实验二十四 GaussDB运维与故障处理-SQL慢问题

蒋士峰DBA修行之路2025-11-25 12:16

进行本实验前需要具备如下能力：

  1. 具备linux操作系统运维能力，熟练使用Linux运维命令，包括但不限top、iostat等；

  2. 熟练使用GaussDB数据库，具有理解和分析GaussDB数据库执行计划的能力；

  3. 熟悉GaussDB数据库的SQL优化方法。

实验环境准备

1.确定sysbench等压测工具安装成功。

sql 复制代码 
[root@gs01 lib64]# sysbench --version
sysbench 1.0.17-71abd99
[root@gs01 lib64]#

2.GaussDB中创建用户，之后sysbench压测使用。

sql 复制代码 
gsql -d postgres -p 8000 -r

create user sysbench with sysadmin password 'Huawei@1234';

3.验证sysbench用户是否可以登录。

sql 复制代码 
[Ruby@gs01 cm_agent]$ gsql -d postgres -p 8000 -U sysbench -W Huawei@1234 -h 192.168.3.60 -r
gsql ((GaussDB Kernel 505.2.1.SPC0800 build 85996fbb) compiled at 2025-07-03 01:15:58 commit 10558 last mr 24271 release)
SSL connection (cipher: ECDHE-RSA-AES128-GCM-SHA256, bits: 128)
Type "help" for help.

gaussdb=> 
gaussdb=> \l
                          List of databases
   Name    | Owner | Encoding  | Collate | Ctype | Access privileges 
-----------+-------+-----------+---------+-------+-------------------
 postgres  | Ruby  | SQL_ASCII | C       | C     | 
 template0 | Ruby  | SQL_ASCII | C       | C     | =c/Ruby          +
           |       |           |         |       | Ruby=CTc/Ruby
 template1 | Ruby  | SQL_ASCII | C       | C     | =c/Ruby          +
           |       |           |         |       | Ruby=CTc/Ruby
 templatea | Ruby  | SQL_ASCII | C       | C     | =c/Ruby          +
           |       |           |         |       | Ruby=CTc/Ruby
 templatem | Ruby  | SQL_ASCII | C       | C     | =c/Ruby          +
           |       |           |         |       | Ruby=CTc/Ruby
(5 rows)

4.创建测试库。

sql 复制代码 
[Ruby@gs01 cm_agent]$ gsql -d postgres -p 8000 -U sysbench -W Huawei@1234  -c "create database db_test;"
CREATE DATABASE
[Ruby@gs01 cm_agent]$

5.初始化数据。

sql 复制代码 
[root@gs01 share]# sysbench oltp_read_write --tables=1 --table-size=5000000 --threads=8 --time=600 --db-driver=pgsql --pgsql-db=db_test --pgsql-user=sysbench --pgsql-password=Huawei@1234 --pgsql-host=192.168.3.60 --pgsql-port=8000 --report-interval=5 prepare
sysbench 1.0.17-71abd99 (using bundled LuaJIT 2.1.0-beta2)

Initializing worker threads...

FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed
FATAL: Connection to database failed: none of the server's SASL authentication mechanisms are supported

FATAL: `sysbench.cmdline.call_command' function failed: ./oltp_common.lua:88: connection creation failed

解决办法：

sql 复制代码 
[Ruby@gs01 cm_agent]$ gs_guc reload -Z datanode -N all -I all -c "password_encryption_type=0"
The gs_guc run with the following arguments: [gs_guc -Z datanode -N all -I all -c password_encryption_type=0 reload ].
Begin to perform the total nodes: 1.
Popen count is 1, Popen success count is 1, Popen failure count is 0.
Begin to perform gs_guc for datanodes.
Command count is 1, Command success count is 1, Command failure count is 0.

Total instances: 1. Failed instances: 0.
ALL: Success to perform gs_guc!

[Ruby@gs01 cm_agent]$ gs_guc reload -Z datanode -N all -I all -h "host all all 0.0.0.0/0 md5"
The gs_guc run with the following arguments: [gs_guc -Z datanode -N all -I all -h host all all 0.0.0.0/0 md5 reload ].
Begin to perform the total nodes: 1.
Popen count is 1, Popen success count is 1, Popen failure count is 0.
Begin to perform gs_guc for datanodes.
Command count is 1, Command success count is 1, Command failure count is 0.

Total instances: 1. Failed instances: 0.
ALL: Success to perform gs_guc!

删除sysbench用户重新创建：

sql 复制代码 
gsql -d postgres -p 8000 -r

drop database db_test;
drop user sysbench;

create database db_test;
create user sysbench with sysadmin password 'Huawei@1234';

6.重新执行数据初始化。

sql 复制代码 
[root@gs01 share]# sysbench oltp_read_write --tables=1 --table-size=5000000 --threads=8 --time=600 --db-driver=pgsql --pgsql-db=db_test --pgsql-user=sysbench --pgsql-password=Huawei@1234 --pgsql-host=192.168.3.60 --pgsql-port=8000 --report-interval=5 prepare
sysbench 1.0.17-71abd99 (using bundled LuaJIT 2.1.0-beta2)

Initializing worker threads...

Creating table 'sbtest1'...
Inserting 5000000 records into 'sbtest1'
Creating a secondary index on 'sbtest1'...
[root@gs01 share]#

7.登录GaussDB数据库，查询数据库参数synchronous_commit

sql 复制代码 
gaussdb=# show synchronous_commit;
 synchronous_commit 
--------------------
 off
(1 row)

8.重新设置 synchronous_commit 参数。

sql 复制代码 
[Ruby@gs01 cm_agent]$ gs_guc reload -Z datanode -N all -I all -c "synchronous_commit=local"
The gs_guc run with the following arguments: [gs_guc -Z datanode -N all -I all -c synchronous_commit=local reload ].
Begin to perform the total nodes: 1.
Popen count is 1, Popen success count is 1, Popen failure count is 0.
Begin to perform gs_guc for datanodes.
Command count is 1, Command success count is 1, Command failure count is 0.

Total instances: 1. Failed instances: 0.
ALL: Success to perform gs_guc!

9.手动进行环境检测。

sql 复制代码 
sysbench oltp_read_write --tables=1 --table-size=500000000 --threads=8 --time=600 --db-driver=pgsql --pgsql-db=db_test --pgsql-user=sysbench --pgsql-password=Huawei@1234 --pgsql-host=192.168.3.60 --pgsql-port=8000 --report-interval=5 run

磁盘IO压力大造成的性能劣化实验模拟

1.使用如下脚本模拟常规IO压力下业务的性能情况。

sql 复制代码 
sysbench oltp_read_write --tables=1 --table-size=5000000 --threads=8 --time=600 --db-driver=pgsql --pgsql-db=db_test --pgsql-user=sysbench --pgsql-password=Huawei@1234 --pgsql-host=192.168.3.60 --pgsql-port=8000 --report-interval=5 --percentile=99 run

2.执行如下脚本，构建数据盘IO压力高场景。

sql 复制代码 
touch /data/cluster/master/datanode1/test_in
touch /data/cluster/master/datanode1/test_out
dd if=/dev/zero of="/data/cluster/master/datanode1/test_in" bs=3G count=10 iflag=fullblock
while true;
do
dd if="/data/cluster/master/datanode1/test_in" of="/data/cluster/master/datanode1/test_out" bs=3G count=10 iflag=fullblock
done;

3.再次执行sysbench压测。

sql 复制代码 
sysbench oltp_read_write --tables=1 --table-size=5000000 --threads=8 --time=600 --db-driver=pgsql --pgsql-db=db_test --pgsql-user=sysbench --pgsql-password=Huawei@1234 --pgsql-host=192.168.3.60 --pgsql-port=8000 --report-interval=5 --percentile=99 run

通过实验得知，没有IO读写压力的情况下TPS为357，QPS为7154；有IO读写压力的情况下，TPS为140，QPS为 2806。

相关推荐
蒋士峰DBA修行之路5 小时前
实验二十九 dbmind智能运维工具
gaussdb
蒋士峰DBA修行之路1 天前
实验二十三 GaussDB启动失败定位与处理
gaussdb
Gauss松鼠会3 天前
【GaussDB】使用DBLINK连接到ORACLE
数据库·sql·database·gaussdb
热门推荐
01GitHub 镜像站点02【保姆级教程】免费使用Gemini3的5种方法！免翻墙/国内直连03BongoCat - 跨平台键盘猫动画工具04UV安装并设置国内源05Google Antigravity：无法登录？早期错误、登录修复和用户反馈指南06安娜的档案(Anna’s Archive) 镜像网站/国内最新可访问入口（持续更新）07Linux下V2Ray安装配置指南08Spring Boot 4.0 发布总结：新特性、依赖变更与升级指南09Labelme从安装到标注：零基础完整指南10全球最强模型Grok4，国内已可免费使用！（附教程）