The memory pressure alert

Memory pressure is the warning before a memory crisis. It fires when your heap is getting full but before performance has necessarily degraded, giving you a window to act before your players notice anything.

What triggers it

The alert fires when heap usage exceeds 85% of the configured maximum. At this level, the JVM's garbage collector is working harder to keep memory free, which increases GC pause frequency and can begin affecting TPS. An out of memory error that crashes the server is typically preceded by heap staying above 90 to 95% for several minutes.

What the alert contains

The email includes:

• Current heap usage as a percentage of max and in GB
• GC pressure, the cumulative GC pause time in the last heartbeat window
• Current entity count across all worlds
• Estimated time to 90% ceiling at the current rate, from linear regression on recent heap data
• A link to your dashboard's memory chart

The time-to-ceiling estimate is a rough projection based on recent trend. A server that had a sudden spike in entity count will show a short projection; a server with a slow steady leak will show a longer one.

What to do when it fires

In order of how often each is the cause:

• Entity count: check the entity breakdown panel. Mob farms and item overflow are the most common cause of sudden memory pressure
• Plugin memory leak: a plugin retaining references to objects it should have discarded. Restarting the server will confirm whether the leak is in a plugin, as memory drops and then climbs again at the same rate after restart
• Heap too small: if memory pressure is chronic and TPS is fine, your -Xmx value is simply too low for your current player count and plugin load. Increase it
• GC tuning: on very large heaps, switching from G1GC to ZGC can eliminate stop-the-world pauses at the cost of slightly higher CPU usage