惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Y
Y Combinator Blog
博客园_首页
雷峰网
雷峰网
V
V2EX
博客园 - 司徒正美
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
博客园 - Franky
月光博客
月光博客
Hugging Face - Blog
Hugging Face - Blog
WordPress大学
WordPress大学
T
Tailwind CSS Blog
小众软件
小众软件
博客园 - 叶小钗
美团技术团队
酷 壳 – CoolShell
酷 壳 – CoolShell
Apple Machine Learning Research
Apple Machine Learning Research
IT之家
IT之家
MyScale Blog
MyScale Blog
Blog — PlanetScale
Blog — PlanetScale
大猫的无限游戏
大猫的无限游戏
Jina AI
Jina AI
人人都是产品经理
人人都是产品经理
H
Help Net Security
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻

blag

SQLite prefixes its temp files with `etilqs_` - blag Setsum - order agnostic, additive, subtractive checksum - blag Oldest recorded transaction - blag Replacing a cache service with a database - blag SQLite commits are not durable under default settings - blag PSA: SQLite WAL checksums fail silently and may lose data - blag Rickrolling Turso DB (SQLite rewrite in Rust) - blag Collection of insane and fun facts about SQLite - blag How bloom filters made SQLite 10x faster - blag In search of a faster SQLite - blag Galloping Search - blag Building a distributed log using S3 (under 150 lines of Go) - blag Zero Disk Architecture - blag PSA: Most databases do not do checksums by default - blag Disaggregated Storage - a brief introduction - blag Why does SQLite (in production) have such a bad rep? - blag SQLite Slaps - blag Now - blag Learning C - blag Snapshot Testing - blag Win: contribution to libSQL (SQLite) codebase - blag Errata in Hekaton MVCC paper - blag Internet is wholesome: MVCC edition - blag It is becoming difficult for me to be productive in Python - blag MongoDB secondary only index - blag Introducing CaskDB – a project to teach you writing a key-value store - blag Recurse Center: Winter Break - blag Recurse Center Day 24: Hacking Go compiler to add a new keyword - blag Recurse Center Day 20: Django v4 upgrade (from v1) - blag Recurse Center Day 19 - blag
PSA: SQLite does not do checksums - blag
2024-11-09 · via blag

SQLite does not do checksums by default. I learned this from Alex Miller. What does this mean? If there is disk corruption, the database or application won’t be able to know that the database is ‘corrupt’.

Even a single bit flip can cause havoc. This can happen due to a faulty disk, a bug in the disk driver, or when another application (malicious or otherwise) modifies the database files.

This is not a bug - it’s properly documented:

SQLite assumes that the detection and/or correction of bit errors caused by cosmic rays, thermal noise, quantum fluctuations, device driver bugs, or other mechanisms, is the responsibility of the underlying hardware and operating system. SQLite does not add any redundancy to the database file for the purpose of detecting corruption or I/O errors. SQLite assumes that the data it reads is exactly the same data that it previously wrote.

I created a simple script to demonstrate this:

  1. Create a sample database using this script. It creates a bank database and adds a row for Alice with $83K.

  2. Flip a single bit:

     printf '\x00\x00\x00\x00\x00\x80' | dd of=bank.db bs=1 seek=$((0x1ffd)) count=1 conv=notrunc
    
  3. Alice’s balance is now zero. Sorry, Alice.

It passes PRAGMA integrity_check too. Here’s an ASCII animation if you prefer that:

WAL and Checksums

SQLite has checksums for WAL frames. However, when it detects a corrupt frame, it silently ignores the faulty frame and all subsequent frames. It doesn’t even raise an error!

Ignoring frames might be acceptable, but not raising an error is where it gets me.

Checksum VFS Shim

You can use the Checksum VFS Shim, but there’s one important caveat:

Checksumming only works on databases that have a reserve bytes value of exactly 8

The documentation of reserve bytes explains:

SQLite has the ability to set aside a small number of extra bytes at the end of every page for use by extensions. These extra bytes are used, for example, by the SQLite Encryption Extension to store a nonce and/or cryptographic checksum associated with each page. The “reserved space” size in the 1-byte integer at offset 20 is the number of bytes of space at the end of each page to reserve for extensions. This value is usually 0. The value can be odd.

This means if you’re using any extension that uses reserve bytes, you can’t use the Checksum shim.

Again, this is not a bug. Most databases (except a few) assume that the OS, filesystem, and disk are sound. Whether this matters depends on your application and the guarantees you need.

edit: I wrote a follow up post.