Why should you care?

PostgresVerse payment cluster needs pager when replication lag >30s or disk >85% not email next morning.

See it live — copy this example

Run in pgAdmin or psql.

-- Metrics-friendly queries (postgres_exporter compatible)
SELECT count(*) FILTER (WHERE state = 'active') AS active_sessions
FROM pg_stat_activity WHERE datname = 'PostgresVerse';

SELECT extract(epoch FROM (now() - pg_last_xact_replay_timestamp())) AS replay_lag_sec;

What happened?

• active_sessions gauge for Grafana.
• replay_lag_sec on standby feeds SLA dashboard.
• Exporter scrapes these on interval; Alertmanager fires thresholds.

Practice next

1. Deploy postgres_exporter pointing at PostgresVerse.
2. Import Grafana dashboard ID for postgres.
3. Set alert replication lag and connections >80% max.
4. Add custom metric n_dead_tup from pg_stat_user_tables for orders.
5. Synthetic canary query every 60s from outside VPC.

Remember

Metrics + logs + traces = full picture. Alert on lag, disk, connections, deadlocks. Runbooks linked from pager.

PostgresVerse NOC dashboard

Grafana wall shows lag, QPS, cache hit, top query; on-call resolves before users notice.

Outcome: MTTR under 15 min for DB incidents last quarter.