Tracking: SQL query parsing and matching support #822

Tracks SQL query shapes we want ASAPQuery to support, using the PromQL compliance suites as the reference set. Coverage is checked by the SQL query support runner (#823) (asap-query-engine/tests/sql_query_support.rs), which plans each query and builds the engine execution context without executing it. Its suites mirror promql-compliance/suites case-for-case.

Supported today

• Single-layer SUM / COUNT / MIN / MAX / QUANTILE / COUNT(DISTINCT col) with GROUP BY, e.g. sum by (job) (sum_over_time(data[5m]))
• Time range as BETWEEN or half-open >= ... <, relative (NOW()) or absolute
• Global top-k by SUM or COUNT: ORDER BY <alias> DESC LIMIT k

Unsupported shapes

• sql: aggregation without GROUP BY is rejected #817 Aggregation without GROUP BY, e.g. sum(data), quantile(0.5, m), histogram_quantile(...)
• sql: per-group top-k (LIMIT k BY cols) is rejected #818 Per-group top-k (LIMIT k BY cols), e.g. topk by (job) (3, ...)
• sql: two-layer nested aggregation (subquery in FROM) is rejected #819 Two-layer nested aggregation, e.g. sum by (job) (sum by (job, instance) (data))
• sql: outer SELECT without an aggregate over a subquery is rejected #820 Outer SELECT without an aggregate over a subquery, e.g. topk by (job) (3, sum by (job, instance) (data))
• sql: ORDER BY / LIMIT inside a FROM subquery is silently dropped #821 ORDER BY / LIMIT inside a subquery is silently dropped
• sql: ORDER BY / LIMIT on the outer layer of a two-layer query is rejected #891 ORDER BY / LIMIT on the outer layer of a two-layer query, e.g. topk(5, sum by (job) (sum by (job, instance) (data)))
• Three or more nested layers, e.g. sum by (job) (avg by (job) (sum by (job, instance) (data))). Only one subquery level is parsed.
• AVG: planned as Sum + Count, but the engine only resolves single-statistic SQL queries
• Arithmetic in the projection, e.g. SUM(value) / COUNT(value), SUM(value) + 1, QUANTILE(0.9, value) / QUANTILE(0.5, value)
• More than one aggregate in a SELECT, e.g. SUM(value), COUNT(value)
• COUNT(*)

No SQL equivalent yet (not in the SQL suites)

• Counter rate() / increase()
• PromQL subqueries over time, e.g. max_over_time((sum(m))[10m:])
• Bare selector (raw samples), e.g. user_service_cache_refresh_lag_seconds

Intentionally rejected

• Closed <= upper time bound: only half-open [start, end) ranges match ClickHouse's scan
• No time filter