Profiling
Profiling is a way to diagnose performance problems with your server. A profiler measures certain characteristics of a running server, e.g. how often a method is called and how much time it takes up in a single tick, how long and often the garbage collector runs and much more.
For profiling to be effective, the issue you are diagnosing must be actively occurring.
If you need help with analyzing a performance issue, please bring a spark report to the PaperMC Discord server (#paper-help) for assistance.
spark
Starting with 1.21, Paper bundles the spark profiler, which is the preferred way to profile Paper.
To start profiling your server, run this command:
/spark profiler start --timeout 600
After 10 minutes, this will return a URL to a profiler report, which you can analyze yourself or provide to a developer of a plugin or the Paper support chats.
This is only a fraction of what spark can do, so if you want to learn about the different features of spark or learn to analyze reports yourself, check out spark's documentation here.
If you want to use a version of spark newer than the bundled one, simply place a standalone spark plugin JAR
into the plugins directory and set the paper.preferSparkPlugin
system property to true.
Timings
Paper also bundles the Timings v2 profiler, however Timings has been unmaintained for multiple years and its reports are difficult to read for beginners. It has been deprecated in favor of spark and turned off by default in 1.21, see this discussion for more information.