Health dashboard
The Health tab is the main system-level diagnostics page. It starts with a small set of summary cards, then shows detail sections for specific risks.
Use it when you want to answer questions like:
- Is the instance under checkpoint or I/O pressure?
- Are long transactions blocking cleanup?
- Is wraparound risk building up?
- Are replication slots or statement statistics becoming a problem?
Top-level cards
| Card | What it means |
|---|---|
| Cache hit ratio | Whether reads are mostly served from cache or are falling through to disk more often than expected. |
| Connections | Current connection usage relative to the configured ceiling. |
| Oldest transaction | Whether a long-running transaction may be blocking vacuum progress or tuple cleanup. |
| TXID wraparound risk | Whether transaction ID age or multixact age is approaching a dangerous range. |
The UI shows the current value, severity, and explanation. The exact internal scoring thresholds are not documented in detail because they may change as the product logic improves.
Detail sections
- Checkpoint stats show timed vs requested checkpoints, checkpoint write/sync time, and related pressure signals. On managed providers, checkpoint frequency may partly reflect provider activity rather than only application writes.
- Sequences at risk show sequences that are getting close to their maximum value.
- Replication slots show WAL retained by non-temporary slots. This matters because an inactive or stalled slot can keep WAL files on disk for too long.
- pg_stat_statements capacity shows whether the statement statistics store is evicting entries before they remain useful.
- bgwriter maxwritten shows whether the background writer is repeatedly hitting its configured limit.
- Vacuum horizon blockers show sessions with old snapshots that can hold back dead tuple cleanup.
Failure readiness
Two summary cards estimate recovery posture:
- Crash Recovery (RTO) is a rough signal for how much recovery work may be needed after a crash.
- Data Safety (RPO) is a rough signal based on replica state, sync mode, and WAL retention hazards.
These are operational summaries, not a substitute for tested disaster recovery procedures.
Snooze
Findings can be snoozed locally so accepted conditions do not keep interrupting the current investigation.
Caching
Health data is short-lived cached. Existing content stays visible while a fresh read is loaded in the background.