acryl-datahub 0.15.0rc15__py3-none-any.whl → 0.15.0rc16__py3-none-any.whl

datahub/ingestion/source/unity/usage.py

@@ -7,7 +7,6 @@ from typing import Any, Callable, Dict, Generic, Iterable, List, Optional, Set,
 
 import pyspark
 from databricks.sdk.service.sql import QueryStatementType
-from sqllineage.runner import LineageRunner
 
 from datahub.emitter.mcp import MetadataChangeProposalWrapper
 from datahub.ingestion.api.source_helpers import auto_empty_dataset_usage_statistics
@@ -22,7 +21,9 @@ from datahub.ingestion.source.unity.proxy_types import (
 from datahub.ingestion.source.unity.report import UnityCatalogReport
 from datahub.ingestion.source.usage.usage_common import UsageAggregator
 from datahub.metadata.schema_classes import OperationClass
+from datahub.sql_parsing.sqlglot_lineage import create_lineage_sql_parsed_result
 from datahub.sql_parsing.sqlglot_utils import get_query_fingerprint
+from datahub.utilities.urns.dataset_urn import DatasetUrn
 
 logger = logging.getLogger(__name__)
 
@@ -48,6 +49,7 @@ class UnityCatalogUsageExtractor:
     proxy: UnityCatalogApiProxy
     table_urn_builder: Callable[[TableReference], str]
     user_urn_builder: Callable[[str], str]
+    platform: str = "databricks"
 
     def __post_init__(self):
         self.usage_aggregator = UsageAggregator[TableReference](self.config)
@@ -173,7 +175,7 @@ class UnityCatalogUsageExtractor:
         self, query: Query, table_map: TableMap
     ) -> Optional[QueryTableInfo]:
         with self.report.usage_perf_report.sql_parsing_timer:
-            table_info = self._parse_query_via_lineage_runner(query.query_text)
+            table_info = self._parse_query_via_sqlglot(query.query_text)
             if table_info is None and query.statement_type == QueryStatementType.SELECT:
                 with self.report.usage_perf_report.spark_sql_parsing_timer:
                     table_info = self._parse_query_via_spark_sql_plan(query.query_text)
@@ -191,26 +193,33 @@ class UnityCatalogUsageExtractor:
                     ),
                 )
 
-    def _parse_query_via_lineage_runner(self, query: str) -> Optional[StringTableInfo]:
+    def _parse_query_via_sqlglot(self, query: str) -> Optional[StringTableInfo]:
         try:
-            runner = LineageRunner(query)
+            sql_parser_in_tables = create_lineage_sql_parsed_result(
+                query=query,
+                default_db=None,
+                platform=self.platform,
+                env=self.config.env,
+                platform_instance=None,
+            )
+
             return GenericTableInfo(
                 source_tables=[
-                    self._parse_sqllineage_table(table)
-                    for table in runner.source_tables
+                    self._parse_sqlglot_table(table)
+                    for table in sql_parser_in_tables.in_tables
                 ],
                 target_tables=[
-                    self._parse_sqllineage_table(table)
-                    for table in runner.target_tables
+                    self._parse_sqlglot_table(table)
+                    for table in sql_parser_in_tables.out_tables
                 ],
             )
         except Exception as e:
-            logger.info(f"Could not parse query via lineage runner, {query}: {e!r}")
+            logger.info(f"Could not parse query via sqlglot, {query}: {e!r}")
             return None
 
     @staticmethod
-    def _parse_sqllineage_table(sqllineage_table: object) -> str:
-        full_table_name = str(sqllineage_table)
+    def _parse_sqlglot_table(table_urn: str) -> str:
+        full_table_name = DatasetUrn.from_string(table_urn).name
         default_schema = "<default>."
         if full_table_name.startswith(default_schema):
             return full_table_name[len(default_schema) :]

datahub/utilities/partition_executor.py

@@ -268,7 +268,7 @@ class BatchPartitionExecutor(Closeable):
         self.process_batch = process_batch
         self.min_process_interval = min_process_interval
         self.read_from_pending_interval = read_from_pending_interval
-        assert self.max_workers > 1
+        assert self.max_workers >= 1
 
         self._state_lock = threading.Lock()
         self._executor = ThreadPoolExecutor(

datahub/utilities/sql_lineage_parser_impl.py

@@ -1,160 +0,0 @@
-import contextlib
-import logging
-import re
-import unittest
-import unittest.mock
-from typing import Dict, List, Optional, Set
-
-from sqllineage.core.holders import Column, SQLLineageHolder
-from sqllineage.exceptions import SQLLineageException
-
-from datahub.utilities.sql_parser_base import SQLParser, SqlParserException
-
-with contextlib.suppress(ImportError):
-    import sqlparse
-    from networkx import DiGraph
-    from sqllineage.core import LineageAnalyzer
-
-    import datahub.utilities.sqllineage_patch
-logger = logging.getLogger(__name__)
-
-
-class SqlLineageSQLParserImpl(SQLParser):
-    _DATE_SWAP_TOKEN = "__d_a_t_e"
-    _HOUR_SWAP_TOKEN = "__h_o_u_r"
-    _TIMESTAMP_SWAP_TOKEN = "__t_i_m_e_s_t_a_m_p"
-    _DATA_SWAP_TOKEN = "__d_a_t_a"
-    _ADMIN_SWAP_TOKEN = "__a_d_m_i_n"
-    _MYVIEW_SQL_TABLE_NAME_TOKEN = "__my_view__.__sql_table_name__"
-    _MYVIEW_LOOKER_TOKEN = "my_view.SQL_TABLE_NAME"
-
-    def __init__(self, sql_query: str, use_raw_names: bool = False) -> None:
-        super().__init__(sql_query)
-        original_sql_query = sql_query
-        self._use_raw_names = use_raw_names
-
-        # SqlLineageParser makes mistakes on lateral flatten queries, use the prefix
-        if "lateral flatten" in sql_query:
-            sql_query = sql_query[: sql_query.find("lateral flatten")]
-
-        # Replace reserved words that break SqlLineageParser
-        self.token_to_original: Dict[str, str] = {
-            self._DATE_SWAP_TOKEN: "date",
-            self._HOUR_SWAP_TOKEN: "hour",
-            self._TIMESTAMP_SWAP_TOKEN: "timestamp",
-            self._DATA_SWAP_TOKEN: "data",
-            self._ADMIN_SWAP_TOKEN: "admin",
-        }
-        for replacement, original in self.token_to_original.items():
-            # Replace original tokens with replacement. Since table and column name can contain a hyphen('-'),
-            # also prevent original tokens appearing as part of these names with a hyphen from getting substituted.
-            sql_query = re.sub(
-                rf"((?<!-)\b{original}\b)(?!-)",
-                rf"{replacement}",
-                sql_query,
-                flags=re.IGNORECASE,
-            )
-
-        # SqlLineageParser lowercarese tablenames and we need to replace Looker specific token which should be uppercased
-        sql_query = re.sub(
-            rf"(\${{{self._MYVIEW_LOOKER_TOKEN}}})",
-            rf"{self._MYVIEW_SQL_TABLE_NAME_TOKEN}",
-            sql_query,
-        )
-
-        # SqlLineageParser does not handle "encode" directives well. Remove them
-        sql_query = re.sub(r"\sencode [a-zA-Z]*", "", sql_query, flags=re.IGNORECASE)
-
-        # Replace lookml templates with the variable otherwise sqlparse can't parse ${
-        sql_query = re.sub(r"(\${)(.+)(})", r"\2", sql_query)
-        if sql_query != original_sql_query:
-            logger.debug(f"Rewrote original query {original_sql_query} as {sql_query}")
-
-        self._sql = sql_query
-        self._stmt_holders: Optional[List[LineageAnalyzer]] = None
-        self._sql_holder: Optional[SQLLineageHolder] = None
-        try:
-            self._stmt = [
-                s
-                for s in sqlparse.parse(
-                    # first apply sqlparser formatting just to get rid of comments, which cause
-                    # inconsistencies in parsing output
-                    sqlparse.format(
-                        self._sql.strip(),
-                        strip_comments=True,
-                        use_space_around_operators=True,
-                    ),
-                )
-                if s.token_first(skip_cm=True)
-            ]
-
-            with unittest.mock.patch(
-                "sqllineage.core.handlers.source.SourceHandler.end_of_query_cleanup",
-                datahub.utilities.sqllineage_patch.end_of_query_cleanup_patch,
-            ):
-                with unittest.mock.patch(
-                    "sqllineage.core.holders.SubQueryLineageHolder.add_column_lineage",
-                    datahub.utilities.sqllineage_patch.add_column_lineage_patch,
-                ):
-                    self._stmt_holders = [
-                        LineageAnalyzer().analyze(stmt) for stmt in self._stmt
-                    ]
-                    self._sql_holder = SQLLineageHolder.of(*self._stmt_holders)
-        except SQLLineageException as e:
-            raise SqlParserException(
-                f"SQL lineage analyzer error '{e}' for query: '{self._sql}"
-            ) from e
-
-    def get_tables(self) -> List[str]:
-        result: List[str] = []
-        if self._sql_holder is None:
-            logger.error("sql holder not present so cannot get tables")
-            return result
-        for table in self._sql_holder.source_tables:
-            table_normalized = re.sub(
-                r"^<default>.",
-                "",
-                (
-                    str(table)
-                    if not self._use_raw_names
-                    else f"{table.schema.raw_name}.{table.raw_name}"
-                ),
-            )
-            result.append(str(table_normalized))
-
-        # We need to revert TOKEN replacements
-        for token, replacement in self.token_to_original.items():
-            result = [replacement if c == token else c for c in result]
-        result = [
-            self._MYVIEW_LOOKER_TOKEN if c == self._MYVIEW_SQL_TABLE_NAME_TOKEN else c
-            for c in result
-        ]
-
-        # Sort tables to make the list deterministic
-        result.sort()
-
-        return result
-
-    def get_columns(self) -> List[str]:
-        if self._sql_holder is None:
-            raise SqlParserException("sql holder not present so cannot get columns")
-        graph: DiGraph = self._sql_holder.graph  # For mypy attribute checking
-        column_nodes = [n for n in graph.nodes if isinstance(n, Column)]
-        column_graph = graph.subgraph(column_nodes)
-
-        target_columns = {column for column, deg in column_graph.out_degree if deg == 0}
-
-        result: Set[str] = set()
-        for column in target_columns:
-            # Let's drop all the count(*) and similard columns which are expression actually if it does not have an alias
-            if not any(ele in column.raw_name for ele in ["*", "(", ")"]):
-                result.add(str(column.raw_name))
-
-        # Reverting back all the previously renamed words which confuses the parser
-        result = {"date" if c == self._DATE_SWAP_TOKEN else c for c in result}
-        result = {
-            "timestamp" if c == self._TIMESTAMP_SWAP_TOKEN else c for c in list(result)
-        }
-
-        # swap back renamed date column
-        return list(result)

datahub/utilities/sql_parser.py

@@ -1,94 +0,0 @@
-import logging
-import multiprocessing
-import traceback
-from multiprocessing import Process, Queue
-from typing import Any, List, Optional, Tuple
-
-from datahub.utilities.sql_lineage_parser_impl import SqlLineageSQLParserImpl
-from datahub.utilities.sql_parser_base import SQLParser
-
-logger = logging.getLogger(__name__)
-
-
-def sql_lineage_parser_impl_func_wrapper(
-    queue: Optional[multiprocessing.Queue], sql_query: str, use_raw_names: bool = False
-) -> Optional[Tuple[List[str], List[str], Any]]:
-    """
-    The wrapper function that computes the tables and columns using the SqlLineageSQLParserImpl
-    and puts the results on the shared IPC queue. This is used to isolate SqlLineageSQLParserImpl
-    functionality in a separate process, and hence protect our sources from memory leaks originating in
-    the sqllineage module.
-    :param queue: The shared IPC queue on to which the results will be put.
-    :param sql_query: The SQL query to extract the tables & columns from.
-    :param use_raw_names: Parameter used to ignore sqllineage's default lowercasing.
-    :return: None.
-    """
-    exception_details: Optional[Tuple[BaseException, str]] = None
-    tables: List[str] = []
-    columns: List[str] = []
-    try:
-        parser = SqlLineageSQLParserImpl(sql_query, use_raw_names)
-        tables = parser.get_tables()
-        columns = parser.get_columns()
-    except BaseException as e:
-        exc_msg = traceback.format_exc()
-        exception_details = (e, exc_msg)
-        logger.debug(exc_msg)
-
-    if queue is not None:
-        queue.put((tables, columns, exception_details))
-        return None
-    else:
-        return (tables, columns, exception_details)
-
-
-class SqlLineageSQLParser(SQLParser):
-    def __init__(
-        self,
-        sql_query: str,
-        use_external_process: bool = False,
-        use_raw_names: bool = False,
-    ) -> None:
-        super().__init__(sql_query, use_external_process)
-        if use_external_process:
-            self.tables, self.columns = self._get_tables_columns_process_wrapped(
-                sql_query, use_raw_names
-            )
-        else:
-            return_tuple = sql_lineage_parser_impl_func_wrapper(
-                None, sql_query, use_raw_names
-            )
-            if return_tuple is not None:
-                (
-                    self.tables,
-                    self.columns,
-                    some_exception,
-                ) = return_tuple
-
-    @staticmethod
-    def _get_tables_columns_process_wrapped(
-        sql_query: str, use_raw_names: bool = False
-    ) -> Tuple[List[str], List[str]]:
-        # Invoke sql_lineage_parser_impl_func_wrapper in a separate process to avoid
-        # memory leaks from sqllineage module used by SqlLineageSQLParserImpl. This will help
-        # shield our sources like lookml & redash, that need to parse a large number of SQL statements,
-        # from causing significant memory leaks in the datahub cli during ingestion.
-        queue: multiprocessing.Queue = Queue()
-        process: multiprocessing.Process = Process(
-            target=sql_lineage_parser_impl_func_wrapper,
-            args=(queue, sql_query, use_raw_names),
-        )
-        process.start()
-        tables, columns, exception_details = queue.get(block=True)
-        if exception_details is not None:
-            raise exception_details[0](f"Sub-process exception: {exception_details[1]}")
-        return tables, columns
-
-    def get_tables(self) -> List[str]:
-        return self.tables
-
-    def get_columns(self) -> List[str]:
-        return self.columns
-
-
-DefaultSQLParser = SqlLineageSQLParser

datahub/utilities/sql_parser_base.py

@@ -1,21 +0,0 @@
-from abc import ABCMeta, abstractmethod
-from typing import List
-
-
-class SqlParserException(Exception):
-    """Raised when sql parser fails"""
-
-    pass
-
-
-class SQLParser(metaclass=ABCMeta):
-    def __init__(self, sql_query: str, use_external_process: bool = True) -> None:
-        self._sql_query = sql_query
-
-    @abstractmethod
-    def get_tables(self) -> List[str]:
-        pass
-
-    @abstractmethod
-    def get_columns(self) -> List[str]:
-        pass