feat(ingestion/superset): superset column level lineage (#12786)

Author: PeteMango

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

@@ -23,6 +23,7 @@
     make_dataset_urn,
     make_dataset_urn_with_platform_instance,
     make_domain_urn,
+    make_schema_field_urn,
     make_user_urn,
 )
 from datahub.emitter.mcp_builder import add_domain_to_entity_wu
@@ -72,6 +73,9 @@
     DashboardInfoClass,
     DatasetLineageTypeClass,
     DatasetPropertiesClass,
+    FineGrainedLineageClass,
+    FineGrainedLineageDownstreamTypeClass,
+    FineGrainedLineageUpstreamTypeClass,
     GlobalTagsClass,
     OwnerClass,
     OwnershipClass,
@@ -80,6 +84,10 @@
     UpstreamClass,
     UpstreamLineageClass,
 )
+from datahub.sql_parsing.sqlglot_lineage import (
+    SqlParsingResult,
+    create_lineage_sql_parsed_result,
+)
 from datahub.utilities import config_clean
 from datahub.utilities.lossy_collections import LossyList
 from datahub.utilities.registries.domain_registry import DomainRegistry
@@ -342,7 +350,7 @@ def get_dataset_info(self, dataset_id: int) -> dict:
         )
         if dataset_response.status_code != 200:
             logger.warning(f"Failed to get dataset info: {dataset_response.text}")
-            dataset_response.raise_for_status()
+            return {}
         return dataset_response.json()
 
     def get_datasource_urn_from_id(
@@ -680,6 +688,88 @@ def gen_dataset_urn(self, datahub_dataset_name: str) -> str:
             env=self.config.env,
         )
 
+    def generate_virtual_dataset_lineage(
+        self,
+        parsed_query_object: SqlParsingResult,
+        datasource_urn: str,
+    ) -> UpstreamLineageClass:
+        cll = (
+            parsed_query_object.column_lineage
+            if parsed_query_object.column_lineage is not None
+            else []
+        )
+
+        fine_grained_lineages: List[FineGrainedLineageClass] = []
+
+        for cll_info in cll:
+            downstream = (
+                [make_schema_field_urn(datasource_urn, cll_info.downstream.column)]
+                if cll_info.downstream and cll_info.downstream.column
+                else []
+            )
+            upstreams = [
+                make_schema_field_urn(column_ref.table, column_ref.column)
+                for column_ref in cll_info.upstreams
+            ]
+            fine_grained_lineages.append(
+                FineGrainedLineageClass(
+                    downstreamType=FineGrainedLineageDownstreamTypeClass.FIELD,
+                    downstreams=downstream,
+                    upstreamType=FineGrainedLineageUpstreamTypeClass.FIELD_SET,
+                    upstreams=upstreams,
+                )
+            )
+
+        upstream_lineage = UpstreamLineageClass(
+            upstreams=[
+                UpstreamClass(
+                    type=DatasetLineageTypeClass.TRANSFORMED,
+                    dataset=input_table_urn,
+                )
+                for input_table_urn in parsed_query_object.in_tables
+            ],
+            fineGrainedLineages=fine_grained_lineages,
+        )
+        return upstream_lineage
+
+    def generate_physical_dataset_lineage(
+        self,
+        dataset_response: dict,
+        upstream_dataset: str,
+        datasource_urn: str,
+    ) -> UpstreamLineageClass:
+        # To generate column level lineage, we can manually decode the metadata
+        # to produce the ColumnLineageInfo
+        columns = dataset_response.get("result", {}).get("columns", [])
+        fine_grained_lineages: List[FineGrainedLineageClass] = []
+
+        for column in columns:
+            column_name = column.get("column_name", "")
+            if not column_name:
+                continue
+
+            downstream = [make_schema_field_urn(datasource_urn, column_name)]
+            upstreams = [make_schema_field_urn(upstream_dataset, column_name)]
+            fine_grained_lineages.append(
+                FineGrainedLineageClass(
+                    downstreamType=FineGrainedLineageDownstreamTypeClass.FIELD,
+                    downstreams=downstream,
+                    upstreamType=FineGrainedLineageUpstreamTypeClass.FIELD_SET,
+                    upstreams=upstreams,
+                )
+            )
+
+        upstream_lineage = UpstreamLineageClass(
+            upstreams=[
+                UpstreamClass(
+                    type=DatasetLineageTypeClass.TRANSFORMED,
+                    dataset=upstream_dataset,
+                )
+            ],
+            fineGrainedLineages=fine_grained_lineages,
+        )
+        return upstream_lineage
+
     def construct_dataset_from_dataset_data(
         self, dataset_data: dict
     ) -> DatasetSnapshot:
@@ -700,6 +790,14 @@ def construct_dataset_from_dataset_data(
         upstream_warehouse_platform = (
             dataset_response.get("result", {}).get("database", {}).get("backend")
         )
+        upstream_warehouse_db_name = (
+            dataset_response.get("result", {}).get("database", {}).get("database_name")
+        )
+
+        # if we have rendered sql, we always use that and defualt back to regular sql
+        sql = dataset_response.get("result", {}).get(
+            "rendered_sql"
+        ) or dataset_response.get("result", {}).get("sql")
 
         # Preset has a way of naming their platforms differently than
         # how datahub names them, so map the platform name to the correct naming
@@ -712,22 +810,28 @@ def construct_dataset_from_dataset_data(
         if upstream_warehouse_platform in warehouse_naming:
             upstream_warehouse_platform = warehouse_naming[upstream_warehouse_platform]
 
-        # TODO: Categorize physical vs virtual upstream dataset
-        # mark all upstream dataset as physical for now, in the future we would ideally like
-        # to differentiate physical vs virtual upstream datasets
-        tag_urn = f"urn:li:tag:{self.platform}:physical"
         upstream_dataset = self.get_datasource_urn_from_id(
             dataset_response, upstream_warehouse_platform
         )
-        upstream_lineage = UpstreamLineageClass(
-            upstreams=[
-                UpstreamClass(
-                    type=DatasetLineageTypeClass.TRANSFORMED,
-                    dataset=upstream_dataset,
-                    properties={"externalUrl": dataset_url},
-                )
-            ]
-        )
+
+        # Sometimes the field will be null instead of not existing
+        if sql == "null" or not sql:
+            tag_urn = f"urn:li:tag:{self.platform}:physical"
+            upstream_lineage = self.generate_physical_dataset_lineage(
+                dataset_response, upstream_dataset, datasource_urn
+            )
+        else:
+            tag_urn = f"urn:li:tag:{self.platform}:virtual"
+            parsed_query_object = create_lineage_sql_parsed_result(
+                query=sql,
+                default_db=upstream_warehouse_db_name,
+                platform=upstream_warehouse_platform,
+                platform_instance=None,
+                env=self.config.env,
+            )
+            upstream_lineage = self.generate_virtual_dataset_lineage(
+                parsed_query_object, datasource_urn
+            )
 
         dataset_info = DatasetPropertiesClass(
             name=dataset.table_name,