CacheParty

A faster drop-in replacement for lancache: one binary that caches game downloads for your whole LAN.

Like lancachenet/monolithic, it sits behind lancache-dns. Clients resolve CDN names to this machine; it caches plain-HTTP downloads from Steam, Battle.net, Epic, Riot, Xbox/WSUS, PlayStation and anything else in cache-domains.

Drop-in means the same environment variables and the same access.log format. Non-goals: caching HTTPS (:443 is a blind SNI passthrough), reading nginx's on-disk cache format.

Advantages over stock lancache

Faster downloads

• Streaming request coalescing. Many clients asking for the same uncached chunk cause one upstream fetch, and every client receives bytes as they arrive. nginx's proxy_cache_lock makes waiters poll for about 500 ms and wait for the whole slice. In benchmarks with 50 clients, cold time to first byte drops from about 500 ms to 50–135 ms, and cold throughput for Steam-style many-small-file downloads is about 1.6x.
• Adaptive read-ahead. After a miss, CacheParty fetches the chunks a client will need next, growing the window on slow origins while sharing it fairly between downloads.
• No internal proxy hop. Redirects are followed inside the upstream client instead of through a second server on localhost:3128.
• Fast domain lookup. A hash map plus a reversed-label suffix tree instead of thousands of regexes per request.

Easier to run

• Internet download speed limit. UPSTREAM_RATE_LIMIT (e.g. 500m) caps how much internet bandwidth the cache uses, so a big download doesn't slow down the rest of the network. Cache hits still run at full LAN speed.
• Fast restarts. A persistent chunk index (bbolt) means startup does not walk millions of files.
• Built-in dashboard. Live throughput, cache hit share, and per-service, per-game, per-client and per-CDN-host traffic, at http://<host>:8080/, plus Prometheus /metrics. See Dashboard, metrics and logs.
• Live settings. Cache size, free-space floor, max age, read-ahead, upstream rate limit and bypass domains can be changed from the dashboard without a restart.
• Cache repair and purge. Bad chunks are repaired on the fly when requested, cacheparty verify checks the whole cache without reading chunk contents, and cacheparty purge (or the dashboard) removes one game. See Cache repair.
• Bypass domains. BYPASS_DOMAINS sends chosen hosts straight through uncached.

Safer caching

• Never serves wrong bytes. If an object changes upstream, all of its chunks are dropped; truncated upstream responses abort the client rather than looking complete.
• Stricter domain matching. Hosts are matched on whole DNS labels, so evilsteamcontent.com is not cached as Steam.

See Compatibility for every behavior difference from nginx lancache and the known limitations.

Quick start

docker run -d --name cacheparty --restart unless-stopped \
  -p 80:80 -p 443:443 -p 8080:8080 \
  -v /srv/cacheparty:/data \
  -v /mnt/cache:/data/cache \
  ghcr.io/antiremy/cacheparty

/data holds logs, settings, dashboard stats and the cache-domains checkout; /data/cache holds the cached chunks and their index, so mount it from the big disk. Swap in your own host paths. CacheParty can't read nginx's cache format, so give it an empty directory rather than an existing monolithic cache.

The image is published to GHCR and Docker Hub for amd64 and arm64: latest tracks main, and releases are tagged by version (1.2.3, 1.2, 1). Each GitHub release also has static linux/amd64 and linux/arm64 binaries. To build the image yourself, run docker build -t cacheparty ..

Then point lancache-dns at the host's IP and open http://<host>:8080/ for the dashboard. The image is built on distroless/static, runs as root so it can bind :80/:443, and has a HEALTHCHECK that calls /cacheparty healthcheck (a request to /lancache-heartbeat).

Set DASHBOARD_PASSWORD to allow changing settings and repairing the cache from the dashboard; without it the dashboard is read-only. All variables are listed in Configuration.

Run locally without root:

NOFETCH=true DOMAINS_DIR=./testdata/domains LOG_DIR=/tmp/gc-logs \
HTTP_ADDR=127.0.0.1:8081 SNI_ADDR=127.0.0.1:8443 METRICS_ADDR=127.0.0.1:8080 \
  go run ./cmd/cacheparty
curl -H 'Host: example.com' http://127.0.0.1:8081/
curl http://127.0.0.1:8080/metrics
open http://127.0.0.1:8080/         # dashboard

Documentation

• Configuration: environment variables, live settings, memory limits, log rotation.
• Dashboard, metrics and logs: what the dashboard shows, /metrics, and the log formats.
• Cache repair: cacheparty verify and cacheparty purge.
• How it works: listeners, cache keys and chunking, coalescing, read-ahead, bypass rules, upstream fetching, storage and eviction.
• Compatibility: differences from nginx lancache and known limitations.
• Benchmarks: methodology and results against monolithic.
• Development: tests, linting and package layout.

License

MIT. See LICENSE.