fix(ingest/bigquery): fix lineage if multiple sql expression passed i…

…n and destination table set (datahub-project#10212)

Co-authored-by: Harshal Sheth <[email protected]>

metadata-ingestion/src/datahub/ingestion/source/bigquery_v2/bigquery_audit.py

@@ -81,7 +81,9 @@ def get_table_and_shard(table_name: str) -> Tuple[Optional[str], Optional[str]]:
     @classmethod
     def from_string_name(cls, table: str) -> "BigqueryTableIdentifier":
         parts = table.split(".")
-        return cls(parts[0], parts[1], parts[2])
+        # If the table name contains dollar sign, it is a referrence to a partitioned table and we have to strip it
+        table = parts[2].split("$", 1)[0]
+        return cls(parts[0], parts[1], table)
 
     def raw_table_name(self):
         return f"{self.project_id}.{self.dataset}.{self.table}"

metadata-ingestion/src/datahub/ingestion/source/bigquery_v2/lineage.py

@@ -17,6 +17,7 @@
 )
 
 import humanfriendly
+import sqlglot
 from google.cloud.datacatalog import lineage_v1
 from google.cloud.logging_v2.client import Client as GCPLoggingClient
 
@@ -737,20 +738,37 @@ def _create_lineage_map(
 
             # Try the sql parser first.
             if self.config.lineage_use_sql_parser:
+                logger.debug(
+                    f"Using sql parser for lineage extraction for destination table: {destination_table.table_identifier.get_table_name()}, queryType: {e.statementType}, query: {e.query}"
+                )
                 if e.statementType == "SELECT":
                     # We wrap select statements in a CTE to make them parseable as insert statement.
                     # This is a workaround for the sql parser to support the case where the user runs a query and inserts the result into a table..
-                    query = f"""create table `{destination_table.table_identifier.get_table_name()}` AS
-                    (
-                        {e.query}
-                    )"""
+                    try:
+                        parsed_queries = sqlglot.parse(e.query, "bigquery")
+                        if parsed_queries[-1]:
+                            query = f"""create table `{destination_table.get_sanitized_table_ref().table_identifier.get_table_name()}` AS
+                            (
+                                {parsed_queries[-1].sql(dialect='bigquery')}
+                            )"""
+                        else:
+                            query = e.query
+                    except Exception:
+                        logger.debug(
+                            f"Failed to parse select-based lineage query {e.query} for table {destination_table}."
+                            "Sql parsing will likely fail for this query, which will result in a fallback to audit log."
+                        )
+                        query = e.query
                 else:
                     query = e.query
                 raw_lineage = sqlglot_lineage(
                     query,
                     schema_resolver=sql_parser_schema_resolver,
                     default_db=e.project_id,
                 )
+                logger.debug(
+                    f"Input tables: {raw_lineage.in_tables}, Output tables: {raw_lineage.out_tables}"
+                )
                 if raw_lineage.debug_info.table_error:
                     logger.debug(
                         f"Sql Parser failed on query: {e.query}. It won't cause any major issues, but "

metadata-ingestion/src/datahub/ingestion/source/bigquery_v2/usage.py

@@ -204,7 +204,7 @@ def usage_statistics_query(top_n: int) -> str:
                     r.resource,
                     q.query,
                     COUNT(r.key) as query_count,
-                    ROW_NUMBER() over (PARTITION BY r.timestamp, r.resource, q.query ORDER BY COUNT(r.key) DESC, q.query) as rank
+                    ROW_NUMBER() over (PARTITION BY r.timestamp, r.resource ORDER BY COUNT(r.key) DESC) as rank
                 FROM
                     read_events r
                     INNER JOIN query_events q ON r.name = q.key
@@ -256,6 +256,7 @@ class UsageStatistic:
 
     def usage_statistics(self, top_n: int) -> Iterator[UsageStatistic]:
         query = self.usage_statistics_query(top_n)
+
         rows = self.read_events.sql_query_iterator(
             query, refs=[self.query_events, self.column_accesses]
         )

metadata-ingestion/src/datahub/ingestion/source/usage/usage_common.py

@@ -74,6 +74,12 @@ def make_usage_workunit(
 
     top_sql_queries: Optional[List[str]] = None
     if query_freq is not None:
+        if top_n_queries < len(query_freq):
+            logger.warn(
+                f"Top N query limit exceeded on {str(resource)}.  Max number of queries {top_n_queries} <  {len(query_freq)}. Truncating top queries to {top_n_queries}."
+            )
+            query_freq = query_freq[0:top_n_queries]
+
         budget_per_query: int = int(queries_character_limit / top_n_queries)
         top_sql_queries = [
             trim_query(

metadata-ingestion/tests/unit/test_bigquery_usage.py

@@ -190,7 +190,7 @@ def config() -> BigQueryV2Config:
         end_time=TS_2 + timedelta(minutes=1),
         usage=BigQueryUsageConfig(
             include_top_n_queries=True,
-            top_n_queries=3,
+            top_n_queries=30,
             bucket_duration=BucketDuration.DAY,
             include_operational_stats=False,
         ),
@@ -878,6 +878,125 @@ def test_usage_counts_no_columns(
         assert not caplog.records
 
 
+def test_usage_counts_no_columns_and_top_n_limit_hit(
+    caplog: pytest.LogCaptureFixture,
+    usage_extractor: BigQueryUsageExtractor,
+    config: BigQueryV2Config,
+) -> None:
+    config.usage.top_n_queries = 1
+
+    job_name = "job_name"
+    ref = BigQueryTableRef(
+        BigqueryTableIdentifier(PROJECT_1, DATABASE_1.name, TABLE_1.name)
+    )
+    events = [
+        AuditEvent.create(
+            ReadEvent(
+                jobName=job_name,
+                timestamp=TS_1,
+                actor_email=ACTOR_1,
+                resource=ref,
+                fieldsRead=[],
+                readReason="JOB",
+                payload=None,
+            ),
+        ),
+        AuditEvent.create(
+            ReadEvent(
+                jobName="job_name_2",
+                timestamp=TS_1,
+                actor_email=ACTOR_1,
+                resource=ref,
+                fieldsRead=[],
+                readReason="JOB",
+                payload=None,
+            ),
+        ),
+        AuditEvent.create(
+            ReadEvent(
+                jobName="job_name_3",
+                timestamp=TS_1,
+                actor_email=ACTOR_1,
+                resource=ref,
+                fieldsRead=[],
+                readReason="JOB",
+                payload=None,
+            ),
+        ),
+        AuditEvent.create(
+            QueryEvent(
+                job_name=job_name,
+                timestamp=TS_1,
+                actor_email=ACTOR_1,
+                query="SELECT * FROM table_1",
+                statementType="SELECT",
+                project_id=PROJECT_1,
+                destinationTable=None,
+                referencedTables=[ref],
+                referencedViews=[],
+                payload=None,
+            )
+        ),
+        AuditEvent.create(
+            QueryEvent(
+                job_name="job_name_2",
+                timestamp=TS_1,
+                actor_email=ACTOR_1,
+                query="SELECT * FROM table_1",
+                statementType="SELECT",
+                project_id=PROJECT_1,
+                destinationTable=None,
+                referencedTables=[ref],
+                referencedViews=[],
+                payload=None,
+            )
+        ),
+        AuditEvent.create(
+            QueryEvent(
+                job_name="job_name_3",
+                timestamp=TS_1,
+                actor_email=ACTOR_1,
+                query="SELECT my_column FROM table_1",
+                statementType="SELECT",
+                project_id=PROJECT_1,
+                destinationTable=None,
+                referencedTables=[ref],
+                referencedViews=[],
+                payload=None,
+            )
+        ),
+    ]
+    caplog.clear()
+    with caplog.at_level(logging.WARNING):
+        workunits = usage_extractor._get_workunits_internal(
+            events, [TABLE_REFS[TABLE_1.name]]
+        )
+        expected = [
+            make_usage_workunit(
+                table=TABLE_1,
+                dataset_usage_statistics=DatasetUsageStatisticsClass(
+                    timestampMillis=int(TS_1.timestamp() * 1000),
+                    eventGranularity=TimeWindowSizeClass(
+                        unit=BucketDuration.DAY, multiple=1
+                    ),
+                    totalSqlQueries=3,
+                    topSqlQueries=["SELECT * FROM table_1"],
+                    uniqueUserCount=1,
+                    userCounts=[
+                        DatasetUserUsageCountsClass(
+                            user=ACTOR_1_URN,
+                            count=3,
+                            userEmail=ACTOR_1,
+                        ),
+                    ],
+                    fieldCounts=[],
+                ),
+            )
+        ]
+        compare_workunits(workunits, expected)
+        assert not caplog.records
+
+
 @freeze_time(FROZEN_TIME)
 @patch.object(BigQueryUsageExtractor, "_generate_usage_workunits")
 def test_operational_stats(