1,gpload环境准备
环境准备请参考博主以前的文章gpfdist部署实战:http://blog.csdn.net/mchdba/article/details/72540806 ,安装好gpfdist后,gpload也自动有了,可以自动使用。
安装完后,可以启动gpfdist服务:nohup /data/greenplum/bin/gpfdist -d /data/greenplum/ -p 8090> /home/gpadmin/gpfdist.log &
2,gpload简介
Greenplum的gpload工具通过可读外部表和并行化文件服务器gpfdist(或gpfdists)来加载数据。gpload处理并行化的基于文件的外部表设置,以及允许我们使用单个YAML文件来配置数据格式,外部表定义,以及gpfdist或gpfdists。
要使用gpload工具有几个前提条件必须满足:
1. 使用gpload的服务器必须安装Python 2.6.2或者以上版本,pygresql工具和pyyaml工具(数据库服务端已经安装了python和需要的python库文件)
2. 必须装gpfdist程序,并把它设置到环境变量PATH中(可以从数据库服务器端的安装目录的子目录bin中拷贝该工具)
3. 确保使用gpload工具的ETL服务器与Greenplum所有服务器的联通性,以及机器名解析正确
4. 确保greenplum集群的所有服务器之间的网络是联通的。
gpload通过它的控制文件(YAML格式控制文件)指定装载的细节信息。所以编写控制文件是用好gpload的关键所在。gpload的控制文件采用YAML1.0.0.1文档格式编写,因此它必须得是有效的YAML格式。通过编写的yaml文件的配置,gpload服务与greenplum集群打通,一个yaml配置对应一个greenplum里面的表。
3,查看gpload参数
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- [gpadmin@g01~]$ gpload -? gpload [options] -f configuration file
Options: -h hostname: host to connect to -p port: port to connect to -U username: user to connect as -d database: database to connect to -W: force password authentication -q: quiet mode -D: do not actually load data -v: verbose -V: very verbose -l logfile: log output to logfile --no_auto_trans: do not wrap gpload in transaction --gpfdist_timeout timeout: gpfdist timeout value --version: print version number and exit -?: help
[gpadmin@g01~]$ |
blog源地址:http://blog.csdn.net/mchdba/article/details/72681969,博主黄杉(mchdba),谢绝转载
4,开始gpload实战
创建全局序列,用来获取汇总执行gpload的次数。
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- create SEQUENCE gpload_audit_seq INCREMENT BY 1 MINVALUE 1 NO MAXVALUE START WITH 1; |
创建审计表,记录每一次gpload执行的开始和结束事件信息
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- create table gpload_audit( |
创建gpload加载的表
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- CREATE TABLE expenses |
在gpload服务器上建立临时测试数据文件天
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- [gpadmin@g01gpdextdata]$ more t21.txt 1|aaa 2|zhangsan
[gpadmin@g01gpdextdata]$ [gpadmin@g01gpdextdata]$ more t22.txt 3|wanger 4|mazi
[gpadmin@g01gpdextdata]$ |
在greenplum集群的master库上,建立外部表:
create table t02 ( id integer, name varchar(128) ) location ( 'gpfdist://192.168.2.72:8090/gpextdata/t21.txt', 'gpfdist://192.168.2.72:8090/gpextdata/t22.txt' ) Format 'TEXT' (delimiter as E'|' null as '' escape 'OFF') ; |
创建gpload.yml配置文件
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- [gpadmin@g01~]$ more g2.yml
--- VERSION: 1.0.0.1 DATABASE: yueworld_db USER: mch HOST: 10.254.2.111 PORT: 5432 GPLOAD: INPUT: - SOURCE: LOCAL_HOSTNAME: - 192.168.2.72 PORT: 8090 FILE: - /data/greenplum/gpextdata/t21.txt - /data/greenplum/gpextdata/t22.txt
- COLUMNS: - id: int - name: text - FORMAT: text - DELIMITER: '|' - ERROR_LIMIT: 25 OUTPUT: - TABLE: dw.t02 - MODE: INSERT SQL: - BEFORE: "INSERT INTO gpload_audit VALUES(nextval('gpload_audit_seq'), 'start', 'insert', 'zhangyun_schema.expenses', current_timestamp)" - AFTER: "INSERT INTO gpload_audit VALUES(nextval('gpload_audit_seq'), 'end', 'insert', 'zhangyun_schema.expenses', current_timestamp)" [gpadmin@g01~]$ |
在greenplum db库上赋予mch账号权限:
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- yueworld_db=# grant select,insert on gpload_audit to mch; GRANT yueworld_db=# grant select,insert,update,delete on t01 to mch; GRANT yueworld_db=# yueworld_db=# grant select,usage,update on gpload_audit_seq to mch; GRANT yueworld_db=# grant all on table t02 to mch; GRANT yueworld_db=# |
去gpload服务器上,执行gpload:
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- [gpadmin@g01~]$ gpload -f g2.yml 2017-05-15 17:52:30|INFO|gpload session started 2017-05-15 17:52:30 2017-05-15 17:52:30|INFO|setting schema 'public' for table 't02' 2017-05-15 17:52:30|INFO|started gpfdist -p 8090 -P 8091 -f "/data/greenplum/gpextdata/t21.txt /data/greenplum/gpextdata/t22.txt" -t 30 2017-05-15 17:52:30|INFO|running time: 0.29 seconds 2017-05-15 17:52:30|INFO|rows Inserted = 4 2017-05-15 17:52:30|INFO|rows Updated = 0 2017-05-15 17:52:30|INFO|data formatting errors = 0 2017-05-15 17:52:30|INFO|gpload succeeded [gpadmin@g01~]$ |
在greenplum服务器上查询
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- yueworld_db=# select * from public.t01_ext_1;; id | name ----+------ 1 | aaa 2 | zhangsan 3 | wanger 4 | mazi (8 rows)
yueworld_db=# |
奇怪的是,当你再执行一次gpload -f y2.yml的时候,会持续往外部表里面写数据,而且是将原有的数据重新写入到t02里面去,执行第二次,rows inserted数量是8,等执行第三次的时候,这个rows Inserted会变成16,如下看执行过程:
[gpadmin@g01~]$ gpload -f g2.yml 2017-05-15 17:52:30|INFO|gpload session started 2017-05-15 17:52:30 2017-05-15 17:52:30|INFO|setting schema 'public' for table 't02' 2017-05-15 17:52:30|INFO|started gpfdist -p 8090 -P 8091 -f "/data/greenplum/gpextdata/t21.txt /data/greenplum/gpextdata/t22.txt" -t 30 2017-05-15 17:52:30|INFO|running time: 0.29 seconds 2017-05-15 17:52:30|INFO|rows Inserted = 8 2017-05-15 17:52:30|INFO|rows Updated = 0 2017-05-15 17:52:30|INFO|data formatting errors = 0 2017-05-15 17:52:30|INFO|gpload succeeded [gpadmin@g01~]$ |
Bty:这里只测试了一个表t02,如果实际生产环境中有多个大表的话,可以编写多个yml文件,用gpload启动多进程同时进行数据高速加载。这里曾经做过测试,500w的原始数据,fil://xxx.com方式需要2个小时,直接insert into select …需要4个小时,而gpload结合gpfdist只需要20分钟。
参考文档:https://gpdb.docs.pivotal.io/4330/client_tool_guides/load/unix/gpload.html