、SPI时序MCP2518FD外部CAN FD控制器具有小尺寸和SPI接口，可实现轻松连接。可将CAN FD通道轻松添加到缺少CAN FD外设或没有所需CAN FD通道的微控制器上。MCP2518FD支持经典格式 (CAN 2.0B) 和CAN灵活数据速率 (CAN FD) 格式中的CAN帧格式，符合ISO11898-1:2015标准。框图：本文实验板MCU：CAN FD控制器：MCP2518FDCAN FD收发器：ATA6560开发环境：想了解开发板更详细信息，请点击。阅读详情

GreatSQL 并行Load Data加快数据导入

数据库信息

数据库版本:GreatSQL 8.0.32-25

Clickhouse表需要导入到 GreatSQL 中，表数据量庞大所以选用导出CSV的方式。

测试数据复现操作

load data

MySQL load data 语句能快速将一个文本文件的内容导入到对应的数据库表中（一般文本的一行对应表的一条记录）。数据库应用程序开发中，涉及大批量数据需要插入时，使用 load data 语句的效率比一般的 insert 语句的高很多可以看成select … into outfile语句的反操作，select … into outfile将数据库表中的数据导出保存到一个文件中。

load data 语法

LOAD DATA
    [LOW_PRIORITY | CONCURRENT] [LOCAL]
    INFILE 'file_name'
    [REPLACE | IGNORE]
    INTO TABLE tbl_name
    [PARTITION (partition_name [, partition_name] ...)]
    [CHARACTER SET charset_name]
    [{FIELDS | COLUMNS}
        [TERMINATED BY 'string']
        [[OPTIONALLY] ENCLOSED BY 'char']
        [ESCAPED BY 'char']
    ]
    [LINES
        [STARTING BY 'string']
        [TERMINATED BY 'string']
    ]
    [IGNORE number {LINES | ROWS}]
    [(col_name_or_user_var
        [, col_name_or_user_var] ...)]
    [SET col_name={expr | DEFAULT}
        [, col_name={expr | DEFAULT}] ...]

GreatSQL开启load data并行的方法

#并行load data默认关闭，需要手动开启
 show variables like '%gdb_parallel_load%';
+------------------------------+---------+
| Variable_name                | Value   |
+------------------------------+---------+
| gdb_parallel_load            | OFF     |
| gdb_parallel_load_chunk_size | 4194304 |
| gdb_parallel_load_workers    | 6       |
+------------------------------+---------+
3 rows in set (0.03 sec)

方法一：设置session变量

连接数据库，执行set session gdb_parallel_load=on 如需调整文件块大小或线程数，执行 SET SESSION gdb_parallel_load_chunk_size=65536 或 SET SESSION gdb_parallel_load_workers=16。使用原load data语句执行导入。

方法二：load语句增加hint

LOAD /*+ SET_VAR(gdb_parallel_load=ON) SET_VAR(gdb_parallel_load_chunk_size=65536) SET_VAR(gdb_parallel_load_workers=16) */ DATA INFILE '$MYSQLTEST_VARDIR/parallel_load_outfile.txt' INTO TABLE t1;

gdb_parallel_load 是否开启并行
gdb_parallel_load_chunk_size 文件块大小
gdb_parallel_load_workers 开启多少个线程同时导入

开启gdb_parallel_load=ON。默认配置是gdb_parallel_load_chunk_size=4194304，gdb_parallel_load_workers=6

测试数据创建

#Clickhouse制造测试数据
#建表并随机生成1000000行数据插入
CREATE TABLE test
ENGINE = MergeTree
ORDER BY user_id AS
SELECT
    number,
    concat('user_', toString(number)) AS user_id,
    concat('email_', toString(number), '@example.com') AS email,
    rand() AS random_value
FROM numbers(1, 1000000); 
Query id: a707f30c-180f-4453-bc18-b8e86ee46059
Ok.
0 rows in set. Elapsed: 0.575 sec. Processed 1.00 million rows, 8.00 MB (1.74 million rows/s., 13.92 MB/s.)
Peak memory usage: 157.29 MiB.
#查看表数据库和大小
SELECT
    table AS `表名`,
    sum(rows) AS `总行数`,
    formatReadableSize(sum(data_uncompressed_bytes)) AS `原始大小`,
    formatReadableSize(sum(data_compressed_bytes)) AS `压缩大小`,
    round((sum(data_compressed_bytes) / sum(data_uncompressed_bytes)) * 100, 0) AS `压缩率`
FROM system.parts
WHERE database IN ('mytest')
GROUP BY table
Query id: c107871c-d58d-41ff-9bb9-603ab5ad57c9
┌─表名─┬──总行数─┬─原始大小──┬─压缩大小──┬─压缩率─┐
│ test │ 1000000 │ 46.52 MiB │ 16.29 MiB │     35 │
└──────┴─────────┴───────────┴───────────┴────────┘
1 row in set. Elapsed: 0.010 sec. 

SELECT count(*) FROM test
Query id: 0e49726f-75d2-402f-a83d-1c1534489b51
┌─count()─┐
│ 1000000 │
└─────────┘
1 row in set. Elapsed: 0.004 sec.

创建GreatSQL库对应库表结构

greatsql> CREATE TABLE `mytest1`.`test` (  
    `number`  BIGINT PRIMARY KEY, 
    `user_id` VARCHAR(255),  
    `email` VARCHAR(255),  
    `random_value`  INT  
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4;

导出导入语句样例

Clickhouse导出
{ck_cmd} -q 'SELECT * FROM table FORMAT CSV #{ck_cmd}为clickhouse-client的位置
GreatSQL导入
{gdb_cmd} -e "LOAD  /*+ SET_VAR(gdb_parallel_load=ON) SET_VAR(gdb_parallel_load_chunk_size=65536) SET_VAR(gdb_parallel_load_workers=16) */ DATA LOCAL INFILE 'table.csv' INTO TABLE {new_table} fields terminated by ','"
#{gdb_cmd}为greatsql客户端的位置