PostgreSQL的学习心得和知识总结（一百五十八）|在线调优工具pgtune的实现原理和源码解析

目录结构

注：提前言明本文借鉴了以下博主、书籍或网站的内容，其列表如下：

1、参考书籍：《PostgreSQL数据库内核分析》
2、参考书籍：《数据库事务处理的艺术：事务管理与并发控制》
3、PostgreSQL数据库仓库链接，点击前往
4、日本著名PostgreSQL数据库专家铃木启修网站主页，点击前往
5、参考书籍：《PostgreSQL指南：内幕探索》，点击前往
6、参考书籍：《事务处理概念与技术》
7、pgtune 官方git仓库，点击前往
8、pgtune 官方在线使用，点击前往

1、本文内容全部来源于开源社区 GitHub和以上博主的贡献，本文也免费开源（可能会存在问题，评论区等待大佬们的指正）
2、本文目的：开源共享抛砖引玉一起学习
3、本文不提供任何资源不存在任何交易与任何组织和机构无关
4、大家可以根据需要自行复制粘贴以及作为其他个人用途，但是不允许转载不允许商用（写作不易，还请见谅 💖）
5、本文内容基于PostgreSQL master源码开发而成

在线调优工具pgtune的实现原理和源码解析

文章快速说明索引
功能使用背景说明
功能实现源码解析

文章快速说明索引

学习目标：

做数据库内核开发久了就会有一种少年得志，年少轻狂的错觉，然鹅细细一品觉得自己其实不算特别优秀远远没有达到自己想要的。也许光鲜的表面掩盖了空洞的内在，每每想到于此，皆有夜半临渊如履薄冰之感。为了睡上几个踏实觉，即日起暂缓其他基于PostgreSQL数据库的兼容功能开发，近段时间将着重于学习分享Postgres的基础知识和实践内幕。

学习内容：（详见目录）

1、在线调优工具pgtune的实现原理和源码解析

学习时间：

2024年11月24日 13:46:59

学习产出：

1、PostgreSQL数据库基础知识回顾 1个
2、CSDN 技术博客 1篇
3、PostgreSQL数据库内核深入学习

注：下面我们所有的学习环境是Centos8+PostgreSQL master +Oracle19C+MySQL8.0

postgres=# select version();version                                                   
------------------------------------------------------------------------------------------------------------PostgreSQL 18devel on x86_64-pc-linux-gnu, compiled by gcc (GCC) 8.5.0 20210514 (Red Hat 8.5.0-21), 64-bit
(1 row)postgres=##-----------------------------------------------------------------------------#SQL> select * from v$version;          BANNER        Oracle Database 19c EE Extreme Perf Release 19.0.0.0.0 - Production	
BANNER_FULL	  Oracle Database 19c EE Extreme Perf Release 19.0.0.0.0 - Production Version 19.17.0.0.0	
BANNER_LEGACY Oracle Database 19c EE Extreme Perf Release 19.0.0.0.0 - Production	
CON_ID 0#-----------------------------------------------------------------------------#mysql> select version();
+-----------+
| version() |
+-----------+
| 8.0.27    |
+-----------+
1 row in set (0.06 sec)mysql>

功能使用背景说明

在这里插入图片描述

作用：根据硬件调整 PostgreSQL 配置。

优点：

无需下载或安装任何东西
也可以离线工作
可以作为移动应用程序工作
开源

使用一下，如下是(常用TPCC)一个推荐配置：

在这里插入图片描述

接下来看一下其实现细节，使用上面比较简单，这里不再赘述。后面大家在使用过程中有问题，也可以直接去官方仓库下提问题！

注1：因为这个算是一个通用设置，所能提供的参数较少(如下) 在具体的使用中，其他相关参数的设置大家可以自行分析：

    const configData = [['max_connections', maxConnectionsVal],['shared_buffers', formatValue(sharedBuffersVal)],['effective_cache_size', formatValue(effectiveCacheSizeVal)],['maintenance_work_mem', formatValue(maintenanceWorkMemVal)],['checkpoint_completion_target', checkpointCompletionTargetVal],['wal_buffers', formatValue(walBuffersVal)],['default_statistics_target', defaultStatisticsTargetVal],['random_page_cost', randomPageCostVal],['effective_io_concurrency', effectiveIoConcurrencyVal],['work_mem', formatValue(workMemVal)],['huge_pages', hugePagesVal]]

注2：这不是标准答案，可以视为一种参考建议值大家在使用中可以根据自己的实际情况进行调整！

可选的应用场景，有：

const dbTypeOptions = () => [{label: 'Web application',value: DB_TYPE_WEB},{label: 'Online transaction processing system',value: DB_TYPE_OLTP},{label: 'Data warehouse',value: DB_TYPE_DW},{label: 'Desktop application',value: DB_TYPE_DESKTOP},{label: 'Mixed type of application',value: DB_TYPE_MIXED}
]

因此大家要充分理解自己的业务情况！

功能实现源码解析

下面我们按照参数的顺序，挨个解释pgtune设置值的细节。

一、max_connections 这个值设置多少即为多少；若是没有设置则使用下面的默认值：

参考点击：http://postgres.cn/docs/current/runtime-config-connection.html#GUC-MAX-CONNECTIONS

export const selectMaxConnections = createSelector([selectConnectionNum, selectDBType],(connectionNum, dbType) =>connectionNum? connectionNum // 设置的走设置值，而非下面的默认值: {[DB_TYPE_WEB]: 200,[DB_TYPE_OLTP]: 300,[DB_TYPE_DW]: 40,[DB_TYPE_DESKTOP]: 20,[DB_TYPE_MIXED]: 100}[dbType]
)

二、shared_buffers

参考点击：http://postgres.cn/docs/current/runtime-config-resource.html#GUC-SHARED-BUFFERS

export const selectSharedBuffers = createSelector([selectTotalMemoryInKb, selectDBType, selectOSType, selectDBVersion],(totalMemoryKb, dbType, osType, dbVersion) => {let sharedBuffersValue = {[DB_TYPE_WEB]: Math.floor(totalMemoryKb / 4),[DB_TYPE_OLTP]: Math.floor(totalMemoryKb / 4),[DB_TYPE_DW]: Math.floor(totalMemoryKb / 4),[DB_TYPE_DESKTOP]: Math.floor(totalMemoryKb / 16),[DB_TYPE_MIXED]: Math.floor(totalMemoryKb / 4)}[dbType]if (dbVersion < 10 && OS_WINDOWS === osType) {// Limit shared_buffers to 512MB on Windowsconst winMemoryLimit = (512 * SIZE_UNIT_MAP['MB']) / SIZE_UNIT_MAP['KB']if (sharedBuffersValue > winMemoryLimit) {sharedBuffersValue = winMemoryLimit}}return sharedBuffersValue}
)

该参数通常设置建议为：totalMemoryKb的1/4 或 1/16(看业务场景)
在pg10以下的Windows环境上，建议设置不超过 512MB

三、effective_cache_size 该参数通常建议设置为totalMemoryKb的 3/4

参考点击：http://postgres.cn/docs/current/runtime-config-query.html#GUC-EFFECTIVE-CACHE-SIZE

export const selectEffectiveCacheSize = createSelector([selectTotalMemoryInKb, selectDBType],(totalMemoryKb, dbType) =>({[DB_TYPE_WEB]: Math.floor((totalMemoryKb * 3) / 4),[DB_TYPE_OLTP]: Math.floor((totalMemoryKb * 3) / 4),[DB_TYPE_DW]: Math.floor((totalMemoryKb * 3) / 4),[DB_TYPE_DESKTOP]: Math.floor(totalMemoryKb / 4),[DB_TYPE_MIXED]: Math.floor((totalMemoryKb * 3) / 4)})[dbType]
)

四、maintenance_work_mem

参考点击：http://postgres.cn/docs/current/runtime-config-resource.html#GUC-MAINTENANCE-WORK-MEM

export const selectMaintenanceWorkMem = createSelector([selectTotalMemoryInKb, selectDBType, selectOSType],(totalMemoryKb, dbType, osType) => {let maintenanceWorkMemValue = {[DB_TYPE_WEB]: Math.floor(totalMemoryKb / 16),[DB_TYPE_OLTP]: Math.floor(totalMemoryKb / 16),[DB_TYPE_DW]: Math.floor(totalMemoryKb / 8),[DB_TYPE_DESKTOP]: Math.floor(totalMemoryKb / 16),[DB_TYPE_MIXED]: Math.floor(totalMemoryKb / 16)}[dbType]// Cap maintenance RAM at 2GB on servers with lots of memoryconst memoryLimit = (2 * SIZE_UNIT_MAP['GB']) / SIZE_UNIT_MAP['KB']if (maintenanceWorkMemValue >= memoryLimit) {if (OS_WINDOWS === osType) {// 2048MB (2 GB) will raise error at Windows, so we need remove 1 MB from itmaintenanceWorkMemValue = memoryLimit - (1 * SIZE_UNIT_MAP['MB']) / SIZE_UNIT_MAP['KB']} else {maintenanceWorkMemValue = memoryLimit}}return maintenanceWorkMemValue}
)

该参数通常是totalMemoryKb的 1/16 或 1/8
若是该值大于2G，则设置为2G

五、checkpoint_completion_target 设置为0.9

参考点击：http://postgres.cn/docs/current/runtime-config-wal.html#GUC-CHECKPOINT-COMPLETION-TARGET

export const selectCheckpointCompletionTarget = createSelector([],() => 0.9 // based on https://github.com/postgres/postgres/commit/bbcc4eb2
)

六、wal_buffers

参考点击：http://postgres.cn/docs/current/runtime-config-wal.html#GUC-WAL-BUFFERS

export const selectWalBuffers = createSelector([selectSharedBuffers], (sharedBuffersValue) => {// Follow auto-tuning guideline for wal_buffers added in 9.1, where it's// set to 3% of shared_buffers up to a maximum of 16MB.let walBuffersValue = Math.floor((3 * sharedBuffersValue) / 100)const maxWalBuffer = (16 * SIZE_UNIT_MAP['MB']) / SIZE_UNIT_MAP['KB']if (walBuffersValue > maxWalBuffer) {walBuffersValue = maxWalBuffer}// It's nice of wal_buffers is an even 16MB if it's near that number. Since// that is a common case on Windows, where shared_buffers is clipped to 512MB,// round upwards in that situationconst walBufferNearValue = (14 * SIZE_UNIT_MAP['MB']) / SIZE_UNIT_MAP['KB']if (walBuffersValue > walBufferNearValue && walBuffersValue < maxWalBuffer) {walBuffersValue = maxWalBuffer}// if less, than 32 kb, than set it to minimumif (walBuffersValue < 32) {walBuffersValue = 32}return walBuffersValue
})

遵循 9.1 中添加的 wal_buffers 自动调整指南，将其设置为 shared_buffers 的 3%，最高为 16MB。
如果 wal_buffers 接近该数字，则最好是 16MB。由于这是 Windows 上的常见情况，其中 shared_buffers 被限制为 512MB，因此在这种情况下向上舍入
如果小于 32 kb，则将其设置为最小值

七、default_statistics_target 该参数使用以下默认值

参考点击：http://postgres.cn/docs/current/runtime-config-query.html#GUC-DEFAULT-STATISTICS-TARGET

export const selectDefaultStatisticsTarget = createSelector([selectDBType],(dbType) =>({[DB_TYPE_WEB]: 100,[DB_TYPE_OLTP]: 100,[DB_TYPE_DW]: 500,[DB_TYPE_DESKTOP]: 100,[DB_TYPE_MIXED]: 100})[dbType]
)

八、random_page_cost 根据硬盘的类型，该参数选择以下值

参考点击：http://postgres.cn/docs/current/runtime-config-query.html#GUC-RANDOM-PAGE-COST

export const selectRandomPageCost = createSelector([selectHDType], (hdType) => {return {[HARD_DRIVE_HDD]: 4,[HARD_DRIVE_SSD]: 1.1,[HARD_DRIVE_SAN]: 1.1}[hdType]
})

九、effective_io_concurrency 该参数在Linux系统下，根据硬盘类型设置如下默认值

参考点击：http://postgres.cn/docs/current/runtime-config-resource.html#GUC-EFFECTIVE-IO-CONCURRENCY

export const selectEffectiveIoConcurrency = createSelector([selectOSType, selectHDType],(osType, hdType) => {if (osType !== OS_LINUX) {return null}return {[HARD_DRIVE_HDD]: 2,[HARD_DRIVE_SSD]: 200,[HARD_DRIVE_SAN]: 300}[hdType]}
)

十、parallel setting

在分析work_mem之前，首先先看一下parallel setting相关：

// 默认设置如下const DEFAULT_DB_SETTINGS = {default: {['max_worker_processes']: 8,['max_parallel_workers_per_gather']: 2,['max_parallel_workers']: 8}
}

如下(如果cpuNum未指定或者太小，不再设置相关并行设置进而使用上面默认)：

export const selectParallelSettings = createSelector([selectDBVersion, selectDBType, selectCPUNum],(dbVersion, dbType, cpuNum) => {if (!cpuNum || cpuNum < 4) {return []}let workersPerGather = Math.ceil(cpuNum / 2)if (dbType !== DB_TYPE_DW && workersPerGather > 4) {// 没有明确的证据表明每个新工人都会为每个新核心带来巨大的利益workersPerGather = 4 // no clear evidence, that each new worker will provide big benefit for each noew core}let config = [{key: 'max_worker_processes',value: cpuNum},{key: 'max_parallel_workers_per_gather',value: workersPerGather}]if (dbVersion >= 10) {config.push({key: 'max_parallel_workers',value: cpuNum})}if (dbVersion >= 11) {let parallelMaintenanceWorkers = Math.ceil(cpuNum / 2)if (parallelMaintenanceWorkers > 4) {parallelMaintenanceWorkers = 4 // no clear evidence, that each new worker will provide big benefit for each noew core}config.push({key: 'max_parallel_maintenance_workers',value: parallelMaintenanceWorkers})}return config}
)

max_worker_processes设置成cpunum
max_parallel_workers_per_gather的设置：数仓的场景下 workersPerGather = cpunum/2；其他场景下 workersPerGather通常最大都是 4
在pg10及其以上，max_parallel_workers设置成cpunum
在pg11及其以上，max_parallel_maintenance_workers的设置通常最大都是 4

参考点击：

http://postgres.cn/docs/current/runtime-config-resource.html#GUC-MAX-WORKER-PROCESSES
http://postgres.cn/docs/current/runtime-config-resource.html#GUC-MAX-PARALLEL-WORKERS-PER-GATHER
http://postgres.cn/docs/current/runtime-config-resource.html#GUC-MAX-PARALLEL-WORKERS
http://postgres.cn/docs/current/runtime-config-resource.html#GUC-MAX-PARALLEL-MAINTENANCE-WORKERS

十一、work_mem

参考点击：http://postgres.cn/docs/current/runtime-config-resource.html#GUC-WORK-MEM

export const selectWorkMem = createSelector([selectTotalMemoryInKb,selectSharedBuffers,selectMaxConnections,selectParallelSettings,selectDbDefaultValues,selectDBType],(totalMemoryKb,sharedBuffersValue,maxConnectionsValue,parallelSettingsValue,dbDefaultValues,dbType) => {const parallelForWorkMem = (() => {if (parallelSettingsValue.length) {const maxParallelWorkersPerGather = parallelSettingsValue.find((param) => param['key'] === 'max_parallel_workers_per_gather')if (maxParallelWorkersPerGather &&maxParallelWorkersPerGather['value'] &&maxParallelWorkersPerGather['value'] > 0) {return maxParallelWorkersPerGather['value']}}if (dbDefaultValues['max_parallel_workers_per_gather'] &&dbDefaultValues['max_parallel_workers_per_gather'] > 0) {return dbDefaultValues['max_parallel_workers_per_gather']}return 1})()// work_mem is assigned any time a query calls for a sort, or a hash, or any other structure that needs a space allocation, which can happen multiple times per query. So you're better off assuming max_connections * 2 or max_connections * 3 is the amount of RAM that will actually use in reality. At the very least, you need to subtract shared_buffers from the amount you're distributing to connections in work_mem.// The other thing to consider is that there's no reason to run on the edge of available memory. If you do that, there's a very high risk the out-of-memory killer will come along and start killing PostgreSQL backends. Always leave a buffer of some kind in case of spikes in memory usage. So your maximum amount of memory available in work_mem should be ((RAM - shared_buffers) / (max_connections * 3) / max_parallel_workers_per_gather).const workMemValue =(totalMemoryKb - sharedBuffersValue) / (maxConnectionsValue * 3) / parallelForWorkMemlet workMemResult = {[DB_TYPE_WEB]: Math.floor(workMemValue),[DB_TYPE_OLTP]: Math.floor(workMemValue),[DB_TYPE_DW]: Math.floor(workMemValue / 2),[DB_TYPE_DESKTOP]: Math.floor(workMemValue / 6),[DB_TYPE_MIXED]: Math.floor(workMemValue / 2)}[dbType]// if less, than 64 kb, than set it to minimumif (workMemResult < 64) {workMemResult = 64}return workMemResult}
)

简单分析一下：

parallelForWorkMem的计算：来源于上面并行设置中max_parallel_workers_per_gather
work_mem的计算：

每次查询调用排序、哈希或任何其他需要空间分配的结构时，都会分配 work_mem，每个查询可能会多次发生这种情况。因此，最好假设 max_connections * 2 或 max_connections * 3 是实际使用的 RAM 量。至少，您需要从分配给 work_mem 中连接的量中减去 shared_buffers
另一件需要考虑的事情是，没有理由在可用内存的边缘运行。如果这样做，内存不足杀手出现并开始杀死 PostgreSQL 后端的风险非常高。始终留下某种缓冲区以防内存使用量激增。因此，work_mem 中可用的最大内存量应为 ((RAM - shared_buffers) / (max_connections * 3) / max_parallel_workers_per_gather)
根据不同场景计算出最后的建议值

    // work_mem is assigned any time a query calls for a sort, or a hash, or any other structure that needs a space allocation, which can happen multiple times per query. So you're better off assuming max_connections * 2 or max_connections * 3 is the amount of RAM that will actually use in reality. At the very least, you need to subtract shared_buffers from the amount you're distributing to connections in work_mem.// The other thing to consider is that there's no reason to run on the edge of available memory. If you do that, there's a very high risk the out-of-memory killer will come along and start killing PostgreSQL backends. Always leave a buffer of some kind in case of spikes in memory usage. So your maximum amount of memory available in work_mem should be ((RAM - shared_buffers) / (max_connections * 3) / max_parallel_workers_per_gather).const workMemValue =(totalMemoryKb - sharedBuffersValue) / (maxConnectionsValue * 3) / parallelForWorkMemlet workMemResult = {[DB_TYPE_WEB]: Math.floor(workMemValue),[DB_TYPE_OLTP]: Math.floor(workMemValue),[DB_TYPE_DW]: Math.floor(workMemValue / 2),[DB_TYPE_DESKTOP]: Math.floor(workMemValue / 6),[DB_TYPE_MIXED]: Math.floor(workMemValue / 2)}[dbType]// if less, than 64 kb, than set it to minimumif (workMemResult < 64) {workMemResult = 64}

十二、huge_pages

参考点击：http://postgres.cn/docs/current/runtime-config-resource.html#GUC-HUGE-PAGES

export const selectHugePages = createSelector([selectTotalMemoryInKb],// more 32GB - better also have huge page// 超过 32GB - 最好也有大页面(totalMemoryKBytes) => (totalMemoryKBytes >= 33554432 ? 'try' : 'off')
)