Postgres is enough for 99% of projects

You probably don't need MongoDB, ClickHouse or DynamoDB. Postgres handles JSON, full-text search, time series and queues fine.

New projects often start with an architecture diagram full of databases: Mongo for documents, Elasticsearch for search, Redis for queues, a time-series store for metrics. Each one is another thing to back up, monitor, upgrade and keep consistent. For most projects, Postgres alone covers all of these well enough, and the extra systems add more operational weight than value.

Pick a specialized database when Postgres actually fails on your workload, not because a conference talk or Hacker News thread told you to.

What Postgres already covers

  • Documents. JSONB columns store and query JSON with proper indexes (GIN), so you get schemaless flexibility where you need it and relational structure everywhere else.
  • Full-text search. tsvector and tsquery with GIN indexes and language-aware stemming. Add pg_trgm for fuzzy matching and typo tolerance. That's enough for blogs, online stores and internal tools.
  • Time series. Native partitioning and BRIN indexes handle append-heavy data; the TimescaleDB extension adds compression and time-bucket functions on top.
  • Queues. FOR UPDATE SKIP LOCKED lets several workers pull jobs from a table without stepping on each other; LISTEN/NOTIFY wakes them up.
  • Scheduled jobs. The pg_cron extension runs SQL on a schedule inside the database.
  • Vectors and geo. pgvector for embeddings and similarity search, PostGIS for geospatial queries.
  • Many connections. PgBouncer pools a large number of client connections onto a small number of server connections.

A job queue in plain SQL:

UPDATE jobs
SET status = 'running', started_at = now()
WHERE id = (
  SELECT id FROM jobs
  WHERE status = 'queued'
  ORDER BY created_at
  FOR UPDATE SKIP LOCKED
  LIMIT 1
)
RETURNING id, payload;

Each worker runs this in a loop. A job is claimed by exactly one worker, and because the queue lives in the same database as your data, you can enqueue a job in the same transaction that creates the order.

Why one database is a feature

  • One backup and restore procedure, tested once, instead of several that must stay in sync.
  • Transactions across all your data. No "the order was saved but the search index wasn't updated" bugs.
  • Fewer moving parts to monitor, patch and pay for.
  • One skill set for the team, and one place to look when something is slow.

When Postgres really isn't enough

  • Analytical queries over very large event volumes, where a column store like ClickHouse is built for the job.
  • Search with heavy relevance tuning, faceting and multilingual analysis at scale, where Elasticsearch or OpenSearch earn their keep.
  • Very high-throughput caching or ephemeral key-value data, where Redis is simpler.
  • Multi-region writes with low latency everywhere, which needs a distributed database designed for it.

Before you get there, measure. Run EXPLAIN ANALYZE on the slow queries, add the missing indexes, set up connection pooling, partition big tables, add a read replica. Many "Postgres can't handle it" moments disappear at this step.

Common mistakes

  • Storing everything in JSONB and losing the benefits of a schema.
  • JSONB and full-text queries without indexes, followed by the conclusion that Postgres is slow.
  • No connection pooling in front of a serverless or many-worker app.
  • Ignoring autovacuum and table bloat on busy tables.
  • Backups that have never been restored as a test.

Checklist before adding a second database

  1. Is there a measured problem, with numbers from your own workload?
  2. Have indexes, query plans, pooling and partitioning been tried?
  3. Who will operate, back up and upgrade the new system?
  4. How will data stay consistent between the two?

If you can't answer all four, stay on Postgres a while longer. That's our default when we design a custom web application, and tuning an existing database is a regular part of web app support.

Wondering whether your project needs more than Postgres? Reach us on Telegram or MAX.

Web application development
Related service — let's discuss your project.

Got a project on this topic? Tell us the task — we'll come back with an estimate and a schedule.

Message on Telegram → Message on MAX