Before you try the hands-on learning path
If you decide to install and explore Database Observability, it’s important to keep in mind to start from the Configuration page.
The most reliable way to avoid setup problems is to let Grafana Cloud generate your Alloy configuration for you. The Database Observability Configuration page builds the config for your database and deployment, so you don’t have to assemble the components by hand.
If you hand-write or edit the config, the listed mistakes are the ones that most often stop data from arriving.
Where setup usually goes wrong
| Pitfall | What happens | What to check |
|---|
| Incomplete Alloy config | The database_observability.mysql or database_observability.postgres component is missing its required data_source_name (the database connection) or its forward_to destination, so nothing is collected. | Both data_source_name and forward_to are set on the component. |
| Missing endpoints | The config has no prometheus.remote_write for metrics or loki.write for logs, so telemetry is collected but never reaches Grafana Cloud. | Both write blocks point at your Grafana Cloud Metrics and Grafana Cloud Logs endpoints with valid credentials. |
| Inconsistent job label | The job label doesn’t use the value from the generated config (integrations/db-o11y), or the metrics and logs relabel rules don’t match, so your data isn’t labeled the way Database Observability expects. | Your discovery.relabel and loki.relabel blocks both set job to integrations/db-o11y, and their rules stay consistent with each other. |
| Database side not prepared | Performance Schema (MySQL) or the pg_stat_statements extension (PostgreSQL) isn’t enabled, so the built-in statistics Alloy reads don’t exist. | The prerequisites for your database are in place before Alloy starts. |
Confirm it’s working before you troubleshoot
After Alloy is running, the Configuration page runs status checks for your instance. If data isn’t showing up, start there. A failing check usually points straight at one of the gaps above, which is faster than searching the dashboards for data that never arrived.
