Apache Druid Dashboard

This dashboard covers broker and coordinator nodes: query latency, CPU cost, JVM heap, GC frequency, and HTTP thread pool utilization.

Dashboard Preview

What This Dashboard Monitors

• Query health: Track broker query time and CPU cost per scan query. Spikes in query time with flat CPU point to slow historical nodes; rising CPU with flat query time points to CPU-bound processing on the broker.
• JVM heap pressure: Watch used heap approach committed and max across broker and coordinator. A narrowing gap between used and max signals GC pressure building.
• GC activity: Monitor G1 old-gen and young-gen collection counts and CPU time. Frequent old-gen collections on the broker degrade query throughput; on the coordinator they delay segment assignment.
• Thread pool saturation: Broker and coordinator Jetty thread pools show busy, idle, and total counts. When busy approaches total, the node starts queuing or rejecting HTTP requests.
• Coordinator duty cycle: Track how long each coordinator duty group (Historical Management, Indexing Service, Metadata Store) takes per cycle. Sustained increases indicate segment assignment backlog or metadata store slowness.

Metrics Included

KPI Panels

• Broker Heartbeat: Heartbeat count — shows the broker is up and sending metrics.
• Broker Open Connections: Open Jetty connections on the broker.
• Broker Thread Pool Utilization: Fraction of broker Jetty threads in use. Red above 0.8.
• Coordinator Duty Time: Latest coordinator management cycle duration.

Query Performance

• Broker Query Time: Scan query latency on the broker.
• Broker Query CPU Time: CPU cost per scan query. High CPU with low query time points to CPU-bound processing.

JVM Health

• Broker JVM Heap Memory: Used, committed, and max heap for the broker. Used approaching max signals GC pressure.
• Coordinator JVM Heap Memory: Used, committed, and max heap for the coordinator.

GC Activity

• Broker GC Count: G1 old-gen and young-gen collection cycles per interval on the broker.
• Broker GC CPU Time: CPU time GC consumes per interval. Rising old-gen means GC is stealing CPU from queries.
• Coordinator GC Count: G1 GC cycles per interval on the coordinator. Elevated old-gen often precedes segment assignment slowdowns.

Thread Pool & Coordinator Duties

• Broker Jetty Thread Pool: Busy, idle, and total HTTP threads on the broker. Busy approaching total means the broker is close to rejecting requests.
• Coordinator Jetty Thread Pool: Busy, idle, and total HTTP threads on the coordinator.
• Coordinator Global Duty Times: Time each duty group takes per cycle: Historical Management, Indexing Service, and Metadata Store. Sustained spikes indicate segment assignment backlog.

Dynamic Variable

A service variable (service.name) is pre-set to apache-druid. If you send Druid metrics under a different service name, update it in Dashboard Settings → Variables.

Next Steps

• Explore the Apache Druid metrics setup guide to configure metrics collection
• Set up alerts for query performance and JVM health.
• Learn about querying metrics to create custom visualizations