Modern DB choices listed
A recent thread summarized the modern database landscape startups tend to mix: Postgres for structured data, Redis for caching, DynamoDB for serverless patterns, Elasticsearch for search, and Neo4j for graph problems. The list frames database choice as pragmatic and use‑case driven rather than one‑size‑fits‑all. (x.com)
A database is where an app keeps its facts, and startups in 2026 increasingly split those facts across several systems instead of forcing one engine to do every job. PostgreSQL, Redis, Amazon DynamoDB, Elasticsearch, and Neo4j each target a different kind of work. (postgresql.org) (redis.io) (aws.amazon.com) (elastic.co) (neo4j.com) PostgreSQL is the default for rows and columns such as users, orders, invoices, and permissions. Its documentation centers transactions and concurrency control, which let multiple sessions update data while preserving integrity. (postgresql.org 1) (postgresql.org 2) PostgreSQL also stretches beyond classic tables. The current docs include built-in JSON and JSONB types for semi-structured data, plus full-text search for indexing documents and ranking matches. (postgresql.org 1) (postgresql.org 2) Redis keeps data in memory, which means on fast-access storage rather than disk-first storage, and teams use it when milliseconds matter. Redis documents caching patterns, data structures, publish-subscribe messaging, and optional persistence modes including snapshots and append-only logs. (redis.io 1) (redis.io 2) (redis.io 3) Amazon DynamoDB is Amazon Web Services’ serverless NoSQL database, built for key-value and document workloads that need automatic scaling. Amazon says DynamoDB delivers single-digit millisecond performance at any scale and offers on-demand, pay-per-request pricing for teams that do not want to manage capacity. (docs.aws.amazon.com) (aws.amazon.com) Elasticsearch is built for search, which means turning words into indexes so an app can find relevant results quickly across large text collections. Elastic says documents become searchable in near real time, typically within 1 second, and the engine is tuned for search relevance and analytics at production scale. (elastic.co) (elastic.co) Neo4j handles graph data, where the main question is how things connect rather than which row they sit in. Neo4j’s model uses nodes, relationships, and paths, and its Cypher query language is designed to match patterns such as “customer bought product” or “person knows person.” (neo4j.com) (neo4j.com) That mix reflects how modern software is built. A startup might store billing records in PostgreSQL, session state in Redis, event-heavy traffic in DynamoDB, product catalog search in Elasticsearch, and fraud or recommendation links in Neo4j. (postgresql.org) (redis.io) (docs.aws.amazon.com) (elastic.co) (neo4j.com) The tradeoff is operational complexity. Every extra database adds a new query model, backup plan, pricing model, and failure mode, which is why many teams still start with PostgreSQL and add other systems only when a specific bottleneck appears. (postgresql.org) (aws.amazon.com) (redis.io) So the current database playbook is less about picking a winner than assigning roles. The common pattern is one reliable system of record, then specialized engines for cache, search, scale, or relationship-heavy queries. (postgresql.org) (redis.io) (aws.amazon.com) (elastic.co) (neo4j.com)
