AIP-99: Add LLMSchemaCompareOperator

[gopidesupavan]

closes: #62734

Add a new operator for cross-system schema drift detection powered by LLM reasoning to the common.ai provider.

LLMSchemaCompareOperator introspects schemas from multiple data sources (databases via DbApiHook, object storage via DataFusionEngine) and uses an LLM to identify mismatches that would break data loading. The LLM handles complex cross-system type mapping that simple equality checks miss (e.g., varchar(255) vs string, timestamp vs timestamptz).

from airflow.sdk import dag
import datetime

@dag
def example_llm_schema_compare():

    @task.llm_schema_compare(
        db_conn_ids=["postgres_default"],
        table_names=["customers", "secondory_customers"],
        llm_conn_id="llm_conn_id",
        model_id="google-gla:gemini-2.5-pro",
    )
    def compare_postgres_customer_tables(ds=None):
        return "Identify schema mismatches that would break data loading between systems"

    compare_postgres_customer_tables()

example_llm_schema_compare()

An example output view:

{
  "summary": "The schemas are not compatible. The target table 'secondory_customers' is missing the 'postal_code' column, and the 'address' column has a restrictive VARCHAR(255) type which can cause data truncation from the source TEXT type. Several other VARCHAR columns have different lengths but are compatible.",
  "compatible": false,
  "mismatches": [
    {
      "column": "postal_code",
      "source": "customers",
      "target": "secondory_customers",
      "severity": "critical",
      "description": "The column 'postal_code' is missing in the target table 'secondory_customers'.",
      "source_type": "VARCHAR(20)",
      "target_type": "Column not found",
      "migration_query": "ALTER TABLE secondory_customers ADD COLUMN postal_code VARCHAR(20);",
      "suggested_action": "Add the 'postal_code' column to the target table."
    },
    {
      "column": "address",
      "source": "customers",
      "target": "secondory_customers",
      "severity": "critical",
      "description": "The 'address' column type in the target is VARCHAR(255), which is more restrictive than the source type TEXT and can lead to data truncation.",
      "source_type": "TEXT",
      "target_type": "VARCHAR(255)",
      "migration_query": "ALTER TABLE secondory_customers ALTER COLUMN address TYPE TEXT;",
      "suggested_action": "Change the data type of the 'address' column in the target table to TEXT to match the source."
    },
    {
      "column": "customer_name",
      "source": "customers",
      "target": "secondory_customers",
      "severity": "info",
      "description": "The VARCHAR length for 'customer_name' differs, but the target type is larger, so it is compatible.",
      "source_type": "VARCHAR(100)",
      "target_type": "VARCHAR(255)",
      "migration_query": "Provide migration query",
      "suggested_action": "No action required, but for consistency you may want to align the varchar lengths."
    },
    {
      "column": "contact_name",
      "source": "customers",
      "target": "secondory_customers",
      "severity": "info",
      "description": "The VARCHAR length for 'contact_name' differs, but the target type is larger, so it is compatible.",
      "source_type": "VARCHAR(100)",
      "target_type": "VARCHAR(255)",
      "migration_query": "Provide migration query",
      "suggested_action": "No action required, but for consistency you may want to align the varchar lengths."
    },
    {
      "column": "city",
      "source": "customers",
      "target": "secondory_customers",
      "severity": "info",
      "description": "The VARCHAR length for 'city' differs, but the target type is larger, so it is compatible.",
      "source_type": "VARCHAR(50)",
      "target_type": "VARCHAR(100)",
      "migration_query": "Provide migration query",
      "suggested_action": "No action required, but for consistency you may want to align the varchar lengths."
    },
    {
      "column": "country",
      "source": "customers",
      "target": "secondory_customers",
      "severity": "info",
      "description": "The VARCHAR length for 'country' differs, but the target type is larger, so it is compatible.",
      "source_type": "VARCHAR(50)",
      "target_type": "VARCHAR(100)",
      "migration_query": "Provide migration query",
      "suggested_action": "No action required, but for consistency you may want to align the varchar lengths."
    }
  ]
}

---

Was generative AI tooling used to co-author this PR?

• Yes (please specify the tool below)

---

• Read the Pull Request Guidelines for more information. Note: commit author/co-author name and email in commits become permanently public when merged.
• For fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
• When adding dependency, check compliance with the ASF 3rd Party License Policy.
• For significant user-facing changes create newsfragment: {pr_number}.significant.rst or {issue_number}.significant.rst, in airflow-core/newsfragments.

[kaxil]

Thanks for addressing the first round of feedback — DataFusionEngine lazy import, Literal types, exception handling in _is_dbapi_connection, and the prompt duplication fix all look good. A few new issues from the changes.

[kaxil]

Much better. Round 2 issues all addressed: log format strings fixed, type-equivalence hints folded into DEFAULT_SYSTEM_PROMPT, conn_id restored in Source labels, AirflowException through compat, example DAGs cleaned up.

Two remaining items — one typo bug and one design question.

[kaxil]

LGTM with minor nits (already commented inline):

1. Missing ) in f-string at line 267 of the operator — ({dialect_name} should be ({dialect_name})
2. Docstring for system_prompt says "appended" but the behavior is "replaces" — update the wording
3. Warning logs for PK/FK/index failures should include table_name for context

[gopidesupavan]

LGTM with minor nits (already commented inline):

1. Missing ) in f-string at line 267 of the operator — ({dialect_name} should be ({dialect_name})
2. Docstring for system_prompt says "appended" but the behavior is "replaces" — update the wording
3. Warning logs for PK/FK/index failures should include table_name for context

thank you, if you would like to take a look one more time?