trino-observability-platform

What problem does it solve?

Trino clusters can silently degrade due to queuing, latency spikes, memory pressure, or OOM events, making it hard to detect issues early and diagnose root causes fast.

Core Features & Use Cases

• Prometheus metrics from Trino JMX: Export running/queued/failed queries, execution latency percentiles, memory pool usage, worker heartbeat health, and JVM metrics for alerting and trend analysis.
• Grafana dashboard panels for fast triage: Visualize query activity, latency (P50/P90/P99), free vs total memory, active workers, OOM kill rate, and input throughput to support SLA dashboards and incident investigations.
• OpenTelemetry trace context propagation: Link orchestration spans to Trino queries via session headers for end-to-end tracing across your data platform.
• Structured query logging via a Trino Event Listener: Emit JSON records for slow or failed queries with query text, timings, memory, and failure details to enable high-signal investigations.
• Prometheus alert rules and REST-based query history analysis: Detect worker loss, queue depth, high P99 latency, failure rate anomalies, and memory pressure while supporting slow-query discovery via Trino’s API.

Quick Start

Configure the Trino JMX Prometheus exporter, wire it into Prometheus with scrape targets for coordinator and workers, and import the provided Grafana panels to get immediate visibility into query latency, queue depth, and memory pressure.