CrateDB

Use it when

• Sensor and industrial data arrives in mixed shapes: relational rows, nested JSON, time-series, geospatial, vectors, and full text behind one SQL surface.
• PostgreSQL wire compatibility matters; standard drivers and JDBC connect without an adapter.
• High-cardinality IoT data must stay queryable in near real time, and everything being indexed by default is worth its write cost.
• The alternative is a document store beside a time-series database beside a search index.

Think twice when

• Raw scan throughput on flat tables is the benchmark; ClickHouse and Doris are built for exactly that.
• You rely on transactions or strict consistency; CrateDB is eventually consistent and has no transactions.
• Pre-aggregation is central to your serving model; there are no materialized views, so aggregates run at query time.

How it runs

A shared-nothing cluster where every node ingests, stores, coordinates, and answers queries; tables shard with replicas across nodes, and scaling means adding nodes. Storage is Lucene-based with columnar and row structures per field. CrateDB Cloud is the managed option.