Preserve order in to_dataframe with BQ Storage from queries containing ORDER BY (#7793)

* Preserve order in `to_dataframe` with BQ Storage and queries containing ORDER BY

This fixes an issue where due to reading from multiple stream in
parallel, the order of rows is not preserved. Normally this isn't an
issue, but it is when the rows are query results from an ORDER BY query.

* Compile regex.

* Assert based on truthiness not equality.

Author: tswast

bigquery/google/cloud/bigquery/job.py

@@ -15,6 +15,7 @@
 """Define API Jobs."""
 
 import copy
+import re
 import threading
 
 import six
@@ -45,6 +46,7 @@
 _DONE_STATE = "DONE"
 _STOPPED_REASON = "stopped"
 _TIMEOUT_BUFFER_SECS = 0.1
+_CONTAINS_ORDER_BY = re.compile(r"ORDER\s+BY", re.IGNORECASE)
 
 _ERROR_REASON_TO_EXCEPTION = {
     "accessDenied": http_client.FORBIDDEN,
@@ -92,6 +94,29 @@ def _error_result_to_exception(error_result):
     )
 
 
+def _contains_order_by(query):
+    """Do we need to preserve the order of the query results?
+
+    This function has known false positives, such as with ordered window
+    functions:
+
+    .. code-block:: sql
+
+       SELECT SUM(x) OVER (
+           window_name
+           PARTITION BY...
+           ORDER BY...
+           window_frame_clause)
+       FROM ...
+
+    This false positive failure case means the behavior will be correct, but
+    downloading results with the BigQuery Storage API may be slower than it
+    otherwise would. This is preferable to the false negative case, where
+    results are expected to be in order but are not (due to parallel reads).
+    """
+    return query and _CONTAINS_ORDER_BY.search(query)
+
+
 class Compression(object):
     """The compression type to use for exported files. The default value is
     :attr:`NONE`.
@@ -2546,7 +2571,7 @@ def from_api_repr(cls, resource, client):
         :returns: Job parsed from ``resource``.
         """
         job_id, config = cls._get_resource_config(resource)
-        query = config["query"]["query"]
+        query = _helpers._get_sub_prop(config, ["query", "query"])
         job = cls(job_id, query, client=client)
         job._set_properties(resource)
         return job
@@ -2849,7 +2874,9 @@ def result(self, timeout=None, retry=DEFAULT_RETRY):
         dest_table_ref = self.destination
         dest_table = Table(dest_table_ref, schema=schema)
         dest_table._properties["numRows"] = self._query_results.total_rows
-        return self._client.list_rows(dest_table, retry=retry)
+        rows = self._client.list_rows(dest_table, retry=retry)
+        rows._preserve_order = _contains_order_by(self.query)
+        return rows
 
     def to_dataframe(self, bqstorage_client=None, dtypes=None, progress_bar_type=None):
         """Return a pandas DataFrame from a QueryJob

bigquery/google/cloud/bigquery/table.py

@@ -1348,6 +1348,7 @@ def __init__(
         )
         self._field_to_index = _helpers._field_to_index_mapping(schema)
         self._page_size = page_size
+        self._preserve_order = False
         self._project = client.project
         self._schema = schema
         self._selected_fields = selected_fields
@@ -1496,10 +1497,15 @@ def _to_dataframe_bqstorage(self, bqstorage_client, dtypes, progress_bar=None):
             for field in self._selected_fields:
                 read_options.selected_fields.append(field.name)
 
+        requested_streams = 0
+        if self._preserve_order:
+            requested_streams = 1
+
         session = bqstorage_client.create_read_session(
             self._table.to_bqstorage(),
             "projects/{}".format(self._project),
             read_options=read_options,
+            requested_streams=requested_streams,
         )
 
         # We need to parse the schema manually so that we can rearrange the
@@ -1512,6 +1518,8 @@ def _to_dataframe_bqstorage(self, bqstorage_client, dtypes, progress_bar=None):
         if not session.streams:
             return pandas.DataFrame(columns=columns)
 
+        total_streams = len(session.streams)
+
         # Use _to_dataframe_finished to notify worker threads when to quit.
         # See: https://stackoverflow.com/a/29237343/101923
         self._to_dataframe_finished = False
@@ -1560,7 +1568,7 @@ def get_frames(pool):
 
             return frames
 
-        with concurrent.futures.ThreadPoolExecutor() as pool:
+        with concurrent.futures.ThreadPoolExecutor(max_workers=total_streams) as pool:
             try:
                 frames = get_frames(pool)
             finally:

bigquery/tests/unit/test_job.py

@@ -17,6 +17,7 @@
 import unittest
 
 import mock
+import pytest
 from six.moves import http_client
 
 try:
@@ -59,6 +60,47 @@ def _make_connection(*responses):
     return mock_conn
 
 
+def _make_job_resource(
+    creation_time_ms=1437767599006,
+    started_time_ms=1437767600007,
+    ended_time_ms=1437767601008,
+    started=False,
+    ended=False,
+    etag="abc-def-hjk",
+    endpoint="https://www.googleapis.com",
+    job_type="load",
+    job_id="a-random-id",
+    project_id="some-project",
+    user_email="[email protected]",
+):
+    resource = {
+        "configuration": {job_type: {}},
+        "statistics": {"creationTime": creation_time_ms, job_type: {}},
+        "etag": etag,
+        "id": "{}:{}".format(project_id, job_id),
+        "jobReference": {"projectId": project_id, "jobId": job_id},
+        "selfLink": "{}/bigquery/v2/projects/{}/jobs/{}".format(
+            endpoint, project_id, job_id
+        ),
+        "user_email": user_email,
+    }
+
+    if started or ended:
+        resource["statistics"]["startTime"] = started_time_ms
+
+    if ended:
+        resource["statistics"]["endTime"] = ended_time_ms
+
+    if job_type == "query":
+        resource["configuration"]["query"]["destinationTable"] = {
+            "projectId": project_id,
+            "datasetId": "_temp_dataset",
+            "tableId": "_temp_table",
+        }
+
+    return resource
+
+
 class Test__error_result_to_exception(unittest.TestCase):
     def _call_fut(self, *args, **kwargs):
         from google.cloud.bigquery import job
@@ -974,6 +1016,7 @@ class _Base(object):
     from google.cloud.bigquery.dataset import DatasetReference
     from google.cloud.bigquery.table import TableReference
 
+    ENDPOINT = "https://www.googleapis.com"
     PROJECT = "project"
     SOURCE1 = "http://example.com/source1.csv"
     DS_ID = "dataset_id"
@@ -994,7 +1037,9 @@ def _setUpConstants(self):
         self.WHEN = datetime.datetime.utcfromtimestamp(self.WHEN_TS).replace(tzinfo=UTC)
         self.ETAG = "ETAG"
         self.FULL_JOB_ID = "%s:%s" % (self.PROJECT, self.JOB_ID)
-        self.RESOURCE_URL = "http://example.com/path/to/resource"
+        self.RESOURCE_URL = "{}/bigquery/v2/projects/{}/jobs/{}".format(
+            self.ENDPOINT, self.PROJECT, self.JOB_ID
+        )
         self.USER_EMAIL = "[email protected]"
 
     def _table_ref(self, table_id):
@@ -1004,30 +1049,19 @@ def _table_ref(self, table_id):
 
     def _make_resource(self, started=False, ended=False):
         self._setUpConstants()
-        resource = {
-            "configuration": {self.JOB_TYPE: {}},
-            "statistics": {"creationTime": self.WHEN_TS * 1000, self.JOB_TYPE: {}},
-            "etag": self.ETAG,
-            "id": self.FULL_JOB_ID,
-            "jobReference": {"projectId": self.PROJECT, "jobId": self.JOB_ID},
-            "selfLink": self.RESOURCE_URL,
-            "user_email": self.USER_EMAIL,
-        }
-
-        if started or ended:
-            resource["statistics"]["startTime"] = self.WHEN_TS * 1000
-
-        if ended:
-            resource["statistics"]["endTime"] = (self.WHEN_TS + 1000) * 1000
-
-        if self.JOB_TYPE == "query":
-            resource["configuration"]["query"]["destinationTable"] = {
-                "projectId": self.PROJECT,
-                "datasetId": "_temp_dataset",
-                "tableId": "_temp_table",
-            }
-
-        return resource
+        return _make_job_resource(
+            creation_time_ms=int(self.WHEN_TS * 1000),
+            started_time_ms=int(self.WHEN_TS * 1000),
+            ended_time_ms=int(self.WHEN_TS * 1000) + 1000000,
+            started=started,
+            ended=ended,
+            etag=self.ETAG,
+            endpoint=self.ENDPOINT,
+            job_type=self.JOB_TYPE,
+            job_id=self.JOB_ID,
+            project_id=self.PROJECT,
+            user_email=self.USER_EMAIL,
+        )
 
     def _verifyInitialReadonlyProperties(self, job):
         # root elements of resource
@@ -4684,7 +4718,11 @@ def test_to_dataframe_bqstorage(self):
         job.to_dataframe(bqstorage_client=bqstorage_client)
 
         bqstorage_client.create_read_session.assert_called_once_with(
-            mock.ANY, "projects/{}".format(self.PROJECT), read_options=mock.ANY
+            mock.ANY,
+            "projects/{}".format(self.PROJECT),
+            read_options=mock.ANY,
+            # Use default number of streams for best performance.
+            requested_streams=0,
         )
 
     @unittest.skipIf(pandas is None, "Requires `pandas`")
@@ -5039,3 +5077,93 @@ def test_from_api_repr_normal(self):
         self.assertEqual(entry.pending_units, self.PENDING_UNITS)
         self.assertEqual(entry.completed_units, self.COMPLETED_UNITS)
         self.assertEqual(entry.slot_millis, self.SLOT_MILLIS)
+
+
+@pytest.mark.parametrize(
+    "query,expected",
+    (
+        (None, False),
+        ("", False),
+        ("select name, age from table", False),
+        ("select name, age from table LIMIT 10;", False),
+        ("select name, age from table order by other_column;", True),
+        ("Select name, age From table Order By other_column", True),
+        ("SELECT name, age FROM table ORDER BY other_column;", True),
+        ("select name, age from table order\nby other_column", True),
+        ("Select name, age From table Order\nBy other_column;", True),
+        ("SELECT name, age FROM table ORDER\nBY other_column", True),
+        ("SelecT name, age froM table OrdeR \n\t BY other_column;", True),
+    ),
+)
+def test__contains_order_by(query, expected):
+    from google.cloud.bigquery import job as mut
+
+    if expected:
+        assert mut._contains_order_by(query)
+    else:
+        assert not mut._contains_order_by(query)
+
+
+@pytest.mark.skipif(pandas is None, reason="Requires `pandas`")
+@pytest.mark.skipif(
+    bigquery_storage_v1beta1 is None, reason="Requires `google-cloud-bigquery-storage`"
+)
+@pytest.mark.parametrize(
+    "query",
+    (
+        "select name, age from table order by other_column;",
+        "Select name, age From table Order By other_column;",
+        "SELECT name, age FROM table ORDER BY other_column;",
+        "select name, age from table order\nby other_column;",
+        "Select name, age From table Order\nBy other_column;",
+        "SELECT name, age FROM table ORDER\nBY other_column;",
+        "SelecT name, age froM table OrdeR \n\t BY other_column;",
+    ),
+)
+def test_to_dataframe_bqstorage_preserve_order(query):
+    from google.cloud.bigquery.job import QueryJob as target_class
+
+    job_resource = _make_job_resource(
+        project_id="test-project", job_type="query", ended=True
+    )
+    job_resource["configuration"]["query"]["query"] = query
+    job_resource["status"] = {"state": "DONE"}
+    get_query_results_resource = {
+        "jobComplete": True,
+        "jobReference": {"projectId": "test-project", "jobId": "test-job"},
+        "schema": {
+            "fields": [
+                {"name": "name", "type": "STRING", "mode": "NULLABLE"},
+                {"name": "age", "type": "INTEGER", "mode": "NULLABLE"},
+            ]
+        },
+        "totalRows": "4",
+    }
+    connection = _make_connection(get_query_results_resource, job_resource)
+    client = _make_client(connection=connection)
+    job = target_class.from_api_repr(job_resource, client)
+    bqstorage_client = mock.create_autospec(
+        bigquery_storage_v1beta1.BigQueryStorageClient
+    )
+    session = bigquery_storage_v1beta1.types.ReadSession()
+    session.avro_schema.schema = json.dumps(
+        {
+            "type": "record",
+            "name": "__root__",
+            "fields": [
+                {"name": "name", "type": ["null", "string"]},
+                {"name": "age", "type": ["null", "long"]},
+            ],
+        }
+    )
+    bqstorage_client.create_read_session.return_value = session
+
+    job.to_dataframe(bqstorage_client=bqstorage_client)
+
+    bqstorage_client.create_read_session.assert_called_once_with(
+        mock.ANY,
+        "projects/test-project",
+        read_options=mock.ANY,
+        # Use a single stream to preserve row order.
+        requested_streams=1,
+    )