Skip to main content
Nerve Centre KPIs · Audit Profile · Sentiment Settings MariaDB-specific health audit for MySQL-stack teams running 10.6 / 10.11 / 11.x, often clustered via Galera for synchronous multi-master replication. Answers six questions: (1) is access scoped correctly and is the stats path (performance_schema + REPLICATION CLIENT) actually readable; (2) is the instance up, are connections being accepted, and is the Galera cluster in a Primary component; (3) is p95 query latency in band and which digests entered the top-10 slow list; (4) are async replicas caught up and is Galera flow-control quiet; (5) is disk and buffer-pool headroom sufficient; (6) is a recent restorable backup present. Cross-channel area joins query-volume and pool-saturation signals to commerce-sibling order and checkout funnels to size live revenue at risk.

​
What this audit checks

​
Authentication & access

  • Connection succeeds with the supplied connection_string + credentials (SSL-mode REQUIRED honoured on managed endpoints)
  • Bound user holds SELECT on performance_schema and information_schema (stats KPIs return rows, not access-denied)
  • Bound user holds REPLICATION CLIENT + PROCESS so SHOW REPLICA STATUS and processlist are readable
  • performance_schema is enabled (events_statements_summary_by_digest populated, not empty)

​
Connection & availability

  • Instance reachable and uptime_seconds advancing (no recent unplanned restart within the window)
  • Galera cluster: wsrep_cluster_status is Primary on every node (non-Primary = split-brain, node refuses writes)
  • Galera cluster: wsrep_cluster_size equals the expected node count (node loss = quorum risk)
  • Aborted_connects over 24h within band - spikes signal auth churn or TLS handshake failures

​
Query performance

  • Query latency p95 under the 200ms threshold; p99 under 500ms
  • Slow-query rate (15m) under 5% of statements
  • Top-10 slow digests from events_statements_summary_by_digest captured with rows_examined vs rows_returned ratio for index review
  • InnoDB deadlocks in the last 5m are zero (any deadlock is flagged)

​
Replication & lag

  • Async replication lag (Seconds_Behind_Master) under 10s on each active replica
  • Every replica is STREAMING (state not in RECOVERING / BROKEN / STOPPED)
  • Galera flow-control paused fraction under 10% over 5m (high = one slow node throttling the cluster)
  • At least one healthy standby / SYNCED node is failover-ready

​
Storage & capacity

  • Database disk usage under the 90% threshold with projected days-to-full headroom
  • InnoDB / XtraDB buffer pool hit rate at or above 95% (low = pool starved, reads hitting disk)
  • Connection pool saturation (Threads_connected vs max_connections) under 90%
  • Instance memory usage under 85%

​
Backups & durability

  • Last successful mariabackup / Percona XtraBackup run under 72h old
  • Backup is restorable (non-zero size, completion marker present, not a partial / interrupted run)
  • Binlog retention covers the gap between full backups for point-in-time recovery

​
Cross-channel: revenue at risk

  • QPS spike with no matching order spike (sibling = bigcommerce/shopify/adobe.orders_per_15m flat while mariadb.qps_15m surges = bot / scraper load)
  • Pool saturation across Galera nodes >90% sustained during an ecom traffic burst (= capacity exhausted cluster-wide at checkout)
  • Slow queries co-occurring within a 5m checkout window where sibling checkout_completion dropped >5pp
  • Inventory-table row drift vs sibling product_inventory count (SKUs out of sync between MariaDB and the storefront)

​
Severity thresholds

SignalWarnCritical
connection_error_rate0.51
query_p95_ms200500
replication_lag_sec1030
disk_usage_pct8090
slow_query_count520
galera_flow_control_paused_pct1025
buffer_pool_hit_rate_pct95-
backup_age_hours4872

​
Data sources

  • GET mariadb://{host}:{port}/{database} - Native MariaDB protocol connection (MySQL wire-compatible) - base for all stats queries
  • GET SHOW VARIABLES - Instance identity, version, max_connections, configuration baseline
  • GET SHOW GLOBAL STATUS - Uptime, Threads_connected/running, Aborted_connects, QPS, buffer-pool counters, deadlocks
  • GET performance_schema.events_statements_summary_by_digest - Slow-query digests, p50/p95/p99 latency, rows_examined vs rows_returned
  • GET performance_schema.processlist - Live connection pool: size, in_use, idle, wait queue, app origin
  • GET SHOW REPLICA STATUS - Async replica role, Seconds_Behind_Master, IO/SQL thread state
  • GET information_schema.GLOBAL_STATUS WHERE LIKE 'wsrep_%' - Galera quorum: wsrep_cluster_size, wsrep_cluster_status, wsrep_local_state, wsrep_flow_control_paused