MongoDB Database Architect

Purpose & When-To-Use

Invoke this skill when designing, reviewing, or optimizing MongoDB database architectures for applications requiring flexible schema, document-based data models, horizontal scalability, or high availability.

Trigger Conditions:

• "Design a MongoDB architecture for [use case]"
• "How should I model [entity relationships] in MongoDB?"
• "My MongoDB queries are slow, need indexing recommendations"
• "Plan a sharding strategy for [data volume] with [growth rate]"
• "Optimize MongoDB aggregation pipeline for [query pattern]"
• "Configure replica set for [availability SLA]"
• "Migrate from MongoDB [old version] to 8.0"

Out of Scope:

• SQL database design (use database-postgres-architect)
• Redis caching (use database-redis-architect when available)
• General database migration (use database-migration-generator)

Pre-Checks

1. Time Normalization: Compute NOW_ET using NIST/time.gov semantics (America/New_York, ISO-8601).
2. Input Validation:
   • Workload type specified (OLTP, OLAP, time-series, content, analytics)
   • Data volume estimates available (documents, collections, size)
   • Access patterns described (query types, read/write ratio)
3. Version Check: MongoDB version specified (default to 8.0 if not provided).
4. Deployment Context: Cloud provider or self-hosted, resource constraints (RAM, CPU, storage).
5. Existing Schema: If optimizing existing database, request sample documents and query patterns.

Abort Conditions:

• No workload type or access patterns provided → emit TODO list with required inputs.
• Data volume completely unknown → warn that sizing recommendations will be generic.

Procedure

T1: Quick Schema Review & Index Recommendations (≤2k tokens)

Use Case: Fast path for common scenarios (80% of requests).

Steps:

1. Analyze Access Patterns: Identify top 3-5 most frequent queries.
2. Document Modeling Decision:
   • Embed if 1-to-1 or 1-to-many relationships, low cardinality, read-heavy, atomic updates needed.
   • Reference if many-to-many, frequently changing data, high cardinality, document size >16 MB risk.
3. Index Recommendations (ESR Rule):
   • Equality fields first (exact match filters: {status: "active"})
   • Sort fields next (sort order: .sort({created_at: -1}))
   • Range fields last (range queries: {age: {$gt: 18}})
   • Example: db.users.createIndex({status: 1, created_at: -1, age: 1})
4. Top 3 Bottlenecks: Identify missing indexes, sequential scans, document size issues.
5. Quick Wins: Provide 1-3 immediate optimizations with estimated speedup (e.g., "Add index on email → 100x faster login").

Output: Schema design decision, 3-5 index recommendations with ESR justification, top 3 bottlenecks.

T2: Complete Architecture Design (≤6k tokens)

Use Case: Comprehensive architecture for production deployments.

Steps:

1. Document Schema Design

Embedding vs Referencing Table:

Schema Validation (MongoDB 8.0):

db.createCollection("users", {
  validator: {
    $jsonSchema: {
      bsonType: "object",
      required: ["email", "created_at"],
      properties: {
        email: {bsonType: "string", pattern: "^.+@.+$"},
        age: {bsonType: "int", minimum: 0, maximum: 150},
        created_at: {bsonType: "date"}
      }
    }
  }
})

2. Advanced Indexing Strategy

Index Types and Use Cases:

Covered Query Optimization:

// Covered query: all fields in index, no document access
db.orders.createIndex({user_id: 1, status: 1, total: 1})
db.orders.find({user_id: 12345, status: "shipped"}, {_id: 0, user_id: 1, status: 1, total: 1})
// explain() shows: totalDocsExamined: 0 (covered by index)

ESR Rule Application:

// Query: Find active users, sort by created_at descending, filter age > 18
db.users.find({status: "active", age: {$gt: 18}}).sort({created_at: -1})

// Correct index (ESR):
// Equality: status (exact match)
// Sort: created_at (sort field)
// Range: age (range filter)
db.users.createIndex({status: 1, created_at: -1, age: 1})

3. Sharding Strategy

Shard Key Selection (MongoDB 8.0):

Hashed Sharding Example:

// Enable sharding on database
sh.enableSharding("myapp")

// Shard collection with hashed shard key
sh.shardCollection("myapp.users", {user_id: "hashed"})

// MongoDB 8.0: Move unsharded collection to specific shard
db.adminCommand({moveCollection: "myapp.analytics", toShard: "shard02"})

Ranged Sharding Example (Time-Series):

// Shard by timestamp for time-series data
sh.shardCollection("myapp.events", {timestamp: 1})

// Zone sharding (MongoDB 8.0): Route data by date ranges to specific shards
sh.addShardToZone("shard01", "recent")
sh.updateZoneKeyRange("myapp.events", {timestamp: ISODate("2025-01-01")}, {timestamp: MaxKey}, "recent")

Avoid Scatter-Gather Queries:

• Include shard key in queries: db.users.find({user_id: 12345}) → targets single shard
• Without shard key: db.users.find({email: "test@example.com"}) → scatter-gather across all shards (slow)
• Exception: Large aggregations benefit from parallelism across shards

4. Aggregation Pipeline Optimization

Pipeline Stages (Execution Order Matters):

// Optimized aggregation: $match early, $project late, use indexes
db.orders.aggregate([
  // Stage 1: $match FIRST (uses index, reduces documents)
  {$match: {status: "shipped", created_at: {$gte: ISODate("2025-01-01")}}},

  // Stage 2: $sort (uses index if compound index exists)
  {$sort: {created_at: -1}},

  // Stage 3: $lookup (join with users collection)
  {$lookup: {
    from: "users",
    localField: "user_id",
    foreignField: "_id",
    as: "user_details"
  }},

  // Stage 4: $group (aggregation after filtering)
  {$group: {
    _id: "$user_id",
    total_spent: {$sum: "$total"},
    order_count: {$sum: 1}
  }},

  // Stage 5: $project LAST (reduce network transfer)
  {$project: {_id: 1, total_spent: 1, order_count: 1}}
])

Index Sort Optimization:

• If {status: 1, created_at: -1} index exists, $match + $sort uses index (no in-memory sort).
• Without index, MongoDB sorts in memory (limited by 100 MB unless allowDiskUse: true).

Sharded Aggregation (MongoDB 8.0):

• Aggregations run in parallel on each shard, then merge results.
• Use $match early to enable shard targeting.
• MongoDB 8.0: 32% overall performance improvement in aggregations.

5. Replica Set Configuration

Standard 3-Member Replica Set:

rs.initiate({
  _id: "myReplicaSet",
  members: [
    {_id: 0, host: "mongo1.example.com:27017", priority: 2},  // Primary (high priority)
    {_id: 1, host: "mongo2.example.com:27017", priority: 1},  // Secondary
    {_id: 2, host: "mongo3.example.com:27017", arbiterOnly: true}  // Arbiter (no data)
  ]
})

Read Preferences:

• primary (default): All reads from primary (strong consistency).
• primaryPreferred: Read from primary, fallback to secondary if unavailable.
• secondary: Read from secondary (may read stale data).
• secondaryPreferred: Read from secondary, fallback to primary.
• nearest: Read from lowest-latency member.

Write Concerns:

• w: 1 (default): Acknowledge after primary write (fast, risk of data loss on primary failure).
• w: "majority": Acknowledge after majority of replica set members (slower, durable).
• w: 3: Acknowledge after 3 members (explicit count).
• j: true: Wait for write to journal (disk) before acknowledging.

MongoDB 8.0 Replica Set Enhancements:

• Faster concurrent writes during replication.
• Disable "majority" read concern for PSA (Primary-Secondary-Arbiter) to avoid cache pressure.

6. WiredTiger Cache & Connection Pool Tuning

WiredTiger Cache Sizing (MongoDB 8.0):

# Default: 50% of RAM - 1 GB
storage:
  wiredTiger:
    engineConfig:
      cacheSizeGB: 32  # For 64 GB RAM server (50%)

Guidelines:

• Production: 50-62.5% of available RAM (balance with filesystem cache).
• Cache should hold working set (frequently accessed data).
• Monitor: db.serverStatus().wiredTiger.cache (bytes in cache, eviction activity).
• Too large: Starves OS filesystem cache, degrades performance.
• Too small: High eviction rate, poor query performance.

Connection Pool Configuration:

// Application connection string
mongodb://mongo1.example.com:27017,mongo2.example.com:27017,mongo3.example.com:27017/?replicaSet=myReplicaSet&maxPoolSize=50&minPoolSize=10&maxIdleTimeMS=60000

Settings:

• maxPoolSize: Maximum connections (default 100). Each connection ~1 MB RAM.
• minPoolSize: Minimum connections (default 0). Pre-warm pool for faster queries.
• maxIdleTimeMS: Close idle connections after timeout (default: no timeout).
• Rule of Thumb: maxPoolSize ≈ (expected concurrent operations) + 10-20% buffer.

7. Performance Tuning (MongoDB 8.0)

Configuration Parameters:

# For 64 GB RAM server (OLTP workload)
storage:
  wiredTiger:
    engineConfig:
      cacheSizeGB: 32                    # 50% of RAM
    collectionConfig:
      blockCompressor: snappy             # Default compression
    indexConfig:
      prefixCompression: true             # Index prefix compression

net:
  maxIncomingConnections: 65536           # Max client connections
  compression:
    compressors: snappy                   # Network compression

operationProfiling:
  mode: slowOp                            # Profile slow queries
  slowOpThresholdMs: 100                  # Log queries >100ms

replication:
  replSetName: myReplicaSet
  enableMajorityReadConcern: true         # Disable for PSA if cache pressure

MongoDB 8.0 Performance Improvements:

• 36% faster reads (vectored I/O, reduced memory usage).
• 56% faster bulk inserts (batch processing optimizations).
• 75% query latency reduction (internal benchmarks).
• Embedded sharding config servers (no separate config server replica set).
• Move collections across shards without shard key (MongoDB 8.0 feature).

Output: Complete architecture document with schema design, index definitions, sharding strategy, aggregation examples, replica set config, tuning parameters.

T3: Enterprise Features & Migration Planning (≤12k tokens)

Use Case: Multi-region deployments, queryable encryption, sharding at scale, version migrations.

Steps:

1. Multi-Region Replica Set (Global Deployment)

rs.initiate({
  _id: "globalReplicaSet",
  members: [
    {_id: 0, host: "us-east-1.example.com:27017", priority: 2, tags: {region: "us-east"}},
    {_id: 1, host: "us-east-2.example.com:27017", priority: 1, tags: {region: "us-east"}},
    {_id: 2, host: "eu-west-1.example.com:27017", priority: 1, tags: {region: "eu-west"}},
    {_id: 3, host: "ap-southeast-1.example.com:27017", priority: 1, tags: {region: "ap-southeast"}},
    {_id: 4, host: "arbiter.example.com:27017", arbiterOnly: true}
  ],
  settings: {
    // Read from nearest region
    getLastErrorDefaults: {w: "majority", wtimeout: 5000}
  }
})

// Region-specific read preferences
db.users.find({region: "us-east"}).readPref("nearest", [{region: "us-east"}])

2. Queryable Encryption (MongoDB 8.0)

Use Case: Encrypt sensitive fields (PII, PHI) while allowing queries.

// Create encrypted collection (MongoDB 8.0 expanded support)
db.createCollection("patients", {
  encryptedFields: {
    fields: [
      {
        path: "ssn",
        bsonType: "string",
        queries: {queryType: "equality"}  // Allow equality queries on encrypted field
      },
      {
        path: "medical_record",
        bsonType: "string"
        // No queries specification = cannot query, only store/retrieve
      }
    ]
  }
})

// Query encrypted field
db.patients.find({ssn: "123-45-6789"})  // Allowed (queryable encryption)

3. Cross-Shard Aggregation Optimization

Parallel Execution on Sharded Cluster:

// Aggregation runs in parallel on each shard, then merges
db.orders.aggregate([
  {$match: {created_at: {$gte: ISODate("2025-01-01")}}},  // Shard targeting if sharded by created_at
  {$group: {_id: "$product_id", total_sales: {$sum: "$total"}}},
  {$sort: {total_sales: -1}},
  {$limit: 10}
], {allowDiskUse: true})  // Allow >100 MB sort for large datasets

Optimization:

• Shard by time-based field → $match with date range targets recent shards only.
• MongoDB 8.0: Improved parallelism for FULL OUTER JOIN and aggregations.
• Monitor with db.currentOp() to see query distribution across shards.

4. Migration from MongoDB 6.x/7.x to 8.0

Benefits of MongoDB 8.0:

• 36% faster reads, 56% faster bulk inserts (accessed 2025-10-26T18:17:22-0400, InfoQ MongoDB 8.0).
• Embedded sharding config servers (reduce infrastructure).
• Queryable encryption enhancements.
• Move collections across shards without shard key.

Migration Strategy (Zero-Downtime):

1. Set up MongoDB 8.0 replica set members (add to existing replica set as secondaries).
2. Replicate data (wait for secondaries to sync).
3. Test queries on MongoDB 8.0 secondaries (validate compatibility, performance).
4. Stepdown primary (rs.stepDown()) → elect MongoDB 8.0 member as new primary.
5. Upgrade remaining members (rolling upgrade, one at a time).
6. Set feature compatibility version: db.adminCommand({setFeatureCompatibilityVersion: "8.0"}).
7. Monitor for 24h (rollback if issues detected).

Risks:

• Incompatible drivers (ensure client drivers support MongoDB 8.0).
• Deprecated features removed (check release notes).
• Configuration parameter changes.

5. Monitoring & Observability

Key Metrics:

// Server status
db.serverStatus()

// WiredTiger cache stats
db.serverStatus().wiredTiger.cache
// Monitor: bytes currently in cache, bytes read into cache, eviction activity

// Connection stats
db.serverStatus().connections
// Monitor: current, available, totalCreated

// Operation counters
db.serverStatus().opcounters
// Monitor: insert, query, update, delete, getmore, command

// Slow query profiling
db.system.profile.find({millis: {$gt: 100}}).sort({ts: -1}).limit(10)

// Index usage stats
db.collection.aggregate([{$indexStats: {}}])

Tools:

• MongoDB Atlas: Built-in monitoring, Performance Advisor (index recommendations).
• Self-Hosted: Percona Monitoring and Management (PMM), MongoDB Ops Manager.
• Application Performance Monitoring (APM): Datadog, New Relic, Dynatrace.

Output: Multi-region architecture, queryable encryption setup, cross-shard aggregation strategy, migration plan with risks, monitoring dashboards.

Decision Rules

1. Embedding vs Referencing:

   • If relationship is 1-to-many with <100 related documents → Embed.
   • If related data changes frequently or queried independently → Reference.
   • If document size risk >16 MB or unbounded growth → Reference.

2. Index Creation:

   • Add index if query scans >1000 documents without index.
   • Use compound index (ESR rule) for multi-field queries.
   • Avoid indexes on low-cardinality fields (<10 distinct values).
   • Use partial indexes for large collections with filtered queries.

3. Sharding Trigger:

   • Enable sharding if data size >200 GB or growth rate >50 GB/month.
   • Use hashed shard key for monotonic IDs (avoid hotspots).
   • Use ranged shard key for time-series data with zone sharding.

4. Replica Set Configuration:

   • Use 3-member replica set minimum (1 primary + 2 secondaries or 1 secondary + 1 arbiter).
   • Use 5-member replica set for high availability (1 primary + 4 secondaries).
   • Set w: "majority" for critical writes (durability over speed).
   • Use readPreference: "secondary" for analytics queries (offload primary).

5. WiredTiger Cache Sizing:

   • Allocate 50% of RAM for WiredTiger cache (default).
   • Increase to 62.5% if working set >50% RAM and low filesystem cache usage.
   • Decrease if high OS memory pressure or filesystem cache thrashing.

6. Aggregation Optimization:

   • Place $match as early as possible (reduce documents).
   • Use indexes for $match and $sort stages.
   • Use $project last to reduce network transfer.
   • Enable allowDiskUse: true for >100 MB sorts/groups.

Uncertainty Thresholds:

• If access patterns unclear → request sample queries and usage statistics.
• If data volume highly uncertain → provide scalable architecture with sharding plan.
• If existing schema has >10 collections → focus on top 3 most-queried collections first.

Output Contract

Required Fields:

document_schema:
  - collection_name: string
    embedding_decision: "embed" | "reference"
    justification: string (why embed or reference)
    schema_validation: object (JSON schema)
    sample_document: object

indexes:
  - collection_name: string
    index_name: string
    index_definition: object ({field: 1|-1})
    index_type: "single" | "compound" | "multikey" | "text" | "geospatial" | "hashed" | "wildcard" | "partial"
    esr_justification: string (if compound index)
    estimated_speedup: string (e.g., "50x faster")

sharding_strategy:
  - enabled: boolean
    shard_key: object ({field: "hashed" | 1})
    shard_key_type: "hashed" | "ranged" | "compound"
    justification: string (why this shard key)
    target_chunk_size: string (default: "64 MB")

aggregation_examples:
  - use_case: string
    pipeline: array (aggregation stages)
    optimization_notes: string

replica_set:
  - members: integer (3, 5, etc.)
    configuration: object (rs.initiate() config)
    read_preference: "primary" | "primaryPreferred" | "secondary" | "secondaryPreferred" | "nearest"
    write_concern: object ({w: "majority", j: true})

performance_tuning:
  - wiredtiger_cache_gb: number
    max_connections: integer
    connection_pool_size: integer
    profiling_threshold_ms: integer
    estimated_improvement: string (e.g., "36% faster reads")

migration_plan:  # If upgrading versions
  - current_version: string
    target_version: string
    strategy: "rolling upgrade" | "blue-green" | "snapshot restore"
    steps: array (migration steps)
    risks: array (potential issues)
    downtime_estimate: string

Token Tier Minimums:

• T1: document_schema (embed/reference decision), indexes (top 3-5), bottlenecks (top 3).
• T2: All of T1 + sharding_strategy, aggregation_examples, replica_set, performance_tuning.
• T3: All of T2 + multi-region, queryable_encryption, migration_plan, monitoring.

Examples

ESR Rule for Compound Index:

// Query: Find active users, sort by created_at descending, filter age > 18
db.users.find({status: "active", age: {$gt: 18}}).sort({created_at: -1})

// Index following ESR rule:
// E (Equality): status
// S (Sort): created_at
// R (Range): age
db.users.createIndex({status: 1, created_at: -1, age: 1})

See examples/content-management-mongodb-architecture.txt for a complete architecture example.

Quality Gates

1. Token Budgets:

   • T1 response ≤2k tokens (fast path, common scenarios).
   • T2 response ≤6k tokens (complete architecture).
   • T3 response ≤12k tokens (enterprise features, migrations).

2. Safety Checks:

   • No credentials or connection strings with passwords in output.
   • Schema validation rules enforce data quality (no malformed documents).
   • Audit logging enabled for sensitive collections (PII, PHI).

3. Auditability:

   • All index recommendations include ESR justification.
   • All sharding strategies include shard key selection reasoning.
   • All performance claims cite MongoDB 8.0 benchmarks with access dates.

4. Determinism:

   • Same input (workload, data volume, access patterns) → same architecture recommendations.
   • Index order deterministic (ESR rule applied consistently).

5. Citations:

   • MongoDB 8.0 performance improvements: 36% faster reads, 56% faster bulk inserts (accessed 2025-10-26T18:17:22-0400, InfoQ).
   • ESR Rule (Equality, Sort, Range) for compound indexes (accessed 2025-10-26T18:17:22-0400, MongoDB Blog).
   • WiredTiger cache default: 50% RAM - 1 GB (accessed 2025-10-26T18:17:22-0400, MongoDB Docs).

Resources

Official MongoDB Documentation:

• MongoDB 8.0 Release Notes
• Data Modeling Introduction
• Indexing Strategies
• Sharding Reference
• Aggregation Pipeline
• Replica Set Configuration

Performance & Best Practices:

• Performance Best Practices: Indexing (accessed 2025-10-26T18:17:22-0400)
• Performance Best Practices: Sharding (accessed 2025-10-26T18:17:22-0400)
• MongoDB 8.0 Performance Improvements (accessed 2025-10-26T18:17:22-0400)
• WiredTiger Cache Tuning (accessed 2025-10-26T18:17:22-0400)

Tools:

• MongoDB Compass (GUI for schema exploration and index management)
• MongoDB Atlas (managed cloud database with Performance Advisor)
• Percona Monitoring and Management (PMM) (self-hosted monitoring)