当前位置：首页 > web >正文

PostgreSQL的扩展pg_visibility

web 2025/6/17 15:57:12

PostgreSQL的扩展pg_visibility

pg_visibility 是 PostgreSQL 提供的一个高级扩展，用于检查表的可见性映射(Visibility Map, VM)和页面级可见性信息。这个扩展主要帮助数据库管理员诊断和解决与MVCC(多版本并发控制)相关的性能问题。

一、pg_visibility 扩展概述

核心功能

检查可见性映射：查看哪些数据块被标记为"全部可见"
诊断页面问题：识别可能包含死元组(dead tuples)的页面
维护辅助：帮助优化VACUUM操作

适用场景

分析表膨胀(table bloat)问题
诊断VACUUM操作效果不佳的情况
高级性能调优

二、安装与启用

1. 安装扩展

-- 使用超级用户或具有CREATEEXTENSION权限的用户执行
CREATE EXTENSION pg_visibility;-- 验证安装
SELECT * FROM pg_extension WHERE extname = 'pg_visibility';

2. 查看提供的函数

\df pg_visibility.*

三、主要功能详解

1. 基本功能函数

检查表的可见性映射

-- 检查表的可见性映射状态
SELECT * FROM pg_visibility('表名');-- 示例：检查public.sample表的可见性
SELECT * FROM pg_visibility('public.sample');

检查特定页面的可见性

-- 检查特定页面的详细可见性信息
SELECT * FROM pg_visibility_map('表名', 页号);-- 示例：检查public.orders表的第5页
SELECT * FROM pg_visibility_map('public.orders', 5);

2. 高级诊断功能

检查全部可见的页面

-- 找出表中所有被标记为"全部可见"的页面
SELECT blkno FROM pg_visibility('表名') WHERE all_visible;

检查包含死元组的页面

-- 找出表中包含死元组的页面
SELECT blkno FROM pg_visibility('表名') WHERE all_visible IS FALSE;

四、实际应用场景

1. 表膨胀分析

-- 结合pg_class分析表膨胀情况
SELECT c.relname,pg_size_pretty(pg_total_relation_size(c.oid)) AS total_size,(SELECT count(*) FROM pg_visibility(c.oid) WHERE all_visible) AS visible_blocks,(SELECT count(*) FROM pg_visibility(c.oid) WHERE all_visible IS FALSE) AS non_visible_blocks
FROM pg_class c
WHERE c.relkind = 'r' AND c.relnamespace = (SELECT oid FROM pg_namespace WHERE nspname = 'public')
ORDER BY non_visible_blocks DESC;

2. 优化VACUUM策略

-- 识别需要重点VACUUM的表
WITH vm_stats AS (SELECT c.oid,c.relname,(SELECT count(*) FROM pg_visibility(c.oid) WHERE all_visible IS FALSE) AS bad_blocksFROM pg_class cWHERE c.relkind = 'r'AND c.relnamespace = (SELECT oid FROM pg_namespace WHERE nspname = 'public')
)
SELECT relname,bad_blocks,pg_size_pretty(pg_total_relation_size(oid)) AS size,CASE WHEN bad_blocks > 100 THEN '需要紧急VACUUM'WHEN bad_blocks > 20 THEN '建议近期VACUUM'ELSE '状态良好'END AS recommendation
FROM vm_stats
WHERE bad_blocks > 0
ORDER BY bad_blocks DESC;

3. 验证VACUUM效果

-- 在VACUUM前后比较可见性映射变化
-- VACUUM前
CREATE TEMP TABLE pre_vacuum_visibility AS
SELECT * FROM pg_visibility('large_table');-- 执行VACUUM
VACUUM (VERBOSE, ANALYZE) large_table;-- VACUUM后
SELECT p.blkno,p.all_visible AS before_vacuum,n.all_visible AS after_vacuum
FROM pre_vacuum_visibility pJOIN pg_visibility('large_table') n ON p.blkno = n.blkno
WHERE p.all_visible IS DISTINCT FROM n.all_visible;

五、高级用法

1. 批量检查所有表的可见性状态

-- 生成检查所有用户表的SQL
SELECT format('SELECT %L AS table_name, (SELECT count(*) FROM pg_visibility(%L) WHERE all_visible IS FALSE) AS bad_blocks', n.nspname || '.' || c.relname, n.nspname || '.' || c.relname)
FROM pg_class c
JOIN pg_namespace n ON c.relnamespace = n.oid
WHERE c.relkind = 'r'AND n.nspname NOT LIKE 'pg_%'AND n.nspname != 'information_schema'
\gexec

2. 与pageinspect扩展结合使用

-- 首先安装pageinspect扩展
CREATE EXTENSION IF NOT EXISTS pageinspect;-- 检查特定页面的详细内容
SELECT v.blkno,v.all_visible,(SELECT count(*) FROM heap_page_items(get_raw_page('public.orders', v.blkno)) WHERE t_infomask & 256 = 0) AS dead_tuples
FROM pg_visibility('public.orders') v
WHERE v.all_visible IS FALSE
LIMIT 10;

六、注意事项

性能影响：
- 扫描大型表的可见性映射可能消耗大量I/O资源
- 建议在低峰期执行相关诊断操作
权限要求：
- 需要超级用户或对目标表有SELECT权限
- 某些函数可能需要额外的权限
版本兼容性：
- 不同PostgreSQL版本中函数可能略有不同
- 建议查看对应版本的文档

七、最佳实践

定期监控：

-- 创建定期监控视图
CREATE VIEW visibility_monitor AS
SELECT n.nspname || '.' || c.relname AS table_name,pg_size_pretty(pg_total_relation_size(c.oid)) AS size,(SELECT count(*) FROM pg_visibility(c.oid) WHERE all_visible IS FALSE) AS non_visible_blocks,(SELECT count(*) FROM pg_visibility(c.oid)) AS total_blocks,round((SELECT count(*) FROM pg_visibility(c.oid) WHERE all_visible IS FALSE)::numeric / (SELECT count(*) FROM pg_visibility(c.oid)) * 100, 2) AS pct_non_visible
FROM pg_class c
JOIN pg_namespace n ON c.relnamespace = n.oid
WHERE c.relkind = 'r'AND n.nspname NOT LIKE 'pg_%'AND n.nspname != 'information_schema';

自动化报告：

-- 生成需要关注的表列表
SELECT * FROM visibility_monitor
WHERE non_visible_blocks > 10
ORDER BY pct_non_visible DESC;

与autovacuum配置结合：

-- 根据可见性状态调整autovacuum参数
ALTER TABLE high_churn_table SET (autovacuum_vacuum_scale_factor = 0.05,autovacuum_vacuum_threshold = 1000,autovacuum_analyze_scale_factor = 0.02
);