datus-ai/datus-db-adapters/db-maxcompute-sql

MaxCompute SQL

Generate MaxCompute SQL from the configured datasource context and metadata. MaxCompute is also known as ODPS; its PyODPS SDK and Hive parser fallback do not make it a general Hive or Spark SQL engine.

Namespaces and identifiers

• MaxCompute has no catalog level. A Datus datasource is bound to one project, so do not generate cross-project access.
• In a two-level project, qualify objects as project.table; there is no schema namespace.
• In a schema-enabled project, qualify objects as project.schema.table. Use the configured or metadata-provided schema, which defaults to default when none is supplied.
• Under the Datus adapter contract, a two-part identifier always means project.table, never schema.table. Use all three parts when a schema-qualified identifier is needed.
• Do not emit USE, USE SCHEMA, or SET odps.namespace.schema to switch context. The connector supplies the project, schema, and namespace hint for each job.
• Preserve metadata-provided names and quote reserved words or special identifiers with backticks.

Queries and scan cost

• Add predicates on relevant partition columns when querying partitioned tables. Do not silently enable odps.sql.allow.fullscan.
• LIMIT restricts returned rows only after the distributed scan; it does not reduce bytes scanned or computing cost.
• By default, a global ORDER BY requires a LIMIT. Include a business-appropriate limit rather than inventing an unbounded sort override.
• Use MaxCompute functions and syntax. Do not borrow a MySQL, PostgreSQL, Spark, or Hive construct unless MaxCompute supports it.

Types and expressions

• Projects can use MaxCompute V1, V2, or Hive-compatible data type editions. Follow metadata and explicit project context instead of assuming an edition.
• Prefer explicit CAST when type coercion is material to correctness; implicit conversion rules differ between editions.
• The adapter's hive parser dialect is a parsing fallback, not evidence that odps.sql.hive.compatible is enabled.

Avoid dialect leaks

Before returning SQL, reject invented catalog, warehouse, cluster, distribution, storage-engine, or transaction syntax. Do not assume primary keys, row-level mutation, or Hive compatibility when the table metadata and project settings do not establish them.

Interpret existing ETL and views

• INSERT INTO appends; INSERT OVERWRITE TABLE replaces the table or static partition. Ordinary views are logical definitions; materialized views store data.
• INSERT mappings are positional. Static partition columns are omitted from SELECT; dynamic partition expressions are last and follow the declaration order. Preserve this mapping when following a stored result's lineage.

Task references

• MaxCompute writes and materialized views: INSERT mappings, partition writes, row mutation and materialized-view definitions.

Read these for the requested operation, including SQL-only authoring. Supplied ETL is evidence; follow the user's task.