Something is eating your TPS. You check the player count, it is only 12 players. Should be fine. But TPS sits at 15 and blocks break with a delay. Without data, you are guessing. Spark stops the guessing.
What Spark Does
Spark is a performance profiler for Minecraft servers. It samples the server's main thread hundreds of times per second and builds a tree of where time is being spent. The result is a detailed report that shows you which plugins, systems, or game mechanics are consuming your tick budget.
Installing Spark
Download from spark.lucko.me. Place the .jar in plugins/ and restart.
Spark works on Paper, Spigot, Fabric, Forge, BungeeCord, and Velocity.
Basic Commands
| Command | What It Does |
|---|---|
| /spark tps | Shows current and recent TPS |
| /spark health | Shows TPS, CPU, RAM, and GC activity |
| /spark profiler start | Starts recording a profile |
| /spark profiler stop | Stops recording and generates a report URL |
| /spark profiler start --timeout 300 | Records for 5 minutes then auto-stops |
| /spark gc | Shows garbage collection statistics |
How to Record a Good Profile
- Wait for lag to happen (or trigger it by having players do what normally causes lag)
- Run
/spark profiler start - Wait at least 3-5 minutes with players online doing normal activities
- Run
/spark profiler stop - Open the URL that appears in chat
A 30-second profile is usually too short. You need enough time for patterns to show up.
Reading the Spark Report
The report opens in your browser at spark.lucko.me. Here is how to read it.
The Flame Graph
The flame graph shows stacked bars. The bottom bar is the main server tick. Each bar above it is a sub-task that runs inside the tick. The wider a bar, the more time it consumes.
What to look for:
- Very wide bars near the top: These are the specific operations consuming the most time
- Plugin names: If you see a plugin class name in a wide bar, that plugin is responsible for the load
- Minecraft internal classes: Names like
ServerChunkCache,EntityLiving,HopperBlockEntitytell you which game system is the bottleneck
Common Findings
EntityLiving.travel or EntityLiving.tick taking 30%+ of tick time: Too many entities. Reduce mob spawn limits or increase merge radius.
HopperBlockEntity.tryMoveItems taking 15%+ of tick time: Hopper networks are too large. Increase hopper transfer cooldown.
ChunkProviderServer.getChunkFutureMainThread taking 20%+ of tick time: Chunk loading is overwhelming the server. Pre-generate terrain with Chunky or reduce view distance.
A specific plugin name taking 10%+ of tick time: That plugin is a performance problem. Check its configuration, update it, or find an alternative.
GC (garbage collection) pauses showing in the health report: You are allocating too much or too little RAM. GC pauses above 50ms cause noticeable lag spikes all at once.
Step-by-Step Example
You profile your server and the Spark report shows:
100% Server thread
├── 35% Chunk ticking
│ ├── 20% Entity ticking
│ │ └── 15% Monster pathfinding
│ └── 12% Hopper transfers
├── 25% Plugin: ShopGUI+
│ └── 22% Database query
├── 15% Chunk loading
└── 25% Everything else
Reading this:
- Monster pathfinding uses 15%: Too many monsters. Lower spawn limits in bukkit.yml
- Hopper transfers use 12%: Hopper networks need throttling. Set hopper-transfer to 8 in spigot.yml
- ShopGUI+ uses 25%: The shop plugin is making slow database queries every tick. Switch from SQLite to MySQL, or reduce the plugin's update frequency in its config
- Chunk loading uses 15%: Players exploring new terrain. Pre-generate with Chunky
After making these changes, profile again. Compare the new report with the old one. You should see those percentages drop.
Memory Profiling
Spark also profiles memory allocation:
/spark heapdump
This creates a snapshot of what is in memory. Open the generated file with VisualVM or Eclipse MAT to see which objects use the most RAM.
For most server owners, the health command is sufficient:
/spark health
This shows:
- Current memory usage and max allocation
- GC frequency and pause duration
- CPU usage
- TPS over time
Sharing Spark Reports
Spark reports are hosted on spark.lucko.me with a unique URL. You can share this URL with:
- Plugin developers (to show them their plugin is slow)
- Hosting support (to diagnose hardware issues)
- Community forums (to get optimization help)
The report contains no sensitive data. It shows class names and timing data, not player information or passwords.
When to Profile
- After adding a new plugin
- When TPS drops below 18 during normal play
- After a Minecraft version update
- Before and after optimization changes (to measure improvement)
- Whenever players report lag and you cannot identify the source
Make profiling a regular habit. Profile once a month during peak hours even if nothing seems wrong. You might catch a slow leak before it becomes a crisis.
Space-Node servers support Spark on all plans. Install it through the file manager and start profiling in minutes. Get started here.
