Skip to main content
Nerve Centre KPIs · Audit Profile · Sentiment Settings MongoDB-specific health audit for self-managed and Atlas clusters. Answers six questions: (1) is the monitoring user scoped to clusterMonitor and is SCRAM-SHA-256 / TLS auth correct for the connection string; (2) is the instance reachable and are connection errors and pool saturation under control; (3) is query latency (p95 / p99) within band and are slow ops + COLLSCAN operations climbing; (4) is the replica set healthy with secondaries keeping up and elections quiet; (5) is disk and WiredTiger cache capacity within safe headroom; (6) is a recent successful backup or snapshot in place for durability. Signals are read from serverStatus, db.stats, the profiler, rs.status, and sh.status.

​
What this audit checks

​
Authentication & access

  • Connection string uses mongodb+srv:// (Atlas) or mongodb:// with replica set seed list, not a single bare host
  • Monitoring user holds the clusterMonitor built-in role plus read on the monitored DBs (slow-op KPIs unusable otherwise)
  • SCRAM-SHA-256 auth in effect (MongoDB 4.0+ default) and TLS enforced for Atlas connections
  • Atlas Admin API key (key ID + private key) present and project-scoped when cluster is Atlas-managed

​
Connection & availability

  • db.serverStatus() reachable and instance uptime present (no recent unexpected restart)
  • Connection errors over 24h below threshold (connections refused / network resets)
  • Connection pool saturation under 90% - connections.current / (connections.current + connections.available)
  • Active reader / writer queue (globalLock.activeClients) not backing up under load

​
Query performance

  • Query latency p95 under 200ms from serverStatus latencies.reads.latency / ops
  • Query latency p99 under 500ms (tail latency not masking p95)
  • Slow ops in trailing 15m under 10 - profiler entries with millis above slowms (default 100ms)
  • COLLSCAN operations over 24h under 10 - full collection scans signal missing indexes or an unindexed code path

​
Replication & lag

  • Every secondary replica lag under 10s from rs.status() optimeDate delta vs primary
  • No member stuck in RECOVERING / STARTUP2 / DOWN state (stateStr healthy across the set)
  • Elections over 24h at most 1 - frequent elections indicate primary flapping from network or hardware instability
  • Sharded clusters: chunk-balance skew under 20% and pending chunk migrations bounded (sh.status())

​
Storage & capacity

  • Database disk usage under 90% from db.stats() storage and Atlas capacity surface
  • WiredTiger cache hit rate at or above 95% - 1 - (bytes-read-into-cache / bytes-currently-in-cache)
  • WiredTiger dirty cache under 20% of configured maximum (above triggers eviction pressure)
  • Resident memory (mem.resident) within tier headroom for the Atlas instance class

​
Backups & durability

  • Last successful backup under 72h - mongodump, Atlas continuous backup, or snapshot
  • Atlas Cloud Backup enabled with a retention policy when cluster is Atlas-managed
  • Oplog window long enough to cover the backup cadence plus restore lead time
  • Write concern w:majority in effect for durability-critical writes (per connection string and app defaults)

​
Severity thresholds

SignalWarnCritical
connection_error_rate15
query_p95_ms200500
replication_lag_sec1030
disk_usage_pct8090
slow_query_count1025

​
Data sources

  • GET mongodb://{host}:{port}/{database} - Base connection - replica set or Atlas srv seed list
  • GET db.serverStatus() - opcounters, connections, latencies, globalLock, mem, WiredTiger cache
  • GET db.stats() - Storage size, data size, index size for disk usage
  • GET db.system.profile.find() - Slow-op profiler entries - requires setProfilingLevel(2) or slowms threshold
  • GET rs.status() - Replica set member states, optimeDate lag, election history
  • GET sh.status() - Shard balance, pending chunk migrations (sharded clusters only)
  • GET db.currentOp() - In-flight long-running operations and collection scans