SQL and execution

Queries and projection

Projection is strict: unknown or duplicated columns are an error rather than a silent pass-through. Tables resolve as catalog.schema.table, or unqualified against the session default.

SELECT region, plan, total
FROM   warehouse.default.orders
WHERE  total > 500 AND region <> 'Asia'
ORDER  BY total DESC NULLS LAST
LIMIT  20;

ORDER BY supports multiple keys, per-key direction, and explicit null placement. A LIMIT directly above a sort is fused into a TopN operator rather than sorting the whole input.

Aggregation

SELECT   region,
         count(*)                AS orders,
         sum(total)              AS revenue,
         avg(total)              AS avg_order,
         count(DISTINCT plan)    AS plans,
         sum(DISTINCT total)     AS distinct_total
FROM     orders
GROUP    BY region
HAVING   sum(total) > 1000
ORDER    BY revenue DESC;

SUM, COUNT, AVG, MIN, MAX and COUNT(DISTINCT …) all execute, locally and distributed. Distributed aggregation is partial/final: workers emit partial states that the coordinator merges, so AVG is carried as a weighted state and DISTINCT as an exact mergeable state rather than being recomputed.

Joins

SELECT c.region, count(*) AS orders
FROM   orders o
JOIN   customers c ON o.customer_id = c.id
GROUP  BY c.region;

INNER, LEFT, RIGHT, FULL and CROSS joins execute locally with qualified relation aliases. Distributed INNER equi-joins repartition both sides by hash; a broadcast build side is used when the planner selects it. Broader representative distributed evidence for outer and cross joins now passes; broader skew and failure qualification remains an alpha gate. Join conditions must be equalities — general predicates are not yet supported as join conditions, though they work in WHERE.

Window functions

SELECT region,
       plan,
       total,
       row_number() OVER (PARTITION BY region ORDER BY total DESC) AS rank_in_region,
       lag(total)   OVER (PARTITION BY region ORDER BY ordered)    AS prev_total,
       sum(total)   OVER (PARTITION BY region)                     AS region_total
FROM   orders;

ROW_NUMBER, RANK, DENSE_RANK, LAG and LEAD are available, as is any supported aggregate used with OVER, with PARTITION BY and ORDER BY.

Set operations

SELECT region FROM orders_2025
UNION ALL
SELECT region FROM orders_2026;

SELECT region FROM orders INTERSECT SELECT region FROM targets;
SELECT region FROM orders EXCEPT    SELECT region FROM excluded;

Expressions and functions

SELECT date_trunc('month', ordered) AS month,
       extract(year FROM ordered)   AS yr,
       upper(region)                AS region,
       CASE WHEN total > 500 THEN 'large' ELSE 'small' END AS bucket
FROM   orders
WHERE  ordered >= current_date - 90;

Not generally executable yet

• Arbitrary scalar and correlated subqueries; supported decorrelated forms are limited to the tested binder paths.
• Non-equality join conditions.
• Data-changing DML such as INSERT, UPDATE, DELETE, and MERGE. Catalog DDL is supported through the Engine catalog surface.
• Comprehensive decimal and date/time edge-case behavior.

How capability is claimed

Kaveon does not publish an ANSI SQL conformance percentage, because it has no conformance corpus to back one. A feature is listed here only once its parser, operator, and distributed-fragment path pass together on dev. Work in progress in a branch is not a capability claim.

To check what your build actually does rather than trusting this page, run the statement — a planning failure is explicit:

kaveon --local --data-dir /data/warehouse -e "SELECT row_number() OVER (ORDER BY total) FROM orders LIMIT 1"