test_driver.py

import itertools
from contextlib import contextmanager
from collections import OrderedDict
import datetime
import io
import logging
import os
import sys
import threading
import time
from unittest import loader, skipIf, skipUnless, TestCase
from uuid import uuid4

import numpy as np
import pyarrow
import pytz
import thrift
import pytest
from urllib3.connectionpool import ReadTimeoutError

import databricks.sql as sql
from databricks.sql import (
    STRING,
    BINARY,
    NUMBER,
    DATETIME,
    DATE,
    DatabaseError,
    Error,
    OperationalError,
    RequestError,
)
from tests.e2e.common.predicates import (
    pysql_has_version,
    pysql_supports_arrow,
    compare_dbr_versions,
    is_thrift_v5_plus,
)
from databricks.sql.thrift_api.TCLIService import ttypes
from tests.e2e.common.core_tests import CoreTestMixin, SmokeTestMixin
from tests.e2e.common.large_queries_mixin import LargeQueriesMixin
from tests.e2e.common.timestamp_tests import TimestampTestsMixin
from tests.e2e.common.decimal_tests import DecimalTestsMixin
from tests.e2e.common.retry_test_mixins import (
    Client429ResponseMixin,
    Client503ResponseMixin,
)
from tests.e2e.common.staging_ingestion_tests import PySQLStagingIngestionTestSuiteMixin
from tests.e2e.common.retry_test_mixins import PySQLRetryTestsMixin

from tests.e2e.common.uc_volume_tests import PySQLUCVolumeTestSuiteMixin

from databricks.sql.exc import SessionAlreadyClosedError

log = logging.getLogger(__name__)

unsafe_logger = logging.getLogger("databricks.sql.unsafe")
unsafe_logger.setLevel(logging.DEBUG)
unsafe_logger.addHandler(logging.FileHandler("./tests-unsafe.log"))

# manually decorate DecimalTestsMixin to need arrow support
for name in loader.getTestCaseNames(DecimalTestsMixin, "test_"):
    fn = getattr(DecimalTestsMixin, name)
    decorated = skipUnless(pysql_supports_arrow(), "Decimal tests need arrow support")(
        fn
    )
    setattr(DecimalTestsMixin, name, decorated)


class PySQLPytestTestCase:
    """A mirror of PySQLTest case that doesn't inherit from unittest.TestCase
    so that we can use pytest.mark.parameterize
    """

    error_type = Error
    conf_to_disable_rate_limit_retries = {"_retry_stop_after_attempts_count": 1}
    conf_to_disable_temporarily_unavailable_retries = {
        "_retry_stop_after_attempts_count": 1
    }
    arraysize = 1000
    buffer_size_bytes = 104857600
    POLLING_INTERVAL = 2

    @pytest.fixture(autouse=True)
    def get_details(self, connection_details):
        self.arguments = connection_details.copy()

    def connection_params(self):
        params = {
            "server_hostname": self.arguments["host"],
            "http_path": self.arguments["http_path"],
            **self.auth_params(),
        }

        return params

    def auth_params(self):
        return {
            "access_token": self.arguments.get("access_token"),
        }

    @contextmanager
    def connection(self, extra_params=()):
        connection_params = dict(self.connection_params(), **dict(extra_params))

        log.info("Connecting with args: {}".format(connection_params))
        conn = sql.connect(**connection_params)

        try:
            yield conn
        finally:
            conn.close()

    @contextmanager
    def cursor(self, extra_params=()):
        with self.connection(extra_params) as conn:
            cursor = conn.cursor(
                arraysize=self.arraysize, buffer_size_bytes=self.buffer_size_bytes
            )
            try:
                yield cursor
            finally:
                cursor.close()

    def assertEqualRowValues(self, actual, expected):
        len_actual = len(actual) if actual else 0
        len_expected = len(expected) if expected else 0
        assert len_actual == len_expected
        for act, exp in zip(actual, expected):
            assert len(act) == len(exp)
            for i in range(len(act)):
                assert act[i] == exp[i]


class TestPySQLLargeQueriesSuite(PySQLPytestTestCase, LargeQueriesMixin):
    def get_some_rows(self, cursor, fetchmany_size):
        row = cursor.fetchone()
        if row:
            return [row]
        else:
            return None

    @skipUnless(pysql_supports_arrow(), "needs arrow support")
    @pytest.mark.skip("This test requires a previously uploaded data set")
    def test_cloud_fetch(self):
        # This test can take several minutes to run
        limits = [100000, 300000]
        threads = [10, 25]
        self.arraysize = 100000
        # This test requires a large table with many rows to properly initiate cloud fetch.
        # e2-dogfood host > hive_metastore catalog > main schema has such a table called store_sales.
        # If this table is deleted or this test is run on a different host, a different table may need to be used.
        base_query = "SELECT * FROM store_sales WHERE ss_sold_date_sk = 2452234 "
        for num_limit, num_threads, lz4_compression in itertools.product(
            limits, threads, [True, False]
        ):
            with self.subTest(
                num_limit=num_limit,
                num_threads=num_threads,
                lz4_compression=lz4_compression,
            ):
                cf_result, noop_result = None, None
                query = base_query + "LIMIT " + str(num_limit)
                with self.cursor(
                    {
                        "use_cloud_fetch": True,
                        "max_download_threads": num_threads,
                        "catalog": "hive_metastore",
                    },
                ) as cursor:
                    cursor.execute(query)
                    cf_result = cursor.fetchall()
                with self.cursor({"catalog": "hive_metastore"}) as cursor:
                    cursor.execute(query)
                    noop_result = cursor.fetchall()
                assert len(cf_result) == len(noop_result)
                for i in range(len(cf_result)):
                    assert cf_result[i] == noop_result[i]


class TestPySQLAsyncQueriesSuite(PySQLPytestTestCase):
    def test_execute_async__long_running(self):

        long_running_query = "SELECT COUNT(*) FROM RANGE(10000 * 16) x JOIN RANGE(10000) y ON FROM_UNIXTIME(x.id * y.id, 'yyyy-MM-dd') LIKE '%not%a%date%'"
        with self.cursor() as cursor:
            cursor.execute_async(long_running_query)

            ## Polling after every POLLING_INTERVAL seconds
            while cursor.is_query_pending():
                time.sleep(self.POLLING_INTERVAL)
                log.info("Polling the status in test_execute_async")

            cursor.get_async_execution_result()
            result = cursor.fetchall()

            assert result[0].asDict() == {"count(1)": 0}

    def test_execute_async__small_result(self):
        small_result_query = "SELECT 1"

        with self.cursor() as cursor:
            cursor.execute_async(small_result_query)

            ## Fake sleep for 5 secs
            time.sleep(5)

            ## Polling after every POLLING_INTERVAL seconds
            while cursor.is_query_pending():
                time.sleep(self.POLLING_INTERVAL)
                log.info("Polling the status in test_execute_async")

            cursor.get_async_execution_result()
            result = cursor.fetchall()

            assert result[0].asDict() == {"1": 1}

    def test_execute_async__large_result(self):
        x_dimension = 1000
        y_dimension = 1000
        large_result_query = f"""
                SELECT
                    x.id AS x_id,
                    y.id AS y_id,
                    FROM_UNIXTIME(x.id * y.id, 'yyyy-MM-dd') AS date
                FROM
                    RANGE({x_dimension}) x
                JOIN
                    RANGE({y_dimension}) y
            """

        with self.cursor() as cursor:
            cursor.execute_async(large_result_query)

            ## Fake sleep for 5 secs
            time.sleep(5)

            ## Polling after every POLLING_INTERVAL seconds
            while cursor.is_query_pending():
                time.sleep(self.POLLING_INTERVAL)
                log.info("Polling the status in test_execute_async")

            cursor.get_async_execution_result()
            result = cursor.fetchall()

            assert len(result) == x_dimension * y_dimension


# Exclude Retry tests because they require specific setups, and LargeQueries too slow for core
# tests
class TestPySQLCoreSuite(
    SmokeTestMixin,
    CoreTestMixin,
    DecimalTestsMixin,
    TimestampTestsMixin,
    PySQLPytestTestCase,
    PySQLStagingIngestionTestSuiteMixin,
    PySQLRetryTestsMixin,
    PySQLUCVolumeTestSuiteMixin,
):
    validate_row_value_type = True
    validate_result = True

    # An output column in description evaluates to equal to multiple types
    # - type code returned by the client as string.
    # - also potentially a PEP-249 object like NUMBER, DATETIME etc.
    def expected_column_types(self, type_):
        type_mappings = {
            "boolean": ["boolean", NUMBER],
            "byte": ["tinyint", NUMBER],
            "short": ["smallint", NUMBER],
            "integer": ["int", NUMBER],
            "long": ["bigint", NUMBER],
            "decimal": ["decimal", NUMBER],
            "timestamp": ["timestamp", DATETIME],
            "date": ["date", DATE],
            "binary": ["binary", BINARY],
            "string": ["string", STRING],
            "array": ["array"],
            "struct": ["struct"],
            "map": ["map"],
            "double": ["double", NUMBER],
            "null": ["null"],
        }
        return type_mappings[type_]

    def test_queries(self):
        if not self._should_have_native_complex_types():
            array_type = str
            array_val = "[1,2,3]"
            struct_type = str
            struct_val = '{"a":1,"b":2}'
            map_type = str
            map_val = "{1:2,3:4}"
        else:
            array_type = np.ndarray
            array_val = np.array([1, 2, 3])
            struct_type = dict
            struct_val = {"a": 1, "b": 2}
            map_type = list
            map_val = [(1, 2), (3, 4)]

        null_type = "null" if float(sql.__version__[0:2]) < 2.0 else "string"
        self.range_queries = CoreTestMixin.range_queries + [
            ("NULL", null_type, type(None), None),
            ("array(1, 2, 3)", "array", array_type, array_val),
            ("struct(1 as a, 2 as b)", "struct", struct_type, struct_val),
            ("map(1, 2, 3, 4)", "map", map_type, map_val),
        ]

        self.run_tests_on_queries({})

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_incorrect_query_throws_exception(self):
        with self.cursor({}) as cursor:
            # Syntax errors should contain the invalid SQL
            with pytest.raises(DatabaseError) as cm:
                cursor.execute("^ FOO BAR")
            assert "FOO BAR" in str(cm.value)

            # Database error should contain the missing database
            with pytest.raises(DatabaseError) as cm:
                cursor.execute("USE foo234823498ydfsiusdhf")
            assert "foo234823498ydfsiusdhf" in str(cm.value)

            # SQL with Extraneous input should send back the extraneous input
            with pytest.raises(DatabaseError) as cm:
                cursor.execute("CREATE TABLE IF NOT EXISTS TABLE table_234234234")
            assert "table_234234234" in str(cm.value)

    def test_create_table_will_return_empty_result_set(self):
        with self.cursor({}) as cursor:
            table_name = "table_{uuid}".format(uuid=str(uuid4()).replace("-", "_"))
            try:
                cursor.execute(
                    "CREATE TABLE IF NOT EXISTS {} AS (SELECT 1 AS col_1, '2' AS col_2)".format(
                        table_name
                    )
                )
                assert cursor.fetchall() == []
            finally:
                cursor.execute("DROP TABLE IF EXISTS {}".format(table_name))

    def test_get_tables(self):
        with self.cursor({}) as cursor:
            table_name = "table_{uuid}".format(uuid=str(uuid4()).replace("-", "_"))
            table_names = [table_name + "_1", table_name + "_2"]

            try:
                for table in table_names:
                    cursor.execute(
                        "CREATE TABLE IF NOT EXISTS {} AS (SELECT 1 AS col_1, '2' AS col_2)".format(
                            table
                        )
                    )
                cursor.tables(schema_name="defa%")
                tables = cursor.fetchall()
                tables_desc = cursor.description

                for table in table_names:
                    # Test only schema name and table name.
                    # From other columns, what is supported depends on DBR version.
                    assert ["default", table] in [list(table[1:3]) for table in tables]
                expected = [
                    ("TABLE_CAT", "string", None, None, None, None, None),
                    ("TABLE_SCHEM", "string", None, None, None, None, None),
                    ("TABLE_NAME", "string", None, None, None, None, None),
                    ("TABLE_TYPE", "string", None, None, None, None, None),
                    ("REMARKS", "string", None, None, None, None, None),
                    ("TYPE_CAT", "string", None, None, None, None, None),
                    ("TYPE_SCHEM", "string", None, None, None, None, None),
                    ("TYPE_NAME", "string", None, None, None, None, None),
                    (
                        "SELF_REFERENCING_COL_NAME",
                        "string",
                        None,
                        None,
                        None,
                        None,
                        None,
                    ),
                    ("REF_GENERATION", "string", None, None, None, None, None),
                ]
                assert tables_desc == expected

            finally:
                for table in table_names:
                    cursor.execute("DROP TABLE IF EXISTS {}".format(table))

    def test_get_columns(self):
        with self.cursor({}) as cursor:
            table_name = "table_{uuid}".format(uuid=str(uuid4()).replace("-", "_"))
            table_names = [table_name + "_1", table_name + "_2"]

            try:
                for table in table_names:
                    cursor.execute(
                        "CREATE TABLE IF NOT EXISTS {} AS (SELECT "
                        "1 AS col_1, "
                        "'2' AS col_2, "
                        "named_struct('name', 'alice', 'age', 28) as col_3, "
                        "map('items', 45, 'cost', 228) as col_4, "
                        "array('item1', 'item2', 'item3') as col_5)".format(table)
                    )

                cursor.columns(schema_name="defa%", table_name=table_name + "%")
                cols = cursor.fetchall()
                cols_desc = cursor.description

                # Catalogue name not consistent across DBR versions, so we skip that
                cleaned_response = [list(col[1:6]) for col in cols]
                # We also replace ` as DBR changes how it represents struct names
                for col in cleaned_response:
                    col[4] = col[4].replace("`", "")

                expected = [
                    ["default", table_name + "_1", "col_1", 4, "INT"],
                    ["default", table_name + "_1", "col_2", 12, "STRING"],
                    [
                        "default",
                        table_name + "_1",
                        "col_3",
                        2002,
                        "STRUCT<name: STRING, age: INT>",
                    ],
                    ["default", table_name + "_1", "col_4", 2000, "MAP<STRING, INT>"],
                    ["default", table_name + "_1", "col_5", 2003, "ARRAY<STRING>"],
                    ["default", table_name + "_2", "col_1", 4, "INT"],
                    ["default", table_name + "_2", "col_2", 12, "STRING"],
                    [
                        "default",
                        table_name + "_2",
                        "col_3",
                        2002,
                        "STRUCT<name: STRING, age: INT>",
                    ],
                    ["default", table_name + "_2", "col_4", 2000, "MAP<STRING, INT>"],
                    [
                        "default",
                        table_name + "_2",
                        "col_5",
                        2003,
                        "ARRAY<STRING>",
                    ],
                ]
                assert cleaned_response == expected
                expected = [
                    ("TABLE_CAT", "string", None, None, None, None, None),
                    ("TABLE_SCHEM", "string", None, None, None, None, None),
                    ("TABLE_NAME", "string", None, None, None, None, None),
                    ("COLUMN_NAME", "string", None, None, None, None, None),
                    ("DATA_TYPE", "int", None, None, None, None, None),
                    ("TYPE_NAME", "string", None, None, None, None, None),
                    ("COLUMN_SIZE", "int", None, None, None, None, None),
                    ("BUFFER_LENGTH", "tinyint", None, None, None, None, None),
                    ("DECIMAL_DIGITS", "int", None, None, None, None, None),
                    ("NUM_PREC_RADIX", "int", None, None, None, None, None),
                    ("NULLABLE", "int", None, None, None, None, None),
                    ("REMARKS", "string", None, None, None, None, None),
                    ("COLUMN_DEF", "string", None, None, None, None, None),
                    ("SQL_DATA_TYPE", "int", None, None, None, None, None),
                    ("SQL_DATETIME_SUB", "int", None, None, None, None, None),
                    ("CHAR_OCTET_LENGTH", "int", None, None, None, None, None),
                    ("ORDINAL_POSITION", "int", None, None, None, None, None),
                    ("IS_NULLABLE", "string", None, None, None, None, None),
                    ("SCOPE_CATALOG", "string", None, None, None, None, None),
                    ("SCOPE_SCHEMA", "string", None, None, None, None, None),
                    ("SCOPE_TABLE", "string", None, None, None, None, None),
                    ("SOURCE_DATA_TYPE", "smallint", None, None, None, None, None),
                    ("IS_AUTO_INCREMENT", "string", None, None, None, None, None),
                ]
                assert cols_desc == expected
            finally:
                for table in table_names:
                    cursor.execute("DROP TABLE IF EXISTS {}".format(table))

    def test_escape_single_quotes(self):
        with self.cursor({}) as cursor:
            table_name = "table_{uuid}".format(uuid=str(uuid4()).replace("-", "_"))
            # Test escape syntax directly
            cursor.execute(
                "CREATE TABLE IF NOT EXISTS {} AS (SELECT 'you\\'re' AS col_1)".format(
                    table_name
                )
            )
            cursor.execute(
                "SELECT * FROM {} WHERE col_1 LIKE 'you\\'re'".format(table_name)
            )
            rows = cursor.fetchall()
            assert rows[0]["col_1"] == "you're"

            # Test escape syntax in parameter
            cursor.execute(
                "SELECT * FROM {} WHERE {}.col_1 LIKE %(var)s".format(
                    table_name, table_name
                ),
                parameters={"var": "you're"},
            )
            rows = cursor.fetchall()
            assert rows[0]["col_1"] == "you're"

    def test_get_schemas(self):
        with self.cursor({}) as cursor:
            database_name = "db_{uuid}".format(uuid=str(uuid4()).replace("-", "_"))
            try:
                cursor.execute("CREATE DATABASE IF NOT EXISTS {}".format(database_name))
                cursor.schemas()
                schemas = cursor.fetchall()
                schemas_desc = cursor.description
                # Catalogue name not consistent across DBR versions, so we skip that
                assert database_name in [schema[0] for schema in schemas]
                assert schemas_desc == [
                    ("TABLE_SCHEM", "string", None, None, None, None, None),
                    ("TABLE_CATALOG", "string", None, None, None, None, None),
                ]

            finally:
                cursor.execute("DROP DATABASE IF EXISTS {}".format(database_name))

    def test_get_catalogs(self):
        with self.cursor({}) as cursor:
            cursor.catalogs()
            cursor.fetchall()
            catalogs_desc = cursor.description
            assert catalogs_desc == [
                ("TABLE_CAT", "string", None, None, None, None, None)
            ]

    @skipUnless(pysql_supports_arrow(), "arrow test need arrow support")
    def test_get_arrow(self):
        # These tests are quite light weight as the arrow fetch methods are used internally
        # by everything else
        with self.cursor({}) as cursor:
            cursor.execute("SELECT * FROM range(10)")
            table_1 = cursor.fetchmany_arrow(1).to_pydict()
            assert table_1 == OrderedDict([("id", [0])])

            table_2 = cursor.fetchall_arrow().to_pydict()
            assert table_2 == OrderedDict([("id", [1, 2, 3, 4, 5, 6, 7, 8, 9])])

    def test_unicode(self):
        unicode_str = "数据砖"
        with self.cursor({}) as cursor:
            cursor.execute("SELECT '{}'".format(unicode_str))
            results = cursor.fetchall()
            assert len(results) == 1 and len(results[0]) == 1
            assert results[0][0] == unicode_str

    def test_cancel_during_execute(self):
        with self.cursor({}) as cursor:

            def execute_really_long_query():
                cursor.execute(
                    "SELECT SUM(A.id - B.id) "
                    + "FROM range(1000000000) A CROSS JOIN range(100000000) B "
                    + "GROUP BY (A.id - B.id)"
                )

            exec_thread = threading.Thread(target=execute_really_long_query)

            exec_thread.start()
            # Make sure the query has started before cancelling
            time.sleep(15)
            cursor.cancel()
            exec_thread.join(5)
            assert not exec_thread.is_alive()

            # Fetching results should throw an exception
            with pytest.raises((Error, thrift.Thrift.TException)):
                cursor.fetchall()
            with pytest.raises((Error, thrift.Thrift.TException)):
                cursor.fetchone()
            with pytest.raises((Error, thrift.Thrift.TException)):
                cursor.fetchmany(10)

            # We should be able to execute a new command on the cursor
            cursor.execute("SELECT * FROM range(3)")
            assert len(cursor.fetchall()) == 3

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_can_execute_command_after_failure(self):
        with self.cursor({}) as cursor:
            with pytest.raises(DatabaseError):
                cursor.execute("this is a sytnax error")

            cursor.execute("SELECT 1;")

            res = cursor.fetchall()
            self.assertEqualRowValues(res, [[1]])

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_can_execute_command_after_success(self):
        with self.cursor({}) as cursor:
            cursor.execute("SELECT 1;")
            cursor.execute("SELECT 2;")

            res = cursor.fetchall()
            self.assertEqualRowValues(res, [[2]])

    def generate_multi_row_query(self):
        query = "SELECT * FROM range(3);"
        return query

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_fetchone(self):
        with self.cursor({}) as cursor:
            query = self.generate_multi_row_query()
            cursor.execute(query)

            assert cursor.fetchone()[0] == 0
            assert cursor.fetchone()[0] == 1
            assert cursor.fetchone()[0] == 2

            assert cursor.fetchone() == None

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_fetchall(self):
        with self.cursor({}) as cursor:
            query = self.generate_multi_row_query()
            cursor.execute(query)

            self.assertEqualRowValues(cursor.fetchall(), [[0], [1], [2]])

            assert cursor.fetchone() == None

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_fetchmany_when_stride_fits(self):
        with self.cursor({}) as cursor:
            query = "SELECT * FROM range(4)"
            cursor.execute(query)

            self.assertEqualRowValues(cursor.fetchmany(2), [[0], [1]])
            self.assertEqualRowValues(cursor.fetchmany(2), [[2], [3]])

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_fetchmany_in_excess(self):
        with self.cursor({}) as cursor:
            query = "SELECT * FROM range(4)"
            cursor.execute(query)

            self.assertEqualRowValues(cursor.fetchmany(3), [[0], [1], [2]])
            self.assertEqualRowValues(cursor.fetchmany(3), [[3]])

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_iterator_api(self):
        with self.cursor({}) as cursor:
            query = "SELECT * FROM range(4)"
            cursor.execute(query)

            expected_results = [[0], [1], [2], [3]]
            for i, row in enumerate(cursor):
                for j in range(len(row)):
                    assert row[j] == expected_results[i][j]

    def test_temp_view_fetch(self):
        with self.cursor({}) as cursor:
            query = "create temporary view f as select * from range(10)"
            cursor.execute(query)
            # TODO assert on a result
            # once what is being returned has stabilised

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    @skipIf(
        True,
        "Unclear the purpose of this test since urllib3 does not complain when timeout == 0",
    )
    def test_socket_timeout(self):
        #  We expect to see a BlockingIO error when the socket is opened
        #  in non-blocking mode, since no poll is done before the read
        with pytest.raises(OperationalError) as cm:
            with self.cursor({"_socket_timeout": 0}):
                pass

        self.assertIsInstance(cm.exception.args[1], io.BlockingIOError)

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    @skipIf(pysql_has_version(">", "2.8"), "This test has been broken for a while")
    def test_socket_timeout_user_defined(self):
        #  We expect to see a TimeoutError when the socket timeout is only
        #  1 sec for a query that takes longer than that to process
        with pytest.raises(ReadTimeoutError) as cm:
            with self.cursor({"_socket_timeout": 1}) as cursor:
                query = "select * from range(1000000000)"
                cursor.execute(query)

    def test_ssp_passthrough(self):
        for enable_ansi in (True, False):
            with self.cursor(
                {"session_configuration": {"ansi_mode": enable_ansi}}
            ) as cursor:
                cursor.execute("SET ansi_mode")
                assert list(cursor.fetchone()) == ["ansi_mode", str(enable_ansi)]

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_timestamps_arrow(self):
        with self.cursor({"session_configuration": {"ansi_mode": False}}) as cursor:
            for timestamp, expected in self.timestamp_and_expected_results:
                cursor.execute(
                    "SELECT TIMESTAMP('{timestamp}')".format(timestamp=timestamp)
                )
                arrow_table = cursor.fetchmany_arrow(1)
                if self.should_add_timezone():
                    ts_type = pyarrow.timestamp("us", tz="Etc/UTC")
                else:
                    ts_type = pyarrow.timestamp("us")
                assert arrow_table.field(0).type == ts_type
                result_value = arrow_table.column(0).combine_chunks()[0].value
                # To work consistently across different local timezones, we specify the timezone
                # of the expected result to
                # be UTC (what it should be by default on the server)
                aware_timestamp = expected and expected.replace(
                    tzinfo=datetime.timezone.utc
                )
                assert result_value == (
                    aware_timestamp and aware_timestamp.timestamp() * 1000000
                ), "timestamp {} did not match {}".format(timestamp, expected)

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_multi_timestamps_arrow(self):
        with self.cursor({"session_configuration": {"ansi_mode": False}}) as cursor:
            query, expected = self.multi_query()
            expected = [
                [self.maybe_add_timezone_to_timestamp(ts) for ts in row]
                for row in expected
            ]
            cursor.execute(query)
            table = cursor.fetchall_arrow()
            # Transpose columnar result to list of rows
            list_of_cols = [c.to_pylist() for c in table]
            result = [
                [col[row_index] for col in list_of_cols]
                for row_index in range(table.num_rows)
            ]
            assert result == expected

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_timezone_with_timestamp(self):
        if self.should_add_timezone():
            with self.cursor() as cursor:
                cursor.execute("SET TIME ZONE 'Europe/Amsterdam'")
                cursor.execute("select CAST('2022-03-02 12:54:56' as TIMESTAMP)")
                amsterdam = pytz.timezone("Europe/Amsterdam")
                expected = amsterdam.localize(datetime.datetime(2022, 3, 2, 12, 54, 56))
                result = cursor.fetchone()[0]
                assert result == expected

                cursor.execute("select CAST('2022-03-02 12:54:56' as TIMESTAMP)")
                arrow_result_table = cursor.fetchmany_arrow(1)
                arrow_result_value = (
                    arrow_result_table.column(0).combine_chunks()[0].value
                )
                ts_type = pyarrow.timestamp("us", tz="Europe/Amsterdam")

                assert arrow_result_table.field(0).type == ts_type
                assert arrow_result_value == expected.timestamp() * 1000000

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_can_flip_compression(self):
        with self.cursor() as cursor:
            cursor.execute("SELECT array(1,2,3,4)")
            cursor.fetchall()
            lz4_compressed = cursor.active_result_set.lz4_compressed
            # The endpoint should support compression
            assert lz4_compressed
            cursor.connection.lz4_compression = False
            cursor.execute("SELECT array(1,2,3,4)")
            cursor.fetchall()
            lz4_compressed = cursor.active_result_set.lz4_compressed
            assert not lz4_compressed

    def _should_have_native_complex_types(self):
        return pysql_has_version(">=", 2) and is_thrift_v5_plus(self.arguments)

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_arrays_are_not_returned_as_strings_arrow(self):
        if self._should_have_native_complex_types():
            with self.cursor() as cursor:
                cursor.execute("SELECT array(1,2,3,4)")
                arrow_df = cursor.fetchall_arrow()

                list_type = arrow_df.field(0).type
                assert pyarrow.types.is_list(list_type)
                assert pyarrow.types.is_integer(list_type.value_type)

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_structs_are_not_returned_as_strings_arrow(self):
        if self._should_have_native_complex_types():
            with self.cursor() as cursor:
                cursor.execute("SELECT named_struct('foo', 42, 'bar', 'baz')")
                arrow_df = cursor.fetchall_arrow()

                struct_type = arrow_df.field(0).type
                assert pyarrow.types.is_struct(struct_type)

    @skipUnless(pysql_supports_arrow(), "arrow test needs arrow support")
    def test_decimal_not_returned_as_strings_arrow(self):
        if self._should_have_native_complex_types():
            with self.cursor() as cursor:
                cursor.execute("SELECT 5E3BD")
                arrow_df = cursor.fetchall_arrow()

                decimal_type = arrow_df.field(0).type
                assert pyarrow.types.is_decimal(decimal_type)

    def test_close_connection_closes_cursors(self):

        from databricks.sql.thrift_api.TCLIService import ttypes

        with self.connection() as conn:
            cursor = conn.cursor()
            cursor.execute(
                "SELECT id, id `id2`, id `id3` FROM RANGE(1000000) order by RANDOM()"
            )
            ars = cursor.active_result_set

            # We must manually run this check because thrift_backend always forces `has_been_closed_server_side` to True

            # Cursor op state should be open before connection is closed
            status_request = ttypes.TGetOperationStatusReq(
                operationHandle=ars.command_id, getProgressUpdate=False
            )
            op_status_at_server = ars.thrift_backend._client.GetOperationStatus(
                status_request
            )
            assert (
                op_status_at_server.operationState
                != ttypes.TOperationState.CLOSED_STATE
            )

            conn.close()

            # When connection closes, any cursor operations should no longer exist at the server
            with pytest.raises(SessionAlreadyClosedError) as cm:
                op_status_at_server = ars.thrift_backend._client.GetOperationStatus(
                    status_request
                )

    def test_closing_a_closed_connection_doesnt_fail(self, caplog):
        caplog.set_level(logging.DEBUG)
        # Second .close() call is when this context manager exits
        with self.connection() as conn:
            # First .close() call is explicit here
            conn.close()

        assert "Session appears to have been closed already" in caplog.text


# use a RetrySuite to encapsulate these tests which we'll typically want to run together; however keep
# the 429/503 subsuites separate since they execute under different circumstances.
class TestPySQLRetrySuite:
    class HTTP429Suite(Client429ResponseMixin, PySQLPytestTestCase):
        pass  # Mixin covers all

    class HTTP503Suite(Client503ResponseMixin, PySQLPytestTestCase):
        # 503Response suite gets custom error here vs PyODBC
        def test_retry_disabled(self):
            self._test_retry_disabled_with_message(
                "TEMPORARILY_UNAVAILABLE", OperationalError
            )


class TestPySQLUnityCatalogSuite(PySQLPytestTestCase):
    """Simple namespace tests that should be run against a unity-catalog-enabled cluster"""

    @skipIf(pysql_has_version("<", "2"), "requires pysql v2")
    def test_initial_namespace(self):
        table_name = "table_{uuid}".format(uuid=str(uuid4()).replace("-", "_"))
        with self.cursor() as cursor:
            cursor.execute("USE CATALOG {}".format(self.arguments["catalog"]))
            cursor.execute("CREATE TABLE table_{} (col1 int)".format(table_name))
        with self.connection(
            {"catalog": self.arguments["catalog"], "schema": table_name}
        ) as connection:
            cursor = connection.cursor()
            cursor.execute("select current_catalog()")
            assert cursor.fetchone()[0] == self.arguments["catalog"]
            cursor.execute("select current_database()")
            assert cursor.fetchone()[0] == table_name