Production Topologies
Single region, multi-AZ baseline
- 3+ master-eligible data nodes.
- At least one replica for critical tables.
- Zone-aware node spread and shard allocation awareness.
Large analytics cluster
- Separate ingest-heavy and query-heavy workloads by node sizing.
- Increase replicas for read-heavy dashboards.
- Partition large fact tables by time.
Hybrid operational + AI/search workload
- Co-locate structured rows with vectors and text for hybrid search.
- Use
FLOAT_VECTOR+MATCHpipelines for relevance blending. - Blend
FLOAT_VECTOR+FULL TEXT SEARCH+GRAPHfor complete mapping of embeddings + lexical nature + relationships. - Keep query memory protections (breakers) tuned.
Docker Swarm
MonkDB runs in Swarm, but stateful scheduling and volume guarantees need strict ops discipline.
- Prefer node labels for deterministic placement.
- Ensure persistent volumes follow service constraints.
- Control rolling updates to avoid shard churn.
Recommended production defaults
- JVM heap explicitly set.
- TLS enabled for HTTP/PGWire/transport as required.
- Snapshot repositories configured early.
- Monitoring stack enabled before go-live.