[prefetch]

Per-client budgets for the prefetch requests <GioLink> sends, and the site-wide switch for prefetching.

[prefetch]
max_concurrent = 5
max_per_second = 20

Which links prefetch, and when, is chosen per link with <GioLink prefetch>. This section bounds what one client's prefetches may cost the server.

Reference

Behavior

• A request is a prefetch when it carries Purpose: prefetch or Sec-Purpose: prefetch, parameters included: a browser's speculation-rules prerender sends Sec-Purpose: prefetch;prerender. Everything else is never counted.
• A client is its IP address after [server] trusted_proxies resolution. Clients idle for a minute are forgotten.
• An over-budget prefetch is logged at warn (prefetch budget exceeded); one refused because prefetching is off is not. Both count in the metrics.
• A slot is released when its response is ready or when the client gives up, so cancelled prefetches never use up the budget.
• An admitted prefetch is an ordinary request: a page with revalidate is served from the page cache and stored in it like any other GET, with the same Cache-Control: prefetching a cached page costs no render, and a prefetch can fill the cache for the next visitor.

No key in this section logs a startup warning, 0 included.

Examples

Turn prefetching off

[prefetch]
enabled = false

A docs site with many viewport links

[prefetch]
max_concurrent = 10
max_per_second = 50

Good to know

• 0 lifts a budget; it does not refuse prefetches. Use enabled = false for that.
• [prefetch] strategy from earlier docs is refused with a hint: the strategy is chosen per link (<GioLink prefetch="hover" | "viewport" | {false}>).

Related

• Linking & Navigating: Prefetching
• <GioLink>
• [[rate_limits]]

Version history