GetCapabilities takes a minute and clients time out

• Clients abort at 30 seconds with 504 Gateway Time-out nginx/1.24.0.
• curl reports size_download around 18 MB and time_total around 58 s.
• The layer tree spinner ends in service unavailable while a single GetMap still answers in 200 ms.

1. Time the request with curl time_total to separate server time from network time.
2. Read size_download; anything above 5 MB means the document needs trimming.
3. Check the layer count and whether every layer publishes every SRS.
4. Check whether capabilities caching is enabled, since otherwise metadata is rebuilt on each request.
5. Check JVM heap and GC pauses in the server log.

Publish only EPSG:4326 and EPSG:3857 per layer, enable capabilities caching, and put Nginx proxy_cache in front with proxy_cache_valid 200 10m and proxy_cache_lock on. Clients should load /geoserver/rest/layers.json for the layer tree instead of the full capabilities document.

504 Gateway Time-out
nginx/1.24.0

curl -o /tmp/cap.xml -w "状态 %{http_code} 大小 %{size_download} 耗时 %{time_total}s\n" \
  "http://127.0.0.1:8080/geoserver/wms?SERVICE=WMS&VERSION=1.1.1&REQUEST=GetCapabilities"

• Adding RAM or CPU: the cost is document volume, not compute.
• Raising the client timeout to 120 seconds, which turns a slow layer tree into a two-minute one.