Skip to content

GRAFANA_SETUP.md

Latest commit

 

History

History
112 lines (84 loc) · 2.87 KB

GRAFANA_SETUP.md

File metadata and controls

112 lines (84 loc) · 2.87 KB

Grafana + Prometheus Integration

RollingStone now exports metrics to Prometheus for professional visualization in Grafana.

Quick Start

1. Install (One-Time Setup)

brew install prometheus grafana

2. Start Everything

./start.sh

This automatically starts:

3. Set Up Grafana Dashboard

  1. Open http://localhost:3000
  2. Login: admin / admin (change password when prompted)
  3. Go to Dashboards → Import
  4. Upload grafana-dashboard.json
  5. Select "Prometheus" as data source
  6. Click Import

4. View Metrics

Open the "RollingStone LSM Simulator" dashboard to see:

  • L0 file count over time
  • Write/Read/Space amplification
  • Disk utilization
  • LSM size growth
  • Write stall indicators

Running Without Prometheus/Grafana

If you don't have Prometheus/Grafana installed, ./start.sh will still work:

Why This Solves the Charting Problem

Browser-based charts failed at 100× speed:

100× speed = 200 data points/second
After 10 minutes: 120,000 points in browser memory
Result: Memory leak, UI freeze

Prometheus + Grafana handles it:

Prometheus: Separate process, built for time-series
Grafana: Queries only visible timerange, auto-downsamples
Result: Smooth charts even at 1000× speed

Available Metrics

rocksdb_l0_files                    # Number of L0 files
rocksdb_write_amplification         # Write amplification factor
rocksdb_read_amplification          # Read amplification (files checked)
rocksdb_total_size_mb               # Total LSM size in MB
rocksdb_disk_utilization_percent    # Disk utilization (0-100%)
rocksdb_is_stalled                  # Write stall state (0=normal, 1=stalled)

Querying in Prometheus

Open http://localhost:9090 and try queries like:

rocksdb_l0_files                           # Current L0 count
rate(rocksdb_total_size_mb[1m])            # Growth rate
rocksdb_write_amplification > 5            # High write amp alert

Grafana Tips

  • Auto-refresh: Set to 5s for live updates
  • Time range: Last 5m for recent data, Last 1h for trends
  • Annotations: Add markers for config changes
  • Alerts: Set thresholds (e.g., disk > 80%)

Stopping Everything

curl http://localhost:8080/quitquitquit  # Stops all three processes

Or manually:

kill <PROM_PID> <GRAF_PID> <SIM_PID>

Data Persistence

  • Prometheus data: Stored in ./prom-data/ (survives restarts)
  • Grafana dashboards: Stored in /opt/homebrew/var/lib/grafana/
  • Simulator data: In-memory only (resets on restart)

To clear Prometheus history:

rm -rf prom-data/