maxframe 1.0.0rc4__cp38-cp38-macosx_10_9_universal2.whl → 1.1.0__cp38-cp38-macosx_10_9_universal2.whl

maxframe/config/config.py

@@ -380,6 +380,9 @@ default_options.register_option(
 default_options.register_option(
     "session.enable_schema", None, validator=is_null | is_bool, remote=True
 )
+default_options.register_option(
+    "session.enable_high_availability", None, validator=is_null | is_bool, remote=True
+)
 default_options.register_option(
     "session.default_schema", None, validator=is_null | is_string, remote=True
 )

maxframe/conftest.py

@@ -126,7 +126,14 @@ def oss_config():
         oss_rolearn = config.get("oss", "rolearn")
 
         options.service_role_arn = oss_rolearn
-        options.object_cache_url = f"oss://{oss_endpoint}/{oss_bucket_name}"
+        if "test" in oss_endpoint:
+            oss_svc_endpoint = oss_endpoint
+        else:
+            endpoint_parts = oss_endpoint.split(".", 1)
+            if "-internal" not in endpoint_parts[0]:
+                endpoint_parts[0] += "-internal"
+            oss_svc_endpoint = ".".join(endpoint_parts)
+        options.object_cache_url = f"oss://{oss_svc_endpoint}/{oss_bucket_name}"
 
         config.oss_config = (
             oss_access_id,
@@ -141,7 +148,7 @@ def oss_config():
         config.oss_bucket = oss2.Bucket(auth, oss_endpoint, oss_bucket_name)
         config.oss_rolearn = oss_rolearn
         yield config
-    except (ConfigParser.NoSectionError, ConfigParser.NoOptionError, ImportError):
+    except (NoSectionError, NoOptionError, ImportError):
         return None
     finally:
         options.service_role_arn = old_role_arn

maxframe/core/operator/base.py

@@ -86,6 +86,8 @@ class SchedulingHint(Serializable):
     # `gpu` indicates that if the operator should be executed on the GPU.
     gpu = BoolField("gpu", default=None)
     priority = Int32Field("priority", default=None)
+    expect_engine = StringField("expect_engine", default=None)
+    expect_resources = DictField("expect_resources", FieldTypes.string, default=None)
 
     @classproperty
     @lru_cache(1)

maxframe/dataframe/arithmetic/tests/test_arithmetic.py

@@ -22,6 +22,7 @@ import pandas as pd
 import pytest
 
 from ....core import OperatorType
+from ....tests.utils import assert_mf_index_dtype
 from ....utils import dataslots
 from ...core import IndexValue
 from ...datasource.dataframe import from_pandas
@@ -164,7 +165,7 @@ def test_without_shuffle(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -176,7 +177,7 @@ def test_without_shuffle(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -370,7 +371,7 @@ def test_with_one_shuffle(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -403,7 +404,7 @@ def test_with_all_shuffle(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -433,7 +434,7 @@ def test_with_all_shuffle(func_name, func_opts):
     pd.testing.assert_index_equal(
         df6.columns_value.to_pandas(), func_opts.func(data4, data5).columns
     )
-    assert isinstance(df6.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df6.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df6.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -468,7 +469,7 @@ def test_without_shuffle_and_with_one_chunk(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -501,7 +502,7 @@ def test_both_one_chunk(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -534,7 +535,7 @@ def test_with_shuffle_and_one_chunk(func_name, func_opts):
     pd.testing.assert_index_equal(
         df3.columns_value.to_pandas(), func_opts.func(data1, data2).columns
     )
-    assert isinstance(df3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df3.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df3.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -558,7 +559,7 @@ def test_on_same_dataframe(func_name, func_opts):
     pd.testing.assert_index_equal(
         df2.columns_value.to_pandas(), func_opts.func(data, data).columns
     )
-    assert isinstance(df2.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df2.index_value.value, np.int64)
     pd.testing.assert_index_equal(
         df2.index_value.to_pandas(), pd.Index([], dtype=np.int64)
     )
@@ -590,19 +591,19 @@ def test_dataframe_and_scalar(func_name, func_opts):
     pd.testing.assert_series_equal(result.dtypes, expected.dtypes)
 
     pd.testing.assert_index_equal(result.columns_value.to_pandas(), data.columns)
-    assert isinstance(result.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(result.index_value.value, np.int64)
 
     pd.testing.assert_index_equal(result2.columns_value.to_pandas(), data.columns)
-    assert isinstance(result2.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(result2.index_value.value, np.int64)
 
     pd.testing.assert_index_equal(result3.columns_value.to_pandas(), data.columns)
-    assert isinstance(result3.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(result3.index_value.value, np.int64)
 
     pd.testing.assert_index_equal(result4.columns_value.to_pandas(), data.columns)
-    assert isinstance(result4.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(result4.index_value.value, np.int64)
 
     pd.testing.assert_index_equal(result5.columns_value.to_pandas(), data.columns)
-    assert isinstance(result5.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(result5.index_value.value, np.int64)
 
     if "builtin_function_or_method" not in str(type(func_opts.func)):
         # skip NotImplemented test for comparison function
@@ -679,7 +680,7 @@ def test_abs():
     pd.testing.assert_index_equal(
         df2.columns_value.to_pandas(), df1.columns_value.to_pandas()
     )
-    assert isinstance(df2.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df2.index_value.value, np.int64)
     assert df2.shape == (10, 10)
 
 
@@ -697,7 +698,7 @@ def test_not():
     pd.testing.assert_index_equal(
         df2.columns_value.to_pandas(), df1.columns_value.to_pandas()
     )
-    assert isinstance(df2.index_value.value, IndexValue.Int64Index)
+    assert_mf_index_dtype(df2.index_value.value, np.int64)
     assert df2.shape == (10, 10)
 
 

maxframe/dataframe/core.py

@@ -142,6 +142,14 @@ class IndexValue(Serializable):
         _data = NDArrayField("data")
         _dtype = DataTypeField("dtype")
 
+        @property
+        def dtype(self):
+            return getattr(self, "_dtype", None)
+
+        @property
+        def inferred_type(self):
+            return "floating" if self.dtype.kind == "f" else "integer"
+
     class RangeIndex(IndexBase):
         _name = AnyField("name")
         _slice = SliceField("slice")
@@ -243,6 +251,10 @@ class IndexValue(Serializable):
         _data = NDArrayField("data")
         _dtype = DataTypeField("dtype")
 
+        @property
+        def dtype(self):
+            return getattr(self, "_dtype", None)
+
         @property
         def inferred_type(self):
             return "integer"
@@ -254,6 +266,10 @@ class IndexValue(Serializable):
         _data = NDArrayField("data")
         _dtype = DataTypeField("dtype")
 
+        @property
+        def dtype(self):
+            return getattr(self, "_dtype", None)
+
         @property
         def inferred_type(self):
             return "integer"
@@ -265,6 +281,10 @@ class IndexValue(Serializable):
         _data = NDArrayField("data")
         _dtype = DataTypeField("dtype")
 
+        @property
+        def dtype(self):
+            return getattr(self, "_dtype", None)
+
         @property
         def inferred_type(self):
             return "floating"
@@ -1514,8 +1534,7 @@ class BaseDataFrameData(HasShapeTileableData, _ToPandasMixin):
         refresh_index_value(self)
         refresh_dtypes(self)
 
-    def refresh_from_table_meta(self, table_meta: DataFrameTableMeta) -> None:
-        dtypes = table_meta.pd_column_dtypes
+    def refresh_from_dtypes(self, dtypes: pd.Series) -> None:
         self._dtypes = dtypes
         self._columns_value = parse_index(dtypes.index, store_data=True)
         self._dtypes_value = DtypesValue(key=tokenize(dtypes), value=dtypes)
@@ -1523,6 +1542,9 @@ class BaseDataFrameData(HasShapeTileableData, _ToPandasMixin):
         new_shape[-1] = len(dtypes)
         self._shape = tuple(new_shape)
 
+    def refresh_from_table_meta(self, table_meta: DataFrameTableMeta) -> None:
+        self.refresh_from_dtypes(table_meta.pd_column_dtypes)
+
     @property
     def dtypes(self):
         dt = getattr(self, "_dtypes", None)

maxframe/dataframe/datasource/read_odps_query.py

@@ -57,7 +57,7 @@ _EXPLAIN_COLUMN_REGEX = re.compile(r"([^\(]+) \(([^\n]+)\)(?:| AS ([^ ]+))(?:\n|
 _ANONYMOUS_COL_REGEX = re.compile(r"^_c(\d+)$")
 
 _SIMPLE_SCHEMA_COLS_REGEX = re.compile(r"SELECT (([^:]+:[^, ]+[, ]*)+)FROM")
-_SIMPLE_SCHEMA_COL_REGEX = re.compile(r"([^\.]+):([^, ]+)")
+_SIMPLE_SCHEMA_COL_REGEX = re.compile(r"([^ \.\)]+):([^ ]+)")
 
 
 @dataclasses.dataclass
@@ -180,23 +180,30 @@ def _parse_full_explain(explain_string: str) -> OdpsSchema:
 
     job_dag = jobs_sector.build_dag()
     indep_job_names = list(job_dag.iter_indep(reverse=True))
-    if len(indep_job_names) > 1:  # pragma: no cover
-        raise ValueError("Only one final job is allowed in SQL statement")
-
-    tasks_sector = jobs_sector.jobs[indep_job_names[0]]
-    task_dag = tasks_sector.build_dag()
-    indep_task_names = list(task_dag.iter_indep(reverse=True))
-    if len(indep_task_names) > 1:  # pragma: no cover
+    schema_signatures = dict()
+    for job_name in indep_job_names:
+        tasks_sector = jobs_sector.jobs[job_name]
+        task_dag = tasks_sector.build_dag()
+        indep_task_names = list(task_dag.iter_indep(reverse=True))
+        for task_name in indep_task_names:
+            task_sector = tasks_sector.tasks[task_name]
+            if not task_sector.schema:  # pragma: no cover
+                raise ValueError("Cannot detect output schema")
+            if task_sector.output_target != "Screen":
+                raise ValueError("The SQL statement should be an instant query")
+            sig_tuples = sorted(
+                [
+                    (c.column_alias or c.column_name, c.column_type)
+                    for c in task_sector.schema
+                ]
+            )
+            schema_signatures[hash(tuple(sig_tuples))] = task_sector.schema
+    if len(schema_signatures) != 1:
         raise ValueError("Only one final task is allowed in SQL statement")
-
-    task_sector = tasks_sector.tasks[indep_task_names[0]]
-    if not task_sector.schema:  # pragma: no cover
-        raise ValueError("Cannot detect output schema")
-    if task_sector.output_target != "Screen":
-        raise ValueError("The SQL statement should be an instant query")
+    schema = list(schema_signatures.values())[0]
     cols = [
         Column(c.column_alias or c.column_name, validate_data_type(c.column_type))
-        for c in task_sector.schema
+        for c in schema
     ]
     return OdpsSchema(cols)
 
@@ -209,7 +216,7 @@ def _parse_simple_explain(explain_string: str) -> OdpsSchema:
     fields_str = fields_match.group(1)
     cols = []
     for field, type_name in _SIMPLE_SCHEMA_COL_REGEX.findall(fields_str):
-        cols.append(Column(field, validate_data_type(type_name)))
+        cols.append(Column(field, validate_data_type(type_name.rstrip(","))))
     return OdpsSchema(cols)
 
 
@@ -257,12 +264,18 @@ class DataFrameReadODPSQuery(
             )
             index_value = parse_index(idx)
 
-        columns_value = parse_index(self.dtypes.index, store_data=True)
+        if self.dtypes is not None:
+            columns_value = parse_index(self.dtypes.index, store_data=True)
+            shape = (np.nan, len(self.dtypes))
+        else:
+            columns_value = None
+            shape = (np.nan, np.nan)
+
         self.output_types = [OutputType.dataframe]
         return self.new_tileable(
             [],
             None,
-            shape=(len(self.dtypes), np.nan),
+            shape=shape,
             dtypes=self.dtypes,
             index_value=index_value,
             columns_value=columns_value,
@@ -278,6 +291,7 @@ def read_odps_query(
     string_as_binary: bool = None,
     sql_hints: Dict[str, str] = None,
     anonymous_col_prefix: str = _DEFAULT_ANONYMOUS_COL_PREFIX,
+    skip_schema: bool = False,
     **kw,
 ):
     """
@@ -298,6 +312,10 @@ def read_odps_query(
         User specified SQL hints.
     anonymous_col_prefix: str, optional
         Prefix for anonymous columns, '_anon_col_' by default.
+    skip_schema: bool, optional
+        Skip resolving output schema before execution. Once this is configured,
+        the output DataFrame cannot be inputs of other DataFrame operators
+        before execution.
 
     Returns
     -------
@@ -319,28 +337,39 @@ def read_odps_query(
 
     if odps_entry is None:
         raise ValueError("Missing odps_entry parameter")
-    inst = odps_entry.execute_sql(f"EXPLAIN {query}", hints=hints)
-    logger.debug("Explain instance ID: %s", inst.id)
-    explain_str = list(inst.get_task_results().values())[0]
 
-    odps_schema = _parse_explained_schema(explain_str)
-
-    new_columns = []
     col_renames = {}
-    for col in odps_schema.columns:
-        anon_match = _ANONYMOUS_COL_REGEX.match(col.name)
-        if anon_match and col.name not in query:
-            new_name = anonymous_col_prefix + anon_match.group(1)
-            col_renames[col.name] = new_name
-            new_columns.append(Column(new_name, col.type))
-        else:
-            new_columns.append(col)
-
-    dtypes = odps_schema_to_pandas_dtypes(OdpsSchema(new_columns))
+    if not skip_schema:
+        inst = odps_entry.execute_sql(f"EXPLAIN {query}", hints=hints)
+        logger.debug("Explain instance ID: %s", inst.id)
+        explain_str = list(inst.get_task_results().values())[0]
+
+        try:
+            odps_schema = _parse_explained_schema(explain_str)
+        except ValueError as ex:
+            exc = ValueError(str(ex) + "\nExplain instance ID: " + inst.id)
+            raise exc.with_traceback(ex.__traceback__) from None
+
+        new_columns = []
+        for col in odps_schema.columns:
+            anon_match = _ANONYMOUS_COL_REGEX.match(col.name)
+            if anon_match and col.name not in query:
+                new_name = anonymous_col_prefix + anon_match.group(1)
+                col_renames[col.name] = new_name
+                new_columns.append(Column(new_name, col.type))
+            else:
+                new_columns.append(col)
+
+        dtypes = odps_schema_to_pandas_dtypes(OdpsSchema(new_columns))
+    else:
+        dtypes = None
 
     if not index_col:
         index_dtypes = None
     else:
+        if dtypes is None:
+            raise ValueError("Cannot configure index_col when skip_schema is True")
+
         if isinstance(index_col, str):
             index_col = [index_col]
         index_col_set = set(index_col)

maxframe/dataframe/datasource/tests/test_datasource.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import os
+import uuid
 from collections import OrderedDict
 
 import numpy as np
@@ -26,7 +27,14 @@ from ....core import OutputType
 from ....tests.utils import tn
 from ....utils import lazy_import
 from ... import read_odps_query, read_odps_table
-from ...core import DatetimeIndex, Float64Index, IndexValue, Int64Index, MultiIndex
+from ...core import (
+    DatetimeIndex,
+    Float64Index,
+    Index,
+    IndexValue,
+    Int64Index,
+    MultiIndex,
+)
 from ..dataframe import from_pandas as from_pandas_df
 from ..date_range import date_range
 from ..from_tensor import (
@@ -36,7 +44,12 @@ from ..from_tensor import (
 )
 from ..index import from_pandas as from_pandas_index
 from ..index import from_tileable
-from ..read_odps_query import ColumnSchema, _parse_simple_explain, _resolve_task_sector
+from ..read_odps_query import (
+    ColumnSchema,
+    _parse_full_explain,
+    _parse_simple_explain,
+    _resolve_task_sector,
+)
 from ..series import from_pandas as from_pandas_series
 
 ray = lazy_import("ray")
@@ -114,18 +127,22 @@ def test_from_tileable_index():
 
     for o in [df, df[0]]:
         index = o.index
-        assert isinstance(index, Int64Index)
+        assert isinstance(index, (Index, Int64Index))
         assert index.dtype == np.int64
         assert index.name == pd_df.index.name
-        assert isinstance(index.index_value.value, IndexValue.Int64Index)
+        assert isinstance(
+            index.index_value.value, (IndexValue.Int64Index, IndexValue.Index)
+        )
 
     t = mt.random.rand(10, chunk_size=6)
     index = from_tileable(t, name="new_name")
 
-    assert isinstance(index, Float64Index)
+    assert isinstance(index, (Index, Float64Index))
     assert index.dtype == np.float64
     assert index.name == "new_name"
-    assert isinstance(index.index_value.value, IndexValue.Float64Index)
+    assert isinstance(
+        index.index_value.value, (IndexValue.Float64Index, IndexValue.Index)
+    )
 
 
 def test_from_tensor():
@@ -327,7 +344,10 @@ def test_from_odps_query():
     odps_entry.write_table(test_table2, [["A", 10, 4.5]])
 
     with pytest.raises(ValueError) as err_info:
-        read_odps_query(f"CREATE TABLE dummy_table AS SELECT * FROM {table1_name}")
+        read_odps_query(
+            f"CREATE TABLE dummy_table_{uuid.uuid4().hex} "
+            f"AS SELECT * FROM {table1_name}"
+        )
     assert "instant query" in err_info.value.args[0]
 
     query1 = f"SELECT * FROM {table1_name} WHERE col1 > 10"
@@ -343,6 +363,10 @@ def test_from_odps_query():
         ),
     )
 
+    df = read_odps_query(query1, skip_schema=True)
+    assert df.dtypes is None
+    assert df.columns_value is None
+
     df = read_odps_query(query1, index_col="col1")
     assert df.op.query == query1
     assert df.index_value.name == "col1"
@@ -442,3 +466,31 @@ def test_resolve_simple_explain():
     assert schema.columns[0].type == odps_types.string
     assert schema.columns[1].name == "createdate"
     assert schema.columns[1].type == odps_types.bigint
+
+
+def test_resolve_conditional():
+    input_path = os.path.join(
+        os.path.dirname(__file__), "test-data", "task-input-multi-cond.txt"
+    )
+    with open(input_path, "r") as f:
+        sector = f.read()
+
+    expected_col_types = {
+        "cs1": "string",
+        "cs2": "string",
+        "ci1": "bigint",
+        "cs3": "string",
+        "cs4": "string",
+        "cs5": "string",
+        "cs6": "string",
+        "cs7": "string",
+        "cs8": "string",
+        "ci2": "int",
+        "ci3": "bigint",
+        "cs9": "string",
+    }
+
+    schema = _parse_full_explain(sector)
+    for col, (exp_nm, exp_tp) in zip(schema.columns, expected_col_types.items()):
+        assert col.name == exp_nm
+        assert col.type == odps_types.validate_data_type(exp_tp)

maxframe/dataframe/extensions/__init__.py

@@ -18,6 +18,8 @@ from .accessor import (
     IndexMaxFrameAccessor,
     SeriesMaxFrameAccessor,
 )
+from .apply_chunk import df_apply_chunk, series_apply_chunk
+from .flatjson import series_flatjson
 from .flatmap import df_flatmap, series_flatmap
 from .reshuffle import DataFrameReshuffle, df_reshuffle
 
@@ -27,7 +29,10 @@ def _install():
 
     DataFrameMaxFrameAccessor._register("reshuffle", df_reshuffle)
     DataFrameMaxFrameAccessor._register("flatmap", df_flatmap)
+    DataFrameMaxFrameAccessor._register("apply_chunk", df_apply_chunk)
     SeriesMaxFrameAccessor._register("flatmap", series_flatmap)
+    SeriesMaxFrameAccessor._register("flatjson", series_flatjson)
+    SeriesMaxFrameAccessor._register("apply_chunk", series_apply_chunk)
 
     if DataFrameMaxFrameAccessor._api_count:
         for t in DATAFRAME_TYPE: