сжатие (gzip/brotli): ментальная модель

Question

Accepted Answer

Text сжимается ~70–90% — gzip/brotli cheapest big win на wire. Client шлёт `Accept-Encoding`, server marks `Content-Encoding`. Brotli static max level, gzip/low brotli dynamic, never compress already-compressed.

## Разбор

Compression работает потому что text redundant — HTML, JS, CSS, JSON repeat tokens. Images, video, fonts already compressed — gzip over JPEG wastes CPU и может сделать **bigger**.

Negotiation — two-line HTTP handshake:

```
Request:   Accept-Encoding: br, gzip          ← "I can decode these"
Response:  Content-Encoding: br                ← "I used Brotli"
           Vary: Accept-Encoding              ← "cache separately per encoding"
```

Решения: **какой algorithm**, **какой level**, **build time или per request?**

## Итог

Text — compress aggressively; binaries — skip; always `Vary: Accept-Encoding`; static Brotli-11, dynamic moderate level.