Five years ago you'd use Postgres + MongoDB + Elasticsearch + Redis + Pinecone. In 2024, Postgres alone often covers all of it.
What Postgres now does well
- JSON / JSONB - schema-flexible documents with indexing
- Full-text search - built-in, good enough for most apps without Elastic
- Vectors - pgvector + HNSW indexes are production-ready
- Queues - SKIP LOCKED makes Postgres a perfectly fine job queue
- Time-series - TimescaleDB extension competes with InfluxDB
- Partitioning - declarative table partitioning for billion-row tables
Hosted choices in 2024
- Supabase - Postgres + auth + storage + realtime
- Neon - serverless Postgres with branching
- RDS / Aurora - AWS workhorses, mature
- Cloud SQL - GCP equivalent
- Crunchy Bridge - managed Postgres without the lock-in
Bottom line
Default to Postgres. Add a second data store only when Postgres genuinely can't do the job - which is rarer than you'd think.