Database Deep Dives

The most significant new features in MongoDB 8.0 are: major query engine performance improvements (up to 36% faster average query performance, 56% faster on analytical workloads); Queryable Encryption reaching production maturity with support for range queries and prefix searches on encrypted fields; time series collection improvements including arbitrary field updates, faster bucket queries, and columnar compression; new aggregation operators ($percentile, $median, bitwise operators); and expanded Atlas Search capabilities including GA Vector Search and hybrid search.

Queryable Encryption allows MongoDB applications to encrypt sensitive document fields on the client side before storing them, while still supporting database queries on the encrypted data. The MongoDB client driver encrypts field values using cryptographic schemes designed to support specific query operations (equality, range, prefix) on ciphertexts without revealing the plaintext to the server. The database server processes queries on encrypted data and returns encrypted results, which the client decrypts. This means even a full database compromise exposes only useless ciphertext — the encryption keys never leave the application tier.

MongoDB reports average query performance improvements of approximately 36% compared to MongoDB 7.0, with analytical and read-heavy workloads seeing up to 56% improvement. Bulk write operations see up to 40% throughput improvement. Time series queries on bucketed data are up to 2× faster. These are benchmark averages — individual results vary significantly by workload type, query complexity, and data volume. Workloads that exercise the newly extended slot-based execution engine (SBE) coverage — particularly aggregation pipelines with $lookup and $group — see the largest improvements.

MongoDB 8.0 time series collections are suitable for workloads where time series data needs to coexist with other document data in the same platform, and where the query patterns are relatively standard (time range queries, aggregations by time window). For workloads with extreme time series throughput (millions of data points per second), very long retention (years at subsecond resolution), or requiring specialised time series analytics (complex windowed functions, anomaly detection), dedicated time series databases like InfluxDB, TimescaleDB (PostgreSQL extension), or QuestDB may be more appropriate. MongoDB 8.0 closes the gap considerably but purpose-built tools still outperform it at the extremes.

The slot-based execution engine (SBE) is MongoDB's high-performance query execution engine introduced progressively since MongoDB 5.1. Unlike the classic execution engine that evaluates documents one at a time through a tree of operators, SBE uses a vectorised execution model that processes batches of values through a pipeline of typed "slots" — similar to how columnar databases execute queries. This enables better CPU cache utilisation, fewer memory allocations, and significantly faster query execution. MongoDB 8.0 extends SBE coverage to more query patterns including complex aggregation stages that previously fell back to the classic engine.

For replica set upgrades: update your MongoDB drivers to versions compatible with MongoDB 8.0; upgrade secondaries one at a time (rolling upgrade) while keeping the replica set operational; use rs.stepDown() to step down the current primary; upgrade the former primary. For sharded clusters, upgrade mongos routers first, then config servers, then each shard (using the rolling upgrade process). MongoDB requires you to upgrade from 7.0 — you cannot skip major versions. After upgrading, run setFeatureCompatibilityVersion to unlock 8.0-specific features. Test in a staging environment that mirrors production before upgrading.

MongoDB Atlas Vector Search is a feature that allows storing and querying high-dimensional vector embeddings alongside regular document data in Atlas collections. Vector embeddings are numerical representations of unstructured data (text, images, audio) generated by machine learning models. Vector Search enables semantic similarity search — finding documents that are conceptually similar to a query, not just keyword matches. It is used for AI-powered search (semantic product search, document retrieval), recommendation engines, RAG (Retrieval-Augmented Generation) pipelines that ground LLM responses in a document database, and image/media similarity search. The hybrid search capability in MongoDB 8.0 combines vector similarity with traditional full-text scoring in a single query.

Choose MongoDB when your data model is naturally document-shaped (nested objects, arrays, variable-schema records), when you need horizontal sharding for write scalability beyond what a single PostgreSQL node can handle, when your team's productivity benefits from working with JSON-like documents without ORM translation, or when you need MongoDB-specific capabilities like Queryable Encryption, Atlas Search, or change streams for real-time data pipelines. Choose PostgreSQL when you have relational data with complex joins, need the strongest possible ACID guarantees at the transaction level, require SQL compatibility for business intelligence tools, or when your workload is heavily analytical — PostgreSQL's ecosystem (TimescaleDB, PostGIS, pg_vector) is richer for specialised analytical workloads.