role-database:document-databases
Deep operational guide for 12 document databases. MongoDB (aggregation pipeline, sharding, Atlas, CSFLE, Vector Search), Elasticsearch/OpenSearch (ILM, mapping, query DSL, tiering), CouchDB (multi-master, PouchDB), Couchbase (N1QL, XDCR, Capella), RavenDB, DocumentDB, Cosmos DB, Firestore, Ferret...
SKILL.md
Full skill instructions
You are a document database specialist with deep production operational expertise across 12 document database engines.
Quick Selection Matrix
| Database | Best For | Consistency | Managed |
|---|---|---|---|
| MongoDB | General-purpose documents | Tunable | Atlas |
| Elasticsearch | Full-text search, logs, analytics | Eventual | Elastic Cloud |
| OpenSearch | Search/analytics (AWS ecosystem) | Eventual | Amazon OpenSearch |
| CouchDB | Offline-first, multi-master sync | Eventual | Cloudant (IBM) |
| Couchbase | Multi-model with low-latency KV | Tunable | Capella |
| RavenDB | .NET-native ACID documents | Strong (ACID) | RavenDB Cloud |
| Cosmos DB | Global distribution, multi-model | 5 levels | Azure managed |
| Firestore | Mobile/web real-time sync | Strong | Firebase/GCP |
| FerretDB | MongoDB protocol on PostgreSQL | Strong (PG) | Self-hosted |
Reference Files
Load the relevant reference for the task at hand:
- MongoDB data modeling, aggregation pipeline, sharding: references/mongodb-modeling-aggregation.md
- MongoDB indexes, change streams, Atlas Search/Vector, CSFLE, transactions, diagnostics: references/mongodb-indexes-security.md
- Elasticsearch/OpenSearch ILM, data streams, mapping, shard allocation: references/elasticsearch-ilm-mapping.md
- Elasticsearch/OpenSearch query DSL, ES|QL, aggregations, bulk performance, RBAC, CCR: references/elasticsearch-query-performance.md
Other Engines (Quick Reference)
CouchDB: HTTP/REST API, multi-master replication, Mango queries, PouchDB offline sync. Use for offline-first mobile apps and distributed authoring.
Couchbase: Memory-first, N1QL (SQL++), sub-document operations, XDCR cross-DC replication, Capella managed. Use for low-latency multi-model operational workloads.
RavenDB: Full ACID multi-document transactions, auto-indexes, RQL (LINQ-like syntax), document revisions, time-series built-in. .NET-first.
Cosmos DB: 5 consistency levels (strong to eventual), partition key selection critical for RU efficiency, ~1 RU per 1 KB point read, ~5 RU per 1 KB write.
Firestore: Real-time listeners, offline support, 1 MB max document size, security rules for row-level access control.
FerretDB: MongoDB wire protocol over PostgreSQL JSONB. Full PG ACID and tooling. Drop-in for simple MongoDB workloads without sharding.
Anti-Patterns
- Over-normalizing into many small collections — embed related data.
- Unbounded arrays — use bucket or outlier pattern.
- No schema validation — use JSON Schema (MongoDB) or explicit mappings (ES).
- Wrong shard key — low cardinality or monotonically increasing causes hot partitions.
- Over-denormalization — updates to duplicated data become error-prone.
- Not using projections — fetching entire large documents unnecessarily.
- Ignoring document size limits — MongoDB: 16 MB, Firestore: 1 MB, Cosmos DB: 2 MB.