ServerBenchmark

Spark support

- Works alongside Spark when it's installed. Pulls its 5s / 10s / 1m / 5m / 15m TPS, MSPT median and p95, server CPU use and per-collector GC stats into the report - Spark is optional, nothing changes if you don't have it - Can be turned off with use-spark: false in the config

More accurate results

- GC overhead is now measured since the last check instead of since boot, so servers with long uptime score correctly - Current TPS uses Spark's 10 second window when available instead of a 1 minute average - Shows how much CPU the server itself uses next to the whole machine, and flags when something else on the box is hogging it - Player capacity now accounts for real tick headroom when players are online - Disk read and write speeds use the real elapsed sample time

New commands

- /benchmark tps - instant TPS, MSPT, heap, players and entities with no sampling delay - /benchmark compare - diffs your last two results, green for better and red for worse - /benchmark monitor - shows the last background check

Background monitor

- Optional timer that runs a benchmark every few minutes, logs a one line summary and warns staff with serverbenchmark.alerts when issues show up - Off by default, enable under monitor: in config.yml

Report additions

- TPS trend row (1m / 5m / 15m) - Spark section with MSPT percentiles and CPU windows - Per-world chunk, entity and player breakdown - GC collector rows with average pause and frequency - Data source line so you know where the numbers came from - New issues and tips for recent lag, long GC pauses, small heaps, empty worlds with chunks loaded and main thread bound servers - JSON exports include worlds, Spark data and GC details