Skip to main content
Nerve Centre KPIs · Audit Profile · Sentiment Settings CockroachDB-specific health audit for distributed-SQL clusters (Self-Hosted / Dedicated / Serverless). Answers six questions: (1) is SQL + Admin HTTP API access correctly provisioned (VIEWACTIVITY privilege, TLS verify-full, cluster routing); (2) are all nodes live and is the cluster reachable; (3) is statement performance within band (p95 / p99 latency, slow-query rate, distributed-transaction contention); (4) are ranges fully replicated with leases balanced and Raft lag low; (5) is storage and connection capacity within safe headroom; and (6) are managed / scheduled backups recent and durable. Cross-platform area joins statement volume and pool saturation to a commerce sibling to size live revenue at risk.

​
What this audit checks

​
Authentication & access

  • Connection string uses sslmode=verify-full with a valid sslrootcert (not disable / require)
  • SQL user holds the VIEWACTIVITY system privilege so crdb_internal stats tables are readable
  • For CockroachDB Cloud, the options=--cluster=<name> routing param is present and resolves
  • Admin HTTP API URL (port 8080) reachable with its own basic + TLS cert for node + range KPIs

​
Connection & availability

  • SHOW NODES returns and every node reports status=live (no DEAD / SUSPECT / DRAINING)
  • node_count matches the expected cluster size (a drop = node lost)
  • crdb_internal.gossip_nodes confirms all nodes gossiping and reachable
  • Statement error rate over the last 5m below 1% (connection refusals, retries surfacing as errors)

​
Query performance (p95 / slow queries)

  • Statement latency p95 below 200ms from crdb_internal.node_statement_statistics
  • Statement latency p99 below 500ms (tail latency from distributed fan-out)
  • Slow-query rate below 5% over a 15m window
  • Contended statements below 10 contention events / 24h via crdb_internal.cluster_contention_events

​
Replication & lag

  • Unavailable ranges = 0 (any >0 means a range lacks quorum and data is unreadable)
  • Under-replicated ranges = 0 sustained over 5m (below configured replication factor)
  • Range lease balance skew below 25% across nodes (no single hot leaseholder)
  • Raft quiescent lag below 10s; no decommissioning node stuck draining beyond 1h

​
Storage & capacity

  • Database disk usage below 90% from the /_status/vars Prometheus metrics
  • Memory usage below 85% per node
  • Connection pool saturation below 90% (in_use vs max_connections)
  • Transaction retries below 1000 / 24h (high retries = contention hotspot eating capacity)

​
Backups & durability

  • Last successful BACKUP TO S3 / GCS (or Cloud managed backup) within 72h
  • A scheduled backup job exists and its most recent run did not fail
  • Backup target is a durable off-cluster store, not a node-local path

​
Cross-platform: revenue protection

  • Statement-volume spike with no matching ecom order spike (sibling = bigcommerce/shopify.orders_per_15m) signalling runaway or anomalous load
  • Connection pool >90% saturation during a commerce traffic burst (sibling = checkout/order rate) risking dropped sessions
  • Slow statements co-occurring with a checkout-completion drop in the same 5m window (sibling = shopify/bigcommerce.checkout)
  • Inventory-table row drift vs ecom inventory by SKU (sibling = product_inventory) flagging stale stock writes

​
Severity thresholds

​
Data sources

  • GET postgresql://{host}:26257/{database} :: crdb_internal.cluster_settings - Cluster id, version, and configured settings for the DatabaseInstance entity
  • GET postgresql://{host}:26257/{database} :: crdb_internal.node_statement_statistics - Per-statement latency (p50/p95/p99), rows read, slow-query detection
  • GET postgresql://{host}:26257/{database} :: crdb_internal.cluster_contention_events - Distributed-transaction contention events for contended-statement checks
  • GET postgresql://{host}:26257/{database} :: crdb_internal.ranges - Range replication state, unavailable + under-replicated counts, lease distribution
  • GET postgresql://{host}:26257/{database} :: crdb_internal.gossip_nodes - Live node gossip state and reachability
  • GET postgresql://{host}:26257/{database} :: SHOW NODES - Node status (live / dead / draining), leaseholder counts, replica balance
  • GET https://{host}:8080/_status/vars - Admin HTTP API Prometheus metrics - disk, memory, qps, node + range gauges