PostgreSQL的高可用性应采用分层设计，首先明确故障范围、恢复点目标（RPO）和恢复时间目标（RTO）。从单主节点开始，逐步引入离线备份、WAL归档和热备份，最终实现多站点灾难恢复。每一层增加特定能力，以确保系统的稳定性和可恢复性，避免高压操作。

PostgreSQL通过顺序扫描、索引扫描和位图扫描等不同策略读取数据。表以8KB页面存储，每个元组包含可见性信息。顺序扫描逐页读取，索引扫描通过索引查找，位图扫描先构建位图再读取。可见性图优化提升性能，降低CPU开销。选择扫描策略时，PostgreSQL利用成本模型评估最佳方案。

It is 3 AM. A rogue DELETE just wiped 500,000 customer records. Traditional recovery takes hours and risks collateral damage. This guide shows you how to recover accidental DELETEs and UPDATEs in...

In PostgreSQL, a DELETE operation does not immediately erase data from disk. The MVCC mechanism retains deleted rows as dead tuples, and reading these dead tuples is one viable approach to data...

I'm proposing a very ambitious patch set for PostgreSQL 19. Only time will tell whether it ends up in the release, but I can't resist using this space to give you a short demonstration of what it...

© Laurenz Albe 2026 PostgreSQL has supported the (non-standard) ON CONFLICT clause for the INSERT statement since version 9.5. In v19, commit 88327092ff added ON CONFLICT ... DO SELECT. A good...

A while ago, I wrote about the index corruption that you can get after an operating system upgrade, and recently I detailed how to keep the pain of having to rebuild indexes minimal. Since this is...

我成功将PostgreSQL实例升级到18版，记录了升级过程中的数据快照和配置调整，计划在下次直接更新主服务器。

A migration case with only a JAR and no source code shows why Oracle syntax compatibility is not always a fake requirement, and how IvorySQL + Pigsty can absorb legacy debt at low cost.

2026年2月9日，Álvaro Herrera 提交补丁，允许按进程类型设置 log_min_messages，支持以逗号分隔的级别列表，兼容旧版本，方便开发者调试。