Observability
Observability is the operational depth behind “the server feels bad”. It shows whether a problem is the tick loop, memory pressure, garbage collection, or the data pipeline itself.
Core metrics
Section titled “Core metrics”Each server reports CPU, RAM, disk, TPS, MSPT, players, and latency. Metrics are stored with average, p95, min, and max aggregations, and the dashboard exposes granularities from 1 minute to daily.
| Metric | Unit | Watch for |
|---|---|---|
| TPS | ticks/s | Sustained drops below 19. |
| MSPT | ms | Median and p95 rising together. |
| CPU | % | Process versus system share. |
| Memory | % | Process, physical, and swap. |
| Latency | ms | Player-facing ping average and p95. |
| Disk | MB | Growth and instance size. |
Health bands
Section titled “Health bands”These bands match the platform’s alert defaults and are a reasonable starting point:
| Metric | Warning | Critical |
|---|---|---|
| TPS | below 19 | below 17 |
| MSPT | above 40 ms | above 50 ms |
| CPU | above 70% | above 90% |
| Memory | above 75% | above 90% |
| Latency | above 100 ms | above 200 ms |
JVM tabs
Section titled “JVM tabs”Server observability includes JVM-specific views:
- Runtime: thread count, peak and daemon threads, GC collections, total GC time, average pause, and open versus max file descriptors.
- Memory pools: used bytes per pool, with a pool selector.
- GC events: an event table, histogram, and scatter view with per-event names and filters.
Use these together: rising memory with more frequent GC pauses usually points at heap pressure, while high CPU with stable memory points at tick work.
Pipeline health
Section titled “Pipeline health”Pipeline health tells you whether data itself is flowing: batches sent and failed, data points sent, queue depth, authentication blocks, and connection failures. If a chart looks stale, check here before assuming the server is idle.
Network interfaces
Section titled “Network interfaces”The network interfaces view reports RX/TX per interface, with an interface selector, so host-level traffic can be compared with player-facing latency.
Scope and timing
Section titled “Scope and timing”Observability is server-scoped; fleet-level health appears on the command center and operations home. Metrics arrive after collection and processing, so a chart can lag the live server by a short interval.