maxframe 1.0.0rc3__cp311-cp311-macosx_10_9_universal2.whl → 1.1.0__cp311-cp311-macosx_10_9_universal2.whl

maxframe/dataframe/window/expanding.py

@@ -28,6 +28,7 @@ from .aggregation import BaseDataFrameExpandingAgg
 from .core import Window
 
 _window_has_method = pd_release_version >= (1, 3, 0)
+_window_has_center = pd_release_version < (2, 0, 0)
 
 
 class DataFrameExpandingAgg(BaseDataFrameExpandingAgg):
@@ -49,10 +50,11 @@ class Expanding(Window):
     def params(self):
         p = OrderedDict()
 
+        args = ["min_periods", "center", "axis", "method"]
         if not _window_has_method:  # pragma: no cover
-            args = ["min_periods", "center", "axis"]
-        else:
-            args = ["min_periods", "center", "axis", "method"]
+            args = [a for a in args if a != "method"]
+        if not _window_has_center:
+            args = [a for a in args if a != "center"]
 
         for k in args:
             p[k] = getattr(self, k)

maxframe/dataframe/window/tests/test_expanding.py

@@ -29,8 +29,8 @@ def test_expanding():
     with pytest.raises(NotImplementedError):
         _ = df2.expanding(3, axis=1)
 
-    r = df2.expanding(3, center=False)
-    expected = df.expanding(3, center=False)
+    r = df2.expanding(3)
+    expected = df.expanding(3)
     assert repr(r) == repr(expected)
 
     assert "b" in dir(r)

maxframe/io/objects/tests/test_object_io.py

@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 import numpy as np
 import pytest
 from odps import ODPS
@@ -48,15 +49,33 @@ def create_volume(request, oss_config):
         oss_bucket_name,
         oss_endpoint,
     ) = oss_config.oss_config
-    test_location = "oss://%s:%s@%s/%s/%s" % (
-        oss_access_id,
-        oss_secret_access_key,
-        oss_endpoint,
-        oss_bucket_name,
-        oss_test_dir_name,
-    )
+
+    if "test" in oss_endpoint:
+        # offline config
+        test_location = "oss://%s:%s@%s/%s/%s" % (
+            oss_access_id,
+            oss_secret_access_key,
+            oss_endpoint,
+            oss_bucket_name,
+            oss_test_dir_name,
+        )
+        rolearn = None
+    else:
+        # online config
+        endpoint_parts = oss_endpoint.split(".", 1)
+        if "-internal" not in endpoint_parts[0]:
+            endpoint_parts[0] += "-internal"
+        test_location = "oss://%s/%s/%s" % (
+            ".".join(endpoint_parts),
+            oss_bucket_name,
+            oss_test_dir_name,
+        )
+        rolearn = oss_config.oss_rolearn
+
     oss_config.oss_bucket.put_object(oss_test_dir_name + "/", b"")
-    odps_entry.create_external_volume(test_vol_name, location=test_location)
+    odps_entry.create_external_volume(
+        test_vol_name, location=test_location, rolearn=rolearn
+    )
 
     try:
         yield test_vol_name
@@ -75,8 +94,12 @@ def test_simple_object_io(create_volume):
 
     odps_entry = ODPS.from_environments()
 
-    reader = ODPSVolumeReader(odps_entry, create_volume, obj.key)
-    writer = ODPSVolumeWriter(odps_entry, create_volume, obj.key)
+    reader = ODPSVolumeReader(
+        odps_entry, create_volume, obj.key, replace_internal_host=True
+    )
+    writer = ODPSVolumeWriter(
+        odps_entry, create_volume, obj.key, replace_internal_host=True
+    )
 
     handler = get_object_io_handler(obj)()
     handler.write_object(writer, obj, data)
@@ -89,8 +112,12 @@ def test_tensor_object_io(create_volume):
 
     odps_entry = ODPS.from_environments()
 
-    reader = ODPSVolumeReader(odps_entry, create_volume, obj.key)
-    writer = ODPSVolumeWriter(odps_entry, create_volume, obj.key)
+    reader = ODPSVolumeReader(
+        odps_entry, create_volume, obj.key, replace_internal_host=True
+    )
+    writer = ODPSVolumeWriter(
+        odps_entry, create_volume, obj.key, replace_internal_host=True
+    )
 
     handler = get_object_io_handler(obj)()
     handler.write_object(writer, obj, data)

maxframe/io/odpsio/arrow.py

@@ -69,13 +69,24 @@ def arrow_to_pandas(
 
 
 def pandas_to_arrow(
-    df: Any, nthreads=1, ignore_index=False
+    df: Any, nthreads=1, ignore_index=False, ms_cols=None
 ) -> Tuple[ArrowTableType, DataFrameTableMeta]:
     table_meta = build_dataframe_table_meta(df, ignore_index)
     df = df.copy() if callable(getattr(df, "copy", None)) else df
+    table_datetime_cols = None
     if table_meta.type in (OutputType.dataframe, OutputType.series):
         if table_meta.type == OutputType.series:
             df = df.to_frame("_data" if df.name is None else df.name)
+            if ms_cols:
+                table_datetime_cols = {"_data"}
+        elif ms_cols:
+            ms_col_set = set(ms_cols)
+            table_datetime_cols = set()
+            for pd_col, table_col in zip(
+                table_meta.pd_column_dtypes.keys(), table_meta.table_column_names
+            ):
+                if pd_col in ms_col_set:
+                    table_datetime_cols.add(table_col)
         df.columns = pd.Index(table_meta.table_column_names)
         if not ignore_index:
             df = df.rename_axis(table_meta.table_index_column_names).reset_index()
@@ -83,6 +94,12 @@ def pandas_to_arrow(
         df = pd.DataFrame([], columns=[])
     elif table_meta.type == OutputType.index:
         names = [f"_idx_{idx}" for idx in range(len(df.names))]
+        table_datetime_cols = set()
+        if ms_cols:
+            if isinstance(df, pd.MultiIndex):
+                table_datetime_cols = {f"_idx_{idx}" for idx in ms_cols}
+            else:
+                table_datetime_cols = {"_idx_0"}
         df = df.to_frame(name=names[0] if len(names) == 1 else names)
     elif table_meta.type == OutputType.scalar:
         names = ["_idx_0"]
@@ -92,4 +109,15 @@ def pandas_to_arrow(
             df = pd.DataFrame([[df]], columns=names)
     else:  # this could never happen  # pragma: no cover
         raise ValueError(f"Does not support meta type {table_meta.type!r}")
-    return pa.Table.from_pandas(df, nthreads=nthreads, preserve_index=False), table_meta
+    pa_table = pa.Table.from_pandas(df, nthreads=nthreads, preserve_index=False)
+    if table_datetime_cols:
+        col_names = pa_table.schema.names
+        col_datas = []
+        for idx, col_name in enumerate(pa_table.schema.names):
+            if col_name not in table_datetime_cols:
+                col_datas.append(pa_table.column(idx))
+                continue
+            col_data = pa_table.column(idx).cast(pa.timestamp("ms"))
+            col_datas.append(col_data)
+        pa_table = pa.Table.from_arrays(col_datas, names=col_names)
+    return pa_table, table_meta

maxframe/io/odpsio/schema.py

@@ -16,6 +16,7 @@ import string
 from collections import defaultdict
 from typing import Any, Dict, Tuple
 
+import numpy as np
 import pandas as pd
 import pyarrow as pa
 from odps import types as odps_types
@@ -39,6 +40,7 @@ _arrow_to_odps_types = {
     pa.float64(): odps_types.double,
     pa.date32(): odps_types.date,
     pa.timestamp("ms"): odps_types.datetime,
+    pa.timestamp("us"): odps_types.timestamp,
     pa.timestamp("ns"): odps_types.timestamp,
 }
 
@@ -54,7 +56,9 @@ _odps_type_to_arrow = {
     odps_types.double: pa.float64(),
     odps_types.date: pa.date32(),
     odps_types.datetime: pa.timestamp("ms"),
+    odps_types.json: pa.string(),
     odps_types.timestamp: pa.timestamp("ns"),
+    odps_types.timestamp_ntz: pa.timestamp("ns"),
 }
 
 
@@ -166,7 +170,7 @@ def odps_schema_to_pandas_dtypes(
     return arrow_schema.empty_table().to_pandas().dtypes
 
 
-def _is_scalar_object(df_obj: Any) -> bool:
+def is_scalar_object(df_obj: Any) -> bool:
     return (
         isinstance(df_obj, TENSOR_TYPE) and df_obj.shape == ()
     ) or pd_types.is_scalar(df_obj)
@@ -187,7 +191,7 @@ def pandas_to_odps_schema(
     from ... import dataframe as md
     from .arrow import pandas_to_arrow
 
-    if _is_scalar_object(df_obj):
+    if is_scalar_object(df_obj):
         empty_index = None
     elif hasattr(df_obj, "index_value"):
         empty_index = df_obj.index_value.to_pandas()[:0]
@@ -203,20 +207,35 @@ def pandas_to_odps_schema(
     else:
         empty_columns = None
 
+    ms_cols = None
     if isinstance(df_obj, (md.DataFrame, pd.DataFrame)):
         empty_df_obj = pd.DataFrame(
             [], columns=empty_columns, index=empty_index
         ).astype(df_obj.dtypes)
+        ms_cols = [
+            col for col, dt in df_obj.dtypes.items() if dt == np.dtype("datetime64[ms]")
+        ]
     elif isinstance(df_obj, (md.Series, pd.Series)):
         empty_df_obj = pd.Series([], name=df_obj.name, index=empty_index).astype(
             df_obj.dtype
         )
+        ms_cols = df_obj.dtype == np.dtype("datetime64[ms]")
     elif isinstance(df_obj, (md.Index, pd.Index)):
         empty_df_obj = empty_index
+        if isinstance(empty_index, pd.MultiIndex):
+            ms_cols = [
+                idx
+                for idx, dt in enumerate(empty_index.dtypes.values)
+                if dt == np.dtype("datetime64[ms]")
+            ]
+        else:
+            ms_cols = df_obj.dtype == np.dtype("datetime64[ms]")
     else:
         empty_df_obj = df_obj
 
-    arrow_data, table_meta = pandas_to_arrow(empty_df_obj, ignore_index=ignore_index)
+    arrow_data, table_meta = pandas_to_arrow(
+        empty_df_obj, ignore_index=ignore_index, ms_cols=ms_cols
+    )
     return (
         arrow_schema_to_odps_schema(
             arrow_data.schema, unknown_as_string=unknown_as_string
@@ -289,7 +308,7 @@ def build_dataframe_table_meta(
         obj_type = OutputType.series
     elif isinstance(df_obj, (md.Index, pd.Index)):
         obj_type = OutputType.index
-    elif _is_scalar_object(df_obj):
+    elif is_scalar_object(df_obj):
         obj_type = OutputType.scalar
     else:  # pragma: no cover
         raise TypeError(f"Cannot accept type {type(df_obj)}")
@@ -344,10 +363,11 @@ def build_dataframe_table_meta(
     else:
         pd_index_val = index_obj
 
-    if hasattr(pd_index_val, "dtypes"):
-        index_dtypes = pd.Series(pd_index_val.dtypes.values, index=pd_index_val.names)
-    else:
-        index_dtypes = pd.Series([pd_index_val.dtype], index=pd_index_val.names)
+    level_dtypes = [
+        pd_index_val.get_level_values(level).dtype
+        for level in range(pd_index_val.nlevels)
+    ]
+    index_dtypes = pd.Series(level_dtypes, index=pd_index_val.names)
 
     if ignore_index and obj_type != OutputType.index:
         table_index_column_names = []

maxframe/io/odpsio/tableio.py

@@ -20,15 +20,14 @@ from typing import Dict, List, Optional, Union
 
 import pyarrow as pa
 from odps import ODPS
-from odps import __version__ as pyodps_version
 from odps.apis.storage_api import (
     StorageApiArrowClient,
     TableBatchScanResponse,
     TableBatchWriteResponse,
 )
-from odps.config import option_context as pyodps_option_context
 from odps.tunnel import TableTunnel
 from odps.types import OdpsSchema, PartitionSpec, timestamp_ntz
+from odps.utils import call_with_retry
 
 try:
     import pyarrow.compute as pac
@@ -37,26 +36,18 @@ except ImportError:
 
 from ...config import options
 from ...env import ODPS_STORAGE_API_ENDPOINT
-from ...lib.version import Version
+from ...utils import sync_pyodps_options
 from .schema import odps_schema_to_arrow_schema
 
 PartitionsType = Union[List[str], str, None]
 
 _DEFAULT_ROW_BATCH_SIZE = 4096
-_need_convert_timezone = Version(pyodps_version) < Version("0.11.7")
-
-
-@contextmanager
-def _sync_pyodps_timezone():
-    with pyodps_option_context() as cfg:
-        cfg.local_timezone = options.local_timezone
-        yield
 
 
 class ODPSTableIO(ABC):
     def __new__(cls, odps: ODPS):
         if cls is ODPSTableIO:
-            if options.use_common_table:
+            if options.use_common_table or ODPS_STORAGE_API_ENDPOINT in os.environ:
                 return HaloTableIO(odps)
             else:
                 return TunnelTableIO(odps)
@@ -138,7 +129,12 @@ class TunnelMultiPartitionReader:
         self._cur_partition_id = -1
         self._reader_start_pos = 0
 
-        if partitions is None or isinstance(partitions, str):
+        if partitions is None:
+            if not self._table.table_schema.partitions:
+                self._partitions = [None]
+            else:
+                self._partitions = [str(pt) for pt in self._table.partitions]
+        elif isinstance(partitions, str):
             self._partitions = [partitions]
         else:
             self._partitions = partitions
@@ -166,12 +162,14 @@ class TunnelMultiPartitionReader:
             self._cur_partition_id += 1
 
             part_str = self._partitions[self._cur_partition_id]
-            with _sync_pyodps_timezone():
+            req_columns = self._schema.names
+            with sync_pyodps_options():
                 self._cur_reader = self._table.open_reader(
                     part_str,
-                    columns=self._columns,
+                    columns=req_columns,
                     arrow=True,
                     download_id=self._partition_to_download_ids.get(part_str),
+                    append_partitions=True,
                 )
             if self._cur_reader.count + self._reader_start_pos > self._start:
                 start = self._start - self._reader_start_pos
@@ -180,43 +178,15 @@ class TunnelMultiPartitionReader:
                 else:
                     count = min(self._count, self._cur_reader.count - start)
 
-                with _sync_pyodps_timezone():
+                with sync_pyodps_options():
                     self._reader_iter = self._cur_reader.read(start, count)
                 break
             self._reader_start_pos += self._cur_reader.count
         else:
             self._cur_reader = None
 
-    def _fill_batch_partition(self, batch: pa.RecordBatch) -> pa.RecordBatch:
-        pt_spec = PartitionSpec(self._partitions[self._cur_partition_id])
-
-        names = list(batch.schema.names)
-        arrays = []
-        for idx in range(batch.num_columns):
-            col = batch.column(idx)
-            if _need_convert_timezone and isinstance(col.type, pa.TimestampType):
-                if col.type.tz is not None:
-                    target_type = pa.timestamp(
-                        self._schema.types[idx].unit, col.type.tz
-                    )
-                    arrays.append(col.cast(target_type))
-                else:
-                    target_type = pa.timestamp(
-                        self._schema.types[idx].unit, options.local_timezone
-                    )
-                    pd_col = col.to_pandas().dt.tz_localize(options.local_timezone)
-                    arrays.append(pa.Array.from_pandas(pd_col).cast(target_type))
-            else:
-                arrays.append(batch.column(idx))
-
-        for part_col in self._partition_cols or []:
-            names.append(part_col)
-            col_type = self._schema.field_by_name(part_col).type
-            arrays.append(pa.array([pt_spec[part_col]] * batch.num_rows).cast(col_type))
-        return pa.RecordBatch.from_arrays(arrays, names)
-
     def read(self):
-        with _sync_pyodps_timezone():
+        with sync_pyodps_options():
             if self._cur_reader is None:
                 self._open_next_reader()
                 if self._cur_reader is None:
@@ -227,7 +197,7 @@ class TunnelMultiPartitionReader:
                     if batch is not None:
                         if self._row_left is not None:
                             self._row_left -= batch.num_rows
-                        return self._fill_batch_partition(batch)
+                        return batch
                 except StopIteration:
                     self._open_next_reader()
             return None
@@ -244,34 +214,6 @@ class TunnelMultiPartitionReader:
         return pa.Table.from_batches(batches)
 
 
-class TunnelWrappedWriter:
-    def __init__(self, nested_writer):
-        self._writer = nested_writer
-
-    def write(self, data: Union[pa.RecordBatch, pa.Table]):
-        if not any(isinstance(tp, pa.TimestampType) for tp in data.schema.types):
-            self._writer.write(data)
-            return
-        pa_type = type(data)
-        arrays = []
-        for idx in range(data.num_columns):
-            name = data.schema.names[idx]
-            col = data.column(idx)
-            if not isinstance(col.type, pa.TimestampType):
-                arrays.append(col)
-                continue
-            if self._writer.schema[name].type == timestamp_ntz:
-                col = HaloTableArrowWriter._localize_timezone(col, "UTC")
-            else:
-                col = HaloTableArrowWriter._localize_timezone(col)
-            arrays.append(col)
-        data = pa_type.from_arrays(arrays, names=data.schema.names)
-        self._writer.write(data)
-
-    def __getattr__(self, item):
-        return getattr(self._writer, item)
-
-
 class TunnelTableIO(ODPSTableIO):
     @contextmanager
     def open_reader(
@@ -285,7 +227,9 @@ class TunnelTableIO(ODPSTableIO):
         reverse_range: bool = False,
         row_batch_size: int = _DEFAULT_ROW_BATCH_SIZE,
     ):
-        table = self._odps.get_table(full_table_name)
+        with sync_pyodps_options():
+            table = self._odps.get_table(full_table_name)
+
         if partition_columns is True:
             partition_columns = [c.name for c in table.table_schema.partitions]
 
@@ -296,21 +240,22 @@ class TunnelTableIO(ODPSTableIO):
             or (stop is not None and stop < 0)
             or (reverse_range and start is None)
         ):
-            table = self._odps.get_table(full_table_name)
-            tunnel = TableTunnel(self._odps)
-            parts = (
-                [partitions]
-                if partitions is None or isinstance(partitions, str)
-                else partitions
-            )
-            part_to_down_id = dict()
-            total_records = 0
-            for part in parts:
-                down_session = tunnel.create_download_session(
-                    table, async_mode=True, partition_spec=part
+            with sync_pyodps_options():
+                table = self._odps.get_table(full_table_name)
+                tunnel = TableTunnel(self._odps)
+                parts = (
+                    [partitions]
+                    if partitions is None or isinstance(partitions, str)
+                    else partitions
                 )
-                part_to_down_id[part] = down_session.id
-                total_records += down_session.count
+                part_to_down_id = dict()
+                total_records = 0
+                for part in parts:
+                    down_session = tunnel.create_download_session(
+                        table, async_mode=True, partition_spec=part
+                    )
+                    part_to_down_id[part] = down_session.id
+                    total_records += down_session.count
 
         count = None
         if start is not None or stop is not None:
@@ -347,20 +292,14 @@ class TunnelTableIO(ODPSTableIO):
         overwrite: bool = True,
     ):
         table = self._odps.get_table(full_table_name)
-        with _sync_pyodps_timezone():
+        with sync_pyodps_options():
             with table.open_writer(
                 partition=partition,
                 arrow=True,
                 create_partition=partition is not None,
                 overwrite=overwrite,
             ) as writer:
-                # fixme should yield writer directly once pyodps fixes
-                #  related arrow timestamp bug when provided schema and
-                #  table schema is identical.
-                if _need_convert_timezone:
-                    yield TunnelWrappedWriter(writer)
-                else:
-                    yield writer
+                yield writer
 
 
 class HaloTableArrowReader:
@@ -416,7 +355,7 @@ class HaloTableArrowReader:
             split_index=self._cur_split_id + 1,
             **read_rows_kw,
         )
-        self._cur_reader = self._client.read_rows_arrow(req)
+        self._cur_reader = call_with_retry(self._client.read_rows_arrow, req)
         self._cur_split_id += 1
 
     def _convert_timezone(self, batch: pa.RecordBatch) -> pa.RecordBatch:
@@ -488,8 +427,9 @@ class HaloTableArrowWriter:
     def open(self):
         from odps.apis.storage_api import WriteRowsRequest
 
-        self._writer = self._client.write_rows_arrow(
-            WriteRowsRequest(self._write_info.session_id)
+        self._writer = call_with_retry(
+            self._client.write_rows_arrow,
+            WriteRowsRequest(self._write_info.session_id),
         )
 
     @classmethod
@@ -560,28 +500,6 @@ class HaloTableIO(ODPSTableIO):
             for pt in partitions
         ]
 
-    def get_table_record_count(
-        self, full_table_name: str, partitions: PartitionsType = None
-    ):
-        from odps.apis.storage_api import SplitOptions, TableBatchScanRequest
-
-        table = self._odps.get_table(full_table_name)
-        client = StorageApiArrowClient(
-            self._odps, table, rest_endpoint=self._storage_api_endpoint
-        )
-
-        split_option = SplitOptions.SplitMode.SIZE
-
-        scan_kw = {
-            "required_partitions": self._convert_partitions(partitions),
-            "split_options": SplitOptions.get_default_options(split_option),
-        }
-
-        # todo add more options for partition column handling
-        req = TableBatchScanRequest(**scan_kw)
-        resp = client.create_read_session(req)
-        return resp.record_count
-
     @contextmanager
     def open_reader(
         self,
@@ -596,8 +514,8 @@ class HaloTableIO(ODPSTableIO):
     ):
         from odps.apis.storage_api import (
             SessionRequest,
+            SessionStatus,
             SplitOptions,
-            Status,
             TableBatchScanRequest,
         )
 
@@ -625,16 +543,16 @@ class HaloTableIO(ODPSTableIO):
 
         # todo add more options for partition column handling
         req = TableBatchScanRequest(**scan_kw)
-        resp = client.create_read_session(req)
+        resp = call_with_retry(client.create_read_session, req)
 
         session_id = resp.session_id
-        status = resp.status
-        while status == Status.WAIT:
-            resp = client.get_read_session(SessionRequest(session_id))
-            status = resp.status
+        status = resp.session_status
+        while status == SessionStatus.INIT:
+            resp = call_with_retry(client.get_read_session, SessionRequest(session_id))
+            status = resp.session_status
             time.sleep(1.0)
 
-        assert status == Status.OK
+        assert status == SessionStatus.NORMAL
 
         count = None
         if start is not None or stop is not None:
@@ -685,7 +603,7 @@ class HaloTableIO(ODPSTableIO):
         part_strs = self._convert_partitions(partition)
         part_str = part_strs[0] if part_strs else None
         req = TableBatchWriteRequest(partition_spec=part_str, overwrite=overwrite)
-        resp = client.create_write_session(req)
+        resp = call_with_retry(client.create_write_session, req)
 
         session_id = resp.session_id
         writer = HaloTableArrowWriter(client, resp, table.table_schema)
@@ -694,9 +612,13 @@ class HaloTableIO(ODPSTableIO):
         yield writer
 
         commit_msg = writer.close()
-        resp = client.commit_write_session(
-            SessionRequest(session_id=session_id), [commit_msg]
+        resp = call_with_retry(
+            client.commit_write_session,
+            SessionRequest(session_id=session_id),
+            [commit_msg],
         )
         while resp.session_status == SessionStatus.COMMITTING:
-            resp = client.get_write_session(SessionRequest(session_id=session_id))
+            resp = call_with_retry(
+                client.get_write_session, SessionRequest(session_id=session_id)
+            )
         assert resp.session_status == SessionStatus.COMMITTED

maxframe/io/odpsio/tests/test_schema.py

@@ -21,6 +21,7 @@ from odps import types as odps_types
 from .... import dataframe as md
 from .... import tensor as mt
 from ....core import OutputType
+from ....utils import pd_release_version
 from ..schema import (
     arrow_schema_to_odps_schema,
     build_dataframe_table_meta,
@@ -270,10 +271,6 @@ def test_odps_arrow_schema_conversion():
 
     with pytest.raises(TypeError):
         arrow_schema_to_odps_schema(pa.schema([("col1", pa.float16())]))
-    with pytest.raises(TypeError):
-        odps_schema_to_arrow_schema(
-            odps_types.OdpsSchema([odps_types.Column("col1", "json")])
-        )
 
 
 def test_build_column_name():
@@ -296,3 +293,42 @@ def test_build_table_meta(wrap_obj):
     table_meta = build_dataframe_table_meta(test_df)
     expected_cols = ["a_2", "a_3", "a_0", "a_1_0", "a_1_1", "b", "c"]
     assert table_meta.table_column_names == expected_cols
+
+
+@pytest.mark.skipif(
+    pd_release_version[0] < 2, reason="only run under pandas 2.0 or greater"
+)
+def test_table_meta_with_datetime():
+    raw_df = pd.DataFrame(
+        [
+            [1, "abc", "2024-10-01 11:23:12"],
+            [3, "uvw", "2024-10-02 22:55:13"],
+        ],
+        columns=["col1", "col2", "col3"],
+    )
+    df = md.DataFrame(raw_df).astype({"col3": "datetime64[ms]"})
+    schema, _ = pandas_to_odps_schema(df, unknown_as_string=True)
+    assert schema.columns[3].type == odps_types.datetime
+
+    raw_series = pd.Series(
+        ["2024-10-01 11:23:12", "2024-10-02 22:55:13"], dtype="datetime64[ms]"
+    )
+    s = md.Series(raw_series)
+    schema, _ = pandas_to_odps_schema(s, unknown_as_string=True)
+    assert schema.columns[1].type == odps_types.datetime
+
+    raw_index = pd.Index(
+        ["2024-10-01 11:23:12", "2024-10-02 22:55:13"], dtype="datetime64[ms]"
+    )
+    idx = md.Index(raw_index)
+    schema, _ = pandas_to_odps_schema(idx, unknown_as_string=True)
+    assert schema.columns[0].type == odps_types.datetime
+
+    src_df = pd.DataFrame(
+        [[1, "2024-10-01 11:23:12"], [3, "2024-10-02 22:55:13"]],
+        columns=["A", "B"],
+    ).astype({"B": "datetime64[ms]"})
+    raw_multiindex = pd.MultiIndex.from_frame(src_df)
+    multiidx = md.Index(raw_multiindex)
+    schema, _ = pandas_to_odps_schema(multiidx, unknown_as_string=True)
+    assert schema.columns[1].type == odps_types.datetime