Grafana Learn
Sign in Sign up
Section 2 · Set up Database Observability

Architecture and data flow for Database Observability

Estimated time: 2 min

How the data flows

Database Observability architecture

Grafana Alloy runs alongside your database and reads its built-in statistics. Where those statistics come from depends on the database:

  • PostgreSQL: pg_stat_statements for aggregate query statistics and pg_stat_activity for active sessions. The pg_stat_statements extension must be installed and enabled.
  • MySQL: Performance Schema, which is enabled by default in MySQL 8.0 and later

Alloy splits that data into two streams and forwards each to a different Grafana Cloud backend:

  • Metrics (query counts, latency, errors, and resource usage) are scraped by database exporters and sent to Grafana Cloud Metrics.
  • Logs (query samples, explain plans, schema details, and wait events) are collected by the Database Observability components and sent to Grafana Cloud Logs.

Grafana Cloud provides Database Observability dashboards that bring the metrics and logs together into a single view of query performance, starting with the Queries Overview dashboard.

Database Observability is designed to minimize impact on your production databases. To capture an explain plan, it runs EXPLAIN, not EXPLAIN ANALYZE, so the query itself does not run. That has minimal overhead for most queries.

When you collect more than you need

Query samples, schema details, and wait events are sent to Grafana Cloud Logs along with the metrics. If that is more than you need, Requirements and overhead lists three changes:

  • Raise collect_interval, for example from 1 minute to 5 minutes.
  • Lower explain_plans.per_collect_ratio, for example to 0.5 or lower.
  • Exclude schemas you don’t need (exclude_schemas on MySQL) or databases you don’t need (exclude_databases on PostgreSQL).

Tune Alloy collection has a high-volume example of those same settings.