Provides MaxCompute SQL engine syntax guidance for text2sql scenarios. MaxCompute is based on Hive SQL extensions and has significant differences from ANSI standard SQL.
Load corresponding reference files based on question type (can be combined). The load column indicates relative loading weight (heavy ≈ large, medium ≈ moderate), for budget estimation, not exact token counts. Each file's opening paragraph describes its own scope.
| Trigger condition | File | load |
|---|---|---|
| --- | --- | --- |
| NL→SELECT generation / text2sql (determine intent, granularity, table-column mapping, output format first) | references/text2sql_principles.md | light |
| Generate MaxCompute SELECT queries (dialect rules, DQL default must-read) | references/maxcompute_select_guide.md | heavy |
| Match query pattern keywords: Top N / top-N per group / year-over-year / month-over-month / consecutive N days / retention / row-to-column / column-to-row / PIVOT / UNPIVOT / array expansion / LATERAL VIEW / EXPLODE / JSON extraction / GET_JSON_OBJECT / cumulative / running total / Range Join / GROUPING SETS / CUBE / ROLLUP / pagination / paging | references/sql_query_patterns.md | medium |
SQL execution failure requiring diagnosis and recovery (including ODPS-0xxx error codes) | references/sql_common_errors.md | medium |
> Relationship: text2sql_principles.md provides engine-independent NL→SELECT generation principles; maxcompute_select_guide.md is the single authoritative source for MaxCompute DQL dialect rules (unsupported syntax/functions/partitioning/types/SET parameters). sql_query_patterns.md provides query template snippets only, without duplicating rules.
The following scenarios exceed this skill's scope:
共 1 个版本