PostgreSQL性能优化实战——索引、SQL与参数调优

PostgreSQL性能优化实战------索引、SQL与参数调优

1. 引言

数据库性能问题80%来自SQL和索引设计，20%来自配置参数。

2. 慢查询分析

2.1 开启慢查询日志

# postgresql.conf

log_min_duration_statement = 1000    # 记录1秒以上的查询
log_line_prefix = '%t [%p] %u@%d '
log_statement = 'ddl'                # 记录DDL操作
log_checkpoints = on
log_lock_waits = on

2.2 分析慢查询

-- 查看当前慢查询
SELECT pid, usename, datname, 
       round(1000 * (now() - query_start), 2) AS duration_ms,
       query
FROM pg_stat_activity 
WHERE state = 'active' 
  AND now() - query_start > interval '500 milliseconds'
ORDER BY duration_ms DESC;

-- 统计历史慢查询（使用pg_stat_statements扩展）
CREATE EXTENSION IF NOT EXISTS pg_stat_statements;

SELECT queryid, query, calls, 
       total_exec_time, mean_exec_time, max_exec_time,
       rows, shared_blks_hit + shared_blks_read AS total_blocks
FROM pg_stat_statements 
ORDER BY mean_exec_time DESC 
LIMIT 10;

3. 执行计划解读

-- 获取执行计划
EXPLAIN (ANALYZE, BUFFERS, TIMING, FORMAT TEXT)
SELECT * FROM orders WHERE customer_id = 12345 AND order_date > '2024-01-01';

-- 关键指标解读
-- Seq Scan：全表扫描（数据量大时效率低）
-- Index Scan：索引扫描（高效）
-- Bitmap Index Scan：位图索引扫描（适合多条件）
-- Nested Loop：嵌套循环连接
-- Hash Join：哈希连接（适合大表连接）
-- Merge Join：归并连接（适合已排序数据）

4. 索引优化实战

4.1 索引类型选择

索引类型	适用场景	示例
B-Tree	等值、范围查询、排序	CREATE INDEX idx_name ON table(col)
Hash	等值查询	CREATE INDEX idx_hash ON table USING HASH(col)
GIN	数组、全文检索、JSON	CREATE INDEX idx_gin ON table USING GIN(json_col)
BRIN	超大表、线性相关数据	CREATE INDEX idx_brin ON table USING BRIN(created_at)

4.2 复合索引最佳实践

-- 错误示例：索引列顺序不当
CREATE INDEX idx_wrong ON orders (order_date, customer_id);
-- 查询：WHERE customer_id = 123 AND order_date > '2024-01-01'
-- 索引只能用到customer_id条件

-- 正确示例：等值查询列在前，范围查询列在后
CREATE INDEX idx_correct ON orders (customer_id, order_date);

-- 覆盖索引（避免回表）
CREATE INDEX idx_covering ON orders (customer_id, order_date) 
INCLUDE (total_amount, status);

4.3 索引维护

-- 查找未使用的索引
SELECT schemaname, tablename, indexname, idx_scan, idx_tup_read
FROM pg_stat_user_indexes 
WHERE idx_scan = 0 AND idx_tup_read = 0;

-- 查找重复索引
SELECT pg_size_pretty(sum(pg_relation_size(indexrelid))::bigint) AS size,
       array_agg(indexrelid::regclass) AS indexes
FROM pg_index 
GROUP BY indrelid, indkey, indcollation, indclass, indoption
HAVING count(*) > 1;

-- 重建索引（减少膨胀）
REINDEX INDEX CONCURRENTLY idx_name;
REINDEX TABLE CONCURRENTLY table_name;

5. SQL优化技巧

5.1 避免常见低效写法

-- 低效：函数包裹索引列
SELECT * FROM users WHERE lower(email) = 'admin@example.com';
-- 高效
SELECT * FROM users WHERE email = 'admin@example.com';
-- 或创建表达式索引
CREATE INDEX idx_lower_email ON users (lower(email));

-- 低效：隐式类型转换
SELECT * FROM orders WHERE order_id = '12345';  -- order_id是int类型
-- 高效
SELECT * FROM orders WHERE order_id = 12345;

-- 低效：使用SELECT *
SELECT * FROM orders WHERE customer_id = 100;
-- 高效：只取需要的列
SELECT order_id, order_date, total_amount FROM orders WHERE customer_id = 100;

5.2 分页优化

-- 低效：传统OFFSET（大偏移量时慢）
SELECT * FROM orders ORDER BY id LIMIT 10 OFFSET 100000;

-- 高效：游标分页（Keyset Pagination）
SELECT * FROM orders 
WHERE id > 100000 
ORDER BY id LIMIT 10;

-- 高效：延迟关联
SELECT o.* FROM orders o
INNER JOIN (
    SELECT id FROM orders 
    ORDER BY id LIMIT 10 OFFSET 100000
) t ON o.id = t.id;

5.3 批量操作优化

-- 批量INSERT
INSERT INTO logs (user_id, action, created_at) VALUES 
(1, 'login', now()), (2, 'click', now()), (3, 'purchase', now());

-- 使用COPY（极速导入）
COPY users FROM '/data/users.csv' DELIMITER ',' CSV HEADER;

-- 批量UPDATE（使用临时表）
CREATE TEMP TABLE temp_updates AS 
SELECT id, new_value FROM source_table;

UPDATE target t 
SET value = tu.new_value 
FROM temp_updates tu 
WHERE t.id = tu.id;

6. 内存参数调优

# 假设服务器配置：32GB RAM，8核CPU

# 共享内存（核心参数）
shared_buffers = 8GB              # 缓存数据块
huge_pages = try                  # 使用大页内存

# 工作内存（每个连接）
work_mem = 64MB                   # 排序/哈希表
maintenance_work_mem = 2GB        # VACUUM/索引创建
temp_buffers = 32MB               # 临时表

# 写入相关
wal_buffers = 32MB
effective_io_concurrency = 200    # SSD硬盘可调高

# 优化器评估
effective_cache_size = 24GB       # 操作系统缓存大小
random_page_cost = 1.1            # SSD硬盘调低到1.1
cpu_tuple_cost = 0.01

7. 连接池配置（PgBouncer示例）

# pgbouncer.ini
[databases]
mydb = host=localhost port=5432 dbname=mydb

[pgbouncer]
listen_addr = 0.0.0.0
listen_port = 6432
auth_type = md5
auth_file = /etc/pgbouncer/userlist.txt

pool_mode = transaction        # 事务级连接池
max_client_conn = 1000
default_pool_size = 50
reserve_pool_size = 10

8. 性能监控指标

-- 缓存命中率（应>99%）
SELECT round(100 * sum(heap_blks_hit) / nullif(sum(heap_blks_hit + heap_blks_read), 0), 2) AS cache_hit_ratio
FROM pg_statio_user_tables;

-- 索引命中率
SELECT round(100 * sum(idx_blks_hit) / nullif(sum(idx_blks_hit + idx_blks_read), 0), 2) AS idx_hit_ratio
FROM pg_statio_user_indexes;

-- 事务提交率
SELECT round(100 * xact_commit / nullif(xact_commit + xact_rollback, 0), 2) AS commit_ratio
FROM pg_stat_database WHERE datname = current_database();

-- 检查点频率（5分钟内不应超过1次）
SELECT checkpoints_timed, checkpoints_req, 
       round(checkpoints_req * 100 / nullif(checkpoints_timed + checkpoints_req, 0), 2) AS req_ratio
FROM pg_stat_bgwriter;