Asset Cache

ODAC's proxy includes a zero-config, adaptive, in-memory cache for static assets (CSS, JS, fonts, images). Unlike the Page Cache, it requires no opt-in and no headers — it just works. The cache automatically adapts to the machine it runs on: a 64 GB server gets generous caching, a 512 MB VPS gets conservative caching, and if memory runs low the cache disables itself entirely.

How it works

When a request comes in for a static asset, ODAC serves it from memory if it's cached. Otherwise it proxies to your app, and — only if the asset is requested often enough — stores the response for future requests. Popular assets stay hot in memory; rarely-requested files are never admitted, so a one-off request can't pollute the cache.

Like the page cache, it uses stale-while-revalidate: a cached asset is served instantly while ODAC checks the backend asynchronously (throttled to once per second per asset) to see if it changed.

What gets cached

Caching is limited to static asset file extensions:

.avif  .bmp  .css  .eot  .gif  .ico  .jpeg  .jpg  .js
.map   .mjs  .otf  .png  .svg  .ttf  .webp  .woff  .woff2

A response is cached only when all of the following hold:

Admission control — only popular assets are cached

ODAC tracks the request frequency of every asset URL, even before it's cached. An asset is admitted only when its score clears a threshold:

score = request_frequency (req/sec) × asset_priority

Asset priority reflects how much each file type affects perceived page load:

Under normal conditions an asset is admitted at roughly 1 request per 5 seconds (a frequency of 0.2). Because priority is a multiplier, a render-blocking CSS file requested once a second easily beats a frequently-fetched source map. This is also how eviction is ordered: the lowest-scoring entries are dropped first.

Memory-aware: it adapts to your server

ODAC continuously receives the host's memory state and recalculates its limits in near-real-time, so the cache scales with actual free RAM:

• Max cache size = up to 15% of available (free) RAM (minimum 16 MB while enabled).
  • A 64 GB machine with 50 GB free → up to ~7.5 GB cache.
  • A 512 MB VPS with 200 MB free → up to ~30 MB cache.
• Memory pressure — when free RAM drops below 30%, ODAC caches only hot assets (high score) and aggressively evicts low-value entries.
• Memory critical — when free RAM drops below 20%, the cache disables itself entirely and purges everything, automatically re-enabling once memory recovers.

You never configure any of this — it's fully automatic.

Eviction

A background task runs every 10 seconds and removes:

• Cold entries — any asset not requested within the last 2 minutes (each hit resets this timer).
• Low-value entries under pressure — when memory is tight, anything below the hot-asset threshold.
• Lowest-score entries first — when the cache needs room for a new asset, it evicts by frequency × priority, dropping the least valuable assets.

Cache hit headers

When an asset is served from cache, ODAC sets:

X-Odac-Cache: HIT
Age: <seconds since cached>
Server: ODAC

Conditional requests are supported — if your If-None-Match (ETag) or If-Modified-Since matches the cached entry, ODAC returns 304 Not Modified without sending the body.

Auto-refresh & purge

• Changed assets are picked up automatically. During background revalidation, if the backend returns updated content (a 200 instead of 304), ODAC replaces the cached copy. You don't need to fingerprint or version filenames for the cache to stay correct.
• Deploys and restarts purge the cache for the affected domain automatically, so fresh assets are served immediately after you ship.

Relationship to the Page Cache