Skip to main content
Nerve Centre KPIs · Audit Profile · Sentiment Settings Supabase-specific database health audit. Answers six questions: (1) is project access correctly keyed - service_role kept server-side, Management API PAT scoped, RLS in force on the anon key; (2) is the project reachable and the PostgREST / Realtime layer up; (3) is query performance within band - PostgREST p95, Postgres p95, and the pg_stat_statements slow-query tail; (4) are read replicas streaming and is lag bounded; (5) is the Supavisor pool and disk under their tier caps; (6) are daily backups / PITR current and recoverable. Cross-channel area joins pool saturation and slow PostgREST queries to commerce-sibling checkout traffic to size live revenue at risk.

​
What this audit checks

​
Authentication & access

  • service_role JWT present and handled server-side only (never shipped to client / anon contexts)
  • Management API PAT scoped to the org and able to read /v1/projects/
  • Row-Level Security enabled on tables exposed through the anon (public) key
  • Supavisor pooler connection string valid (transaction mode 6543 / session mode 5432) and reachable

​
Connection & availability

  • Project /v1/projects/ returns ACTIVE_HEALTHY (not paused / restricted / read-only)
  • PostgREST /rest/v1 reachable and returning 2xx on a probe request
  • Realtime tenant health endpoint /realtime/v1/api/tenants//health reports connected
  • Project not in disk-full read-only mode (tier disk cap not breached)

​
Query performance

  • PostgREST API latency p95 below 200ms and p99 below 500ms
  • Underlying Postgres query latency p95 below 200ms
  • Slow-query rate below 5% over a 15m window (pg_stat_statements mean_exec_time)
  • Buffer cache hit rate at or above 95% (low hit rate signals undersized cache / missing indexes)

​
Replication & lag

  • Every read replica in STREAMING state via pg_stat_replication (no RECOVERING / BROKEN / STOPPED)
  • Read-replica lag below 10s so read-routed PostgREST queries stay current
  • Replica region set is intentional and each regional replica is reporting last_check_at recently

​
Storage & capacity

  • Database disk usage below 90% of the tier cap (cap breach forces read-only mode)
  • Supavisor pool saturation below 90% of the tier connection cap
  • Connections in use below max_connections with a non-growing Supavisor wait_queue
  • Memory usage below 85% and deadlock count zero over the last 5m

​
Backups & durability

  • Last daily backup age below 72h (Pro+; Free tier has no managed backup - flag the gap)
  • Point-in-Time Recovery enabled on tiers that support it and within retention window
  • Edge Function error rate below 2% so backed write paths are not silently failing

​
Cross-channel: revenue at risk

  • Supavisor pool saturation > 90% concurrent with a commerce-sibling traffic burst (shopify/bigcommerce/adobe.order rate)
  • Slow PostgREST queries co-occurring with a sibling checkout-completion drop in the same 5m bucket
  • Supabase QPS spike with no matching ecom order spike (signals retries / runaway client, not real demand)
  • Products-table row count drifting from the sibling ecom catalog SKU count when Supabase is the source of truth

​
Severity thresholds

SignalWarnCritical
connection_error_rate15
query_p95_ms200500
replication_lag_sec1060
disk_usage_pct8090
slow_query_count520

​
Data sources

  • GET https://{project_ref}.supabase.co/rest/v1/rpc/pg_stat_statements - pg_stat_statements via service_role RPC - slow-query tail, mean_exec_time
  • GET https://{project_ref}.supabase.co/realtime/v1/api/tenants/{ref}/health - Realtime tenant health - channel / connection status
  • GET https://{project_ref}.supabase.co/rest/v1/ - PostgREST availability + latency probe
  • GET https://api.supabase.com/v1/projects/{ref} - Management API (PAT) - project status, tier, region, max_connections
  • GET https://api.supabase.com/v1/projects/{ref}/analytics/endpoints/logs.all - Project log analytics - error rates, PostgREST 5xx, request volume
  • GET pg_stat_activity - Direct Postgres via Supavisor - connections in use, wait queue, pool state
  • GET pg_stat_replication - Direct Postgres via Supavisor - replica role, state, lag_seconds