fix(ingest): avoid shell entities during view lineage generation

metadata-ingestion/src/datahub/ingestion/source/sql/sql_common.py

@@ -1197,6 +1197,8 @@ def _run_sql_parser(
             )
         else:
             self.report.num_view_definitions_parsed += 1
+            if raw_lineage.out_tables != [view_urn]:
+                self.report.num_view_definitions_view_urn_mismatch += 1
         return view_definition_lineage_helper(raw_lineage, view_urn)
 
     def get_db_schema(self, dataset_identifier: str) -> Tuple[Optional[str], str]:

metadata-ingestion/src/datahub/ingestion/source/sql/sql_report.py

@@ -48,6 +48,7 @@ class SQLSourceReport(
     query_combiner: Optional[SQLAlchemyQueryCombinerReport] = None
 
     num_view_definitions_parsed: int = 0
+    num_view_definitions_view_urn_mismatch: int = 0
     num_view_definitions_failed_parsing: int = 0
     num_view_definitions_failed_column_parsing: int = 0
     view_definitions_parsing_failures: LossyList[str] = field(default_factory=LossyList)

metadata-ingestion/src/datahub/ingestion/source/unity/source.py

@@ -974,6 +974,8 @@ def _run_sql_parser(
             )
         else:
             self.report.num_view_definitions_parsed += 1
+            if raw_lineage.out_tables != [view_urn]:
+                self.report.num_view_definitions_view_urn_mismatch += 1
         return view_definition_lineage_helper(raw_lineage, view_urn)
 
     def get_view_lineage(self) -> Iterable[MetadataWorkUnit]:

metadata-ingestion/src/datahub/sql_parsing/sqlglot_lineage.py

@@ -1243,13 +1243,19 @@ def infer_output_schema(result: SqlParsingResult) -> Optional[List[SchemaFieldCl
 def view_definition_lineage_helper(
     result: SqlParsingResult, view_urn: str
 ) -> SqlParsingResult:
-    if result.query_type is QueryType.SELECT:
+    if result.query_type is QueryType.SELECT or (
+        result.out_tables and result.out_tables != [view_urn]
+    ):
         # Some platforms (e.g. postgres) store only <select statement> from view definition
         # `create view V as <select statement>` . For such view definitions, `result.out_tables` and
         # `result.column_lineage[].downstream` are empty in `sqlglot_lineage` response, whereas upstream
         # details and downstream column details are extracted correctly.
         # Here, we inject view V's urn in `result.out_tables` and `result.column_lineage[].downstream`
         # to get complete lineage result.
+
+        # Some platforms(e.g. mssql) may have slightly different view name in view definition than
+        # actual view name used elsewhere. Therefore we overwrite downstream table for such cases as well.
+
         result.out_tables = [view_urn]
         if result.column_lineage:
             for col_result in result.column_lineage: