database-observability

This skill sets up Grafana Cloud Database Observability for MySQL and PostgreSQL, giving query-level insight — RED metrics, sampled query texts, and visual explain plans — without application code changes. It solves the practical problem of diagnosing slow queries and monitoring database performance across self-managed and managed instances (RDS, Aurora, Cloud SQL, Azure Database), including cases where a user simply says "my database is slow" or "monitor RDS".

The skill walks through concrete setup workflows. For PostgreSQL it enables `pg_stat_statements` (via `shared_preload_libraries` and a restart), creates a least-privilege monitoring user granted the `pg_monitor` role and connect rights, and enables the extension per database. For MySQL it creates a least-privilege user with SELECT/PROCESS/REPLICATION CLIENT and performance_schema access, and verifies Performance Schema is on. It then configures the `database_observability.postgres` / `database_observability.mysql` Grafana Alloy components with selectable collectors (pg_stat_statements, query_samples, schema_details, explain_plans), forwarding metrics and logs to Grafana Cloud using credentials pulled from environment variables. Reference files provide the full Alloy config, key PromQL queries, alert-rule YAML for slow queries, error rate, and connection saturation, and trace-correlation setup that links slow spans to specific SQL via OpenTelemetry `db.statement` / `db.system` / `db.name` attributes.

Target users are SREs, DBAs, and platform engineers running MySQL or PostgreSQL who want query-level observability, APM correlation, and latency alerting in Grafana Cloud. This is defensive observability tooling that emphasizes least-privilege access and standard credential handling via environment variables.