Summary

Add parser and post-processing support for SQL ORDER BY and LIMIT on supported grouped aggregate queries.

Queries such as:

SELECT g1, g2, agg(v) AS p99
FROM t
GROUP BY g1, g2
ORDER BY p99 DESC
LIMIT 10

now parse and execute successfully where they previously fell through to:

destination=none_unsupported

Changes

• Capture:

  • ORDER BY items
  • LIMIT N
  • aggregate aliases

during SQL parsing and store them on SQLQueryData.

• Apply sorting and truncation after execute_query_pipeline() returns.

• Add a new SqlPostProcessing side-channel returned from build_sql_query_pieces() alongside QueryExecutionContext:

  • keeps the shared execution context unchanged
  • avoids impact to PromQL and Elastic paths

• matches_sql_pattern() now ignores:

  • ORDER BY
  • LIMIT
  • aggregate aliases

so existing registered templates continue matching queries that add presentation-only clauses.

• SqlPostProcessing::apply() short-circuits when no ordering or limit is present, making the fast path effectively zero-cost.

Validation Rules

ORDER BY items must reference either:

• the aggregate alias
• a GROUP BY column

Unknown identifiers are rejected at parse time to avoid silently returning unsorted results for typo'd columns.

Supported

• single or multi-column ORDER BY
• mixed ASC / DESC
• ordering by aggregate alias
• ordering by GROUP BY columns
• LIMIT N

Still Unsupported

The following remain rejected:

• positional ORDER BY
• expressions in ORDER BY
• OFFSET
• LIMIT BY
• MySQL LIMIT a, b
• WITH FILL
• NULLS FIRST/LAST
• ClickHouse ORDER BY ALL

Tests Added

Added 16 tests covering:

• aggregate alias ordering with LIMIT

• default ascending order

• multi-column mixed directions

• ordering by group-by keys

• limit-only queries

• tie-breaking behavior

• NaN handling

• matches_sql_pattern() compatibility

• no-op fast path

• end-to-end SqlPostProcessing::apply

• rejection of:

  • unknown columns
  • expressions
  • OFFSET