SQL reference

Krishiv uses DataFusion for SQL parsing, expressions, planning, and local execution, then adds Engine-owned paths for placement, streaming, and lakehouse operations. The dialect is broad but pre-release; it is not described as universally compatible with another database.

Run the same query from each public API

cargo run -p krishiv -- sql \
  --query "SELECT 42 AS answer"

let result = session
    .sql("SELECT 42 AS answer")?
    .collect()?;
println!("{}", result.pretty()?);

session.sql("SELECT 42 AS answer").show()

The query is self-contained in all three examples. Names in data-backed queries resolve through the current session catalog, so register a table, file, or provider before referencing it.

Query surface

The checked-in SQL feature matrix covers these commonly used families:

• projection, filtering, sorting, limits, aliases, and DISTINCT;
• inner and outer joins, subqueries, and common table expressions;
• GROUP BY, aggregate functions, grouping sets, ROLLUP, and CUBE;
• ranking, offset, and value window functions;
• UNION, INTERSECT, and EXCEPT;
• table/view DDL and INSERT INTO;
• prepared statements through supported APIs and Flight SQL.

Use explain before relying on a complex query shape:

cargo run -p krishiv -- explain \
  --parquet orders=./orders.parquet \
  --query "SELECT customer_id, SUM(amount) FROM orders GROUP BY customer_id"

Engine extensions

Result and write semantics

• A successful batch result means all required stages completed.
• Plain files are not snapshot-consistent if they change during planning or execution.
• Client-side collect-and-write helpers are local conveniences, not atomic distributed writes.
• Distributed writes require a commit-capable sink.
• Streaming delivery is only as strong as the source, checkpoint storage, durability profile, and sink together.

SQL behavior follows the Engine compatibility policy and can change before the first stable release.