TPC-H 是一个面向分析型业务(AP)的基准测试，它由一系列热点查询组成，这些热点查询都是高度复杂的，因此执行时间往往都比较长。 在本次实验测试中，将手动向数据库加载TPC-H数据，并保存在名为 tpch 的数据库中。默认TPC-H数据库的表缺少索引，数据库的参数并没有做任何优化，因此执行效率会比较差。 本实验比较浅显，使用openGauss的参数自调优(X-Tuner：gs_xtuner)功能，对数据库进行参数优化，以提升数据库运行性能，让大家对X-Tuner参数自调优有一个初步的了解。
环境信息
- OS：CentOS Linux release 7.6.1810
- openGauss：2.0.0
- CPU：1core
- Memory：4GB
测试数据脚本清单如下：
[omm@lab01 ~]$ ls -l ~/tpch-kit-back/
total 1076780
-rw------- 1 omm dbgrp 24196144 Apr 24 15:39 customer.tbl
-rw------- 1 omm dbgrp 3814 Apr 24 15:39 dss.ddl
-rw------- 1 omm dbgrp 753862072 Apr 24 15:39 lineitem.tbl
-rw------- 1 omm dbgrp 287 May 25 10:52 load.sh
-rw------- 1 omm dbgrp 2199 Apr 24 15:16 nation.tbl
-rw------- 1 omm dbgrp 170452161 Apr 24 15:16 orders.tbl
-rw------- 1 omm dbgrp 10553197 Apr 24 15:11 out0
-rw------- 1 omm dbgrp 118184616 Apr 24 15:10 partsupp.tbl
-rw------- 1 omm dbgrp 23935125 Apr 24 15:11 part.tbl
drwx------ 3 omm dbgrp 4096 Apr 24 15:39 queries
-rw------- 1 omm dbgrp 384 Apr 24 15:07 region.tbl
-rw------- 1 omm dbgrp 1399184 Apr 24 15:07 supplier.tbl
1. 配置pip，并安装setuptools-rust模块
[root@lab01 ~]# wget https://bootstrap.pypa.io/get-pip.py
[root@lab01 ~]# python3 get-pip.py
[root@lab01 ~]# pip -V pip 21.1.2 from /usr/local/lib/python3.6/site-packages/
pip (python 3.6)
[root@lab01 ~]# pip install setuptools-rust
2. 安装依赖包
[omm@lab01 xtuner]$ pip install joblib
[omm@lab01 xtuner]$ pip install threadpoolctl
3. 创建数据库并导入数据
-- 创建数据库tpch
[omm@lab01 ~]$ gsql -d postgres -p 26000 -c "create database tpch with encoding='UTF-8';"
-- 创建测试表
[omm@lab01 ~]$ gsql -d tpch -p 26000 -f ~/tpch-kit-back/dss.ddl
-- 加载测试数据并统计分析
[omm@lab01 ~]$ vi load.sh
---------------------------------------
for i in `ls *.tbl`; do
table=${i/.tbl/}
echo "Loading $table..."
sed 's/|$//' $i > /tmp/$i
gsql -d tpch -p 26000 -c "TRUNCATE $table"
gsql -d tpch -p 26000 -c "\\copy $table FROM '/home/omm/tpch-kit-back/$i' CSV DELIMITER '|'"
gsql -d tpch -p 26000 -c "ANALYZE $table"
done
---------------------------------------
sh load.sh
4. 编辑requirements.txt文件
[omm@lab01 ~]$ cd /gauss/app/bin/dbmind/xtuner/
[omm@lab01 xtuner]$ vi requirements.txt
---------------------------------------
删除:
tensorflow>=2.2.0
keras-rl2
---------------------------------------
5. 生成gs_xtuner参数调优工具(需要连接外网)
[omm@lab01 ~]$ cd /gauss/app/bin/dbmind/xtuner
[omm@lab01 xtuner]$ python3 setup.py install --user
6. 执行快速推荐命令（基于已经作业执行的信息进行推荐，信息来源pg_stat_database等）
[omm@lab01 xtuner]$ gs_xtuner recommend --db-name tpch --db-user omm --host 192.168.0.99 --host-user omm --port 26000
Please input the password of database:
Please input the password of host:
Start to recommend knobs. Just a moment, please.
************************************* Knob Recommendation Report ****************************************
INFO:
+---------------------------------------+----------------------+
| Metric | Value |
+---------------------------------------+----------------------+
| workload_type | ap |
| dirty_background_bytes | 0 |
| current_locks_count | 0.0 |
| current_prepared_xacts_count | 0.0 |
| rollback_commit_ratio | 0.0 |
| average_connection_age | 0.004575 |
| checkpoint_proactive_triggering_ratio | 0.00863557858376511 |
| fetched_returned_ratio | 0.055316264644388206 |
| cache_hit_rate | 0.5028061903026831 |
| os_cpu_count | 1 |
| current_connections | 1.0 |
| checkpoint_avg_sync_time | 1.07037996545769 |
| write_tup_speed | 101.161719229361 |
| used_mem | 131846656.0 |
| all_database_size | 2292057.41015625 |
| shared_buffer_heap_hit_rate | 25.917067253117217 |
| current_free_mem | 3270760 |
| temp_file_size | 3573.07285767967 |
| uptime | 38.3688171772222 |
| os_mem_total | 3879956 |
| checkpoint_dirty_writing_time_window | 450.0 |
| read_write_ratio | 47.82294541597867 |
| read_tup_speed | 4837.86775193848 |
| max_processes | 503 |
| track_activity_size | 503.0 |
| search_modify_ratio | 658741.9884425476 |
| ap_index | 7.5 |
| shared_buffer_toast_hit_rate | 76.6304347826087 |
| block_size | 8.0 |
| shared_buffer_tidx_hit_rate | 82.7893175074184 |
| shared_buffer_idx_hit_rate | 97.6601060219748 |
| enable_autovacuum | True |
| is_64bit | True |
| is_hdd | True |
| load_average | [1.19, 0.82, 0.8] |
+---------------------------------------+----------------------+
p.s: The unit of storage is kB.
WARN:
[0].
The number of CPU cores is a little small. Please do not run too high concurrency.
You are recommended to set max_connections based on the number of CPU cores.
If your job does not consume much CPU, you can also increase it.
[1].
The value of wal_buffers is a bit high. Generally, an excessively large value does not bring better performance.
You can also set this parameter to -1.
The database automatically performs adaptation.
*********************************** Recommended Knob Settings **********************************************
+---------------------------+-----------+--------+---------+---------+
| name | recommend | min | max | restart |
+---------------------------+-----------+--------+---------+---------+
| shared_buffers | 121256 | 72752 | 139448 | True |
| max_connections | 134 | 15 | 269 | True |
| effective_cache_size | 2909967 | 121256 | 2909967 | False |
| wal_buffers | 3789 | 2048 | 3789 | True |
| random_page_cost | 3.0 | 2.0 | 3.0 | False |
| default_statistics_target | 1000 | 100 | 1000 | False |
+---------------------------+-----------+--------+---------+---------+
注意：修改该推荐值之前，请确保硬件条件满足，否则可能会造成数据库无法启动的问题。
7. [可选]迭代推荐命令（全局搜索算法，迭代式执行，每轮执行约2分钟）
- 1> 修改配置文件
vi /home/omm/.local/lib/python3.6/site-packages/openGauss_xtuner-2.0.0-py3.6.egg/tuner/xtuner.conf
-------------------------------------------------
修改如下行：
max_iterations = 3 （从100轮改为3）
benchmark_path = /home/omm/queries
------------------------------------------------
- 2> 执行命令，观察Reward数值变化，粉色输出的轮次为当前较优数值
[omm@lab01 ~]$ time gs_xtuner tune --db-name tpch --db-user omm --host localhost --host-user omm --port 26000
Please input the password of database:
Please input the password of host:
Start to recommend knobs. Just a moment, please.
WARN: The database may restart several times during tuning, continue or not [yes|no]:yes
2021-05-26 11:09:12,710: Recorder is starting.
| iter | target | random... |
-------------------------------------
2021-05-26 11:10:58,017: [0] Current reward is -102.935543, knobs: {'random_page_cost': '2.64'}.
2021-05-26 11:10:58,018: [0] Best reward is -102.935543, knobs: {'random_page_cost': '2.64'}.
2021-05-26 11:10:58,018: [1] Database metrics: [0.6400000000000001, 0.6007798155874045, 0.65].
2021-05-26 11:10:58,018: [1] Benchmark score: -102.899098, used mem: 36444544 kB, reward: -102.935543.
| 1 | -102.9 | 0.6426 |
2021-05-26 11:12:30,939: [1] Current reward is -91.541441, knobs: {'random_page_cost': '2'}.
2021-05-26 11:12:30,941: [1] Best reward is -91.541441, knobs: {'random_page_cost': '2'}.
2021-05-26 11:12:30,941: [2] Database metrics: [0.0, 0.6107552017890537, 2.6].
2021-05-26 11:12:30,942: [2] Benchmark score: -91.504996, used mem: 36444544 kB, reward: -91.541441.
| 2 | -91.54 | 0.003251 |
2021-05-26 11:13:38,617: [2] Current reward is -66.684871, knobs: {'random_page_cost': '2.46'}.
2021-05-26 11:13:38,618: [2] Best reward is -66.684871, knobs: {'random_page_cost': '2.46'}.
2021-05-26 11:13:38,618: [3] Database metrics: [0.45999999999999996, 0.621014394376401, 3.47].
2021-05-26 11:13:38,618: [3] Benchmark score: -66.648426, used mem: 36444544 kB, reward: -66.684871.
| 3 | -66.68 | 0.4565 |
2021-05-26 11:14:53,250: [3] Current reward is -73.748742, knobs: {'random_page_cost': '2.9'}.
2021-05-26 11:14:53,252: [3] Best reward is -66.684871, knobs: {'random_page_cost': '2.46'}.
2021-05-26 11:14:53,252: [4] Database metrics: [0.8999999999999999, 0.6286889335789447, 3.65].
2021-05-26 11:14:53,252: [4] Benchmark score: -73.712297, used mem: 36444544 kB, reward: -73.748742.
| 4 | -73.75 | 0.9016 |
2021-05-26 11:15:58,798: [4] Current reward is -64.467620, knobs: {'random_page_cost': '2.45'}.
2021-05-26 11:15:58,799: [4] Best reward is -64.467620, knobs: {'random_page_cost': '2.45'}.
2021-05-26 11:15:58,799: [5] Database metrics: [0.4500000000000002, 0.633784310797396, 3.45].
2021-05-26 11:15:58,799: [5] Benchmark score: -64.431175, used mem: 36444544 kB, reward: -64.467620.
| 5 | -64.47 | 0.4544 |
2021-05-26 11:16:59,097: [5] Current reward is -59.161970, knobs: {'random_page_cost': '2.43'}.
2021-05-26 11:16:59,099: [5] Best reward is -59.161970, knobs: {'random_page_cost': '2.43'}.
2021-05-26 11:16:59,099: [6] Database metrics: [0.43000000000000016, 0.6393591990442545, 3.91].
2021-05-26 11:16:59,099: [6] Benchmark score: -59.125525, used mem: 36444544 kB, reward: -59.161970.
| 6 | -59.16 | 0.4304 |
2021-05-26 11:18:08,157: [6] Current reward is -67.964937, knobs: {'random_page_cost': '2.39'}.
2021-05-26 11:18:08,158: [6] Best reward is -59.161970, knobs: {'random_page_cost': '2.43'}.
2021-05-26 11:18:08,158: [7] Database metrics: [0.3900000000000001, 0.6445245622485726, 4.05].
2021-05-26 11:18:08,158: [7] Benchmark score: -67.928493, used mem: 36444544 kB, reward: -67.964937.
| 7 | -67.96 | 0.3854 |
2021-05-26 11:19:11,917: [7] Current reward is -62.842104, knobs: {'random_page_cost': '2.43'}.
2021-05-26 11:19:11,918: [7] Best reward is -59.161970, knobs: {'random_page_cost': '2.43'}.
2021-05-26 11:19:11,918: [8] Database metrics: [0.43000000000000016, 0.6489102035318035, 3.5].
2021-05-26 11:19:11,918: [8] Benchmark score: -62.805659, used mem: 36444544 kB, reward: -62.842104.
| 8 | -62.84 | 0.4301 |
=====================================
2021-05-26 11:19:11,926: The tuning process is complete. The best reward is -59.161970, best knobs are:
{'random_page_cost': '2.43'}.
****************************************** Knob Recommendation Report *************************************
INFO:
+---------------------------------------+-----------------------+
| Metric | Value |
+---------------------------------------+-----------------------+
| workload_type | ap |
| dirty_background_bytes | 0 |
| current_locks_count | 0.0 |
| current_prepared_xacts_count | 0.0 |
| rollback_commit_ratio | 0.0002477694554770677 |
| average_connection_age | 0.004734 |
| checkpoint_proactive_triggering_ratio | 0.00938967136150235 |
| fetched_returned_ratio | 0.09276922373936373 |
| uptime | 0.224322521666667 |
| cache_hit_rate | 0.6006356117493342 |
| os_cpu_count | 1 |
| current_connections | 1.0 |
| checkpoint_avg_sync_time | 1.06359368331199 |
| search_modify_ratio | 1007080.6984163317 |
| max_processes | 137 |
| track_activity_size | 137.0 |
| all_database_size | 2292057.41015625 |
| temp_file_size | 2694.18229367111 |
| current_free_mem | 3298680 |
| shared_buffer_heap_hit_rate | 36.42339765350299 |
| used_mem | 36444544.0 |
| os_mem_total | 3879956 |
| checkpoint_dirty_writing_time_window | 450.0 |
| ap_index | 7.5 |
| shared_buffer_toast_hit_rate | 74.11273486430062 |
| read_tup_speed | 7942.47638202933 |
| block_size | 8.0 |
| read_write_ratio | 80.26596656844558 |
| shared_buffer_tidx_hit_rate | 84.41330998248687 |
| shared_buffer_idx_hit_rate | 96.54182833084825 |
| write_tup_speed | 98.9516516216125 |
| enable_autovacuum | True |
| is_64bit | True |
| is_hdd | True |
| load_average | [0.62, 1.08, 1.13] |
+---------------------------------------+-----------------------+
p.s: The unit of storage is kB.
WARN:
[0]. The number of CPU cores is a little small. Please do not run too high concurrency. You are recommended to set max_connections based on the number of CPU cores. If your job does not consume much CPU, you can also increase it.
BAD:
[0]. The value of wal_buffers is too high. Generally, a large value does not bring better performance. ********************************************* Recommended Knob Settings ********************************************
+---------------------------+-----------+--------+---------+---------+
| name | recommend | min | max | restart |
+---------------------------+-----------+--------+---------+---------+
| random_page_cost | 2.43 | 2.0 | 3.0 | False |
| shared_buffers | 121256 | 72752 | 139448 | True |
| max_connections | 134 | 15 | 269 | True |
| effective_cache_size | 2909967 | 121256 | 2909967 | False |
| wal_buffers | 3789 | 1894 | 3789 | True |
| default_statistics_target | 1000 | 100 | 1000 | False |
+---------------------------+-----------+--------+---------+---------+
real 10m12.961s user 0m6.827s sys 0m1.076s
Gauss松鼠会是汇集数据库爱好者和关注者的大本营，
大家共同学习、探索、分享数据库前沿知识和技术，
互助解决问题，共建数据库技术交流圈。
openGauss官网
相关文章
暂无评论...
致力于称为最好的程序员导航网站