进行本实验前，需掌握以下技能：

• 熟悉Linux操作系统的运维操作，能够熟练运用常见的Linux命令，如top、iostat等；
• 具备GaussDB数据库的使用经验，能够理解并分析其执行计划；
• 掌握GaussDB中SQL语句的优化策略与方法。

实验环境搭建步骤

1. 确认sysbench等压力测试工具已正确安装并可正常使用。

[root@gs01 lib64]# sysbench --version
sysbench 1.0.17-71abd99
[root@gs01 lib64]#

2. 在GaussDB中创建用于压测的专用用户，供后续sysbench调用。

gsql -d postgres -p 8000 -r

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

3. 验证该sysbench用户是否具备正常登录数据库的能力。

[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. 创建本次测试所需的独立数据库实例。

[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. 执行数据初始化操作，准备基础测试数据集。

[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

6. 若初始化失败，参考以下处理方式：

问题排查与处理方案

当数据初始化出现异常时，可采取如下措施：

• 查看错误信息并定位原因，常见问题如用户权限不足或连接失败。
• 删除原有sysbench用户，并重新创建以确保配置正确。

[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!

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. 完成用户重建后，再次执行数据初始化流程。

[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，检查当前参数 synchronous_commit 的设置值。

gaussdb=# show synchronous_commit;
 synchronous_commit 
--------------------
 off
(1 row)

8. 根据测试需求，调整 synchronous_commit 参数以优化写入性能。

[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. 手动运行环境检测脚本，确保各项服务和配置处于预期状态。

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高负载对性能影响的模拟实验

1. 运行指定脚本，模拟在常规IO负载下系统的业务表现。

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. 执行另一组脚本，人为制造数据盘高IO压力场景，模拟生产环境中可能出现的磁盘瓶颈。

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. 在高IO压力持续状态下，重新启动sysbench进行压测，观察性能变化。

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

实验结果表明：在无额外IO压力的环境下，系统TPS达到357，QPS为7154；而在磁盘IO压力显著增加的情况下，TPS下降至140，QPS降至2806。由此可见，磁盘IO性能对数据库整体吞吐能力具有显著影响。