# # Licensed to the Apache Software Foundation (ASF) under one or more # contributor license agreements. See the NOTICE file distributed with # this work for additional information regarding copyright ownership. # The ASF licenses this file to You under the Apache License, Version 2.0 # (the "License"); you may not use this file except in compliance with # the License. You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # 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 sys from collections import Counter from typing import List, Optional, Type, Union, no_type_check, overload, TYPE_CHECKING from warnings import catch_warnings, simplefilter, warn from pyspark.rdd import _load_from_socket from pyspark.sql.pandas.serializers import ArrowCollectSerializer from pyspark.sql.types import ( IntegralType, ByteType, ShortType, IntegerType, LongType, FloatType, DoubleType, BooleanType, MapType, TimestampType, TimestampNTZType, DayTimeIntervalType, StructType, DataType, ) from pyspark.sql.utils import is_timestamp_ntz_preferred from pyspark.traceback_utils import SCCallSiteSync if TYPE_CHECKING: import numpy as np import pyarrow as pa from py4j.java_gateway import JavaObject from pyspark.sql.pandas._typing import DataFrameLike as PandasDataFrameLike from pyspark.sql import DataFrame class PandasConversionMixin: """ Mix-in for the conversion from Spark to pandas. Currently, only :class:`DataFrame` can use this class. """ def toPandas(self) -> "PandasDataFrameLike": """ Returns the contents of this :class:`DataFrame` as Pandas ``pandas.DataFrame``. This is only available if Pandas is installed and available. .. versionadded:: 1.3.0 Notes ----- This method should only be used if the resulting Pandas ``pandas.DataFrame`` is expected to be small, as all the data is loaded into the driver's memory. Usage with ``spark.sql.execution.arrow.pyspark.enabled=True`` is experimental. Examples -------- >>> df.toPandas() # doctest: +SKIP age name 0 2 Alice 1 5 Bob """ from pyspark.sql.dataframe import DataFrame assert isinstance(self, DataFrame) from pyspark.sql.pandas.utils import require_minimum_pandas_version require_minimum_pandas_version() import numpy as np import pandas as pd from pandas.core.dtypes.common import is_timedelta64_dtype jconf = self.sparkSession._jconf timezone = jconf.sessionLocalTimeZone() if jconf.arrowPySparkEnabled(): use_arrow = True try: from pyspark.sql.pandas.types import to_arrow_schema from pyspark.sql.pandas.utils import require_minimum_pyarrow_version require_minimum_pyarrow_version() to_arrow_schema(self.schema) except Exception as e: if jconf.arrowPySparkFallbackEnabled(): msg = ( "toPandas attempted Arrow optimization because " "'spark.sql.execution.arrow.pyspark.enabled' is set to true; however, " "failed by the reason below:\n %s\n" "Attempting non-optimization as " "'spark.sql.execution.arrow.pyspark.fallback.enabled' is set to " "true." % str(e) ) warn(msg) use_arrow = False else: msg = ( "toPandas attempted Arrow optimization because " "'spark.sql.execution.arrow.pyspark.enabled' is set to true, but has " "reached the error below and will not continue because automatic fallback " "with 'spark.sql.execution.arrow.pyspark.fallback.enabled' has been set to " "false.\n %s" % str(e) ) warn(msg) raise # Try to use Arrow optimization when the schema is supported and the required version # of PyArrow is found, if 'spark.sql.execution.arrow.pyspark.enabled' is enabled. if use_arrow: try: from pyspark.sql.pandas.types import ( _check_series_localize_timestamps, _convert_map_items_to_dict, ) import pyarrow # Rename columns to avoid duplicated column names. tmp_column_names = ["col_{}".format(i) for i in range(len(self.columns))] self_destruct = jconf.arrowPySparkSelfDestructEnabled() batches = self.toDF(*tmp_column_names)._collect_as_arrow( split_batches=self_destruct ) if len(batches) > 0: table = pyarrow.Table.from_batches(batches) # Ensure only the table has a reference to the batches, so that # self_destruct (if enabled) is effective del batches # Pandas DataFrame created from PyArrow uses datetime64[ns] for date type # values, but we should use datetime.date to match the behavior with when # Arrow optimization is disabled. pandas_options = {"date_as_object": True} if self_destruct: # Configure PyArrow to use as little memory as possible: # self_destruct - free columns as they are converted # split_blocks - create a separate Pandas block for each column # use_threads - convert one column at a time pandas_options.update( { "self_destruct": True, "split_blocks": True, "use_threads": False, } ) pdf = table.to_pandas(**pandas_options) # Rename back to the original column names. pdf.columns = self.columns for field in self.schema: if isinstance(field.dataType, TimestampType): pdf[field.name] = _check_series_localize_timestamps( pdf[field.name], timezone ) elif isinstance(field.dataType, MapType): pdf[field.name] = _convert_map_items_to_dict(pdf[field.name]) return pdf else: corrected_panda_types = {} for index, field in enumerate(self.schema): pandas_type = PandasConversionMixin._to_corrected_pandas_type( field.dataType ) corrected_panda_types[tmp_column_names[index]] = ( np.object0 if pandas_type is None else pandas_type ) pdf = pd.DataFrame(columns=tmp_column_names).astype( dtype=corrected_panda_types ) pdf.columns = self.columns return pdf except Exception as e: # We might have to allow fallback here as well but multiple Spark jobs can # be executed. So, simply fail in this case for now. msg = ( "toPandas attempted Arrow optimization because " "'spark.sql.execution.arrow.pyspark.enabled' is set to true, but has " "reached the error below and can not continue. Note that " "'spark.sql.execution.arrow.pyspark.fallback.enabled' does not have an " "effect on failures in the middle of " "computation.\n %s" % str(e) ) warn(msg) raise # Below is toPandas without Arrow optimization. pdf = pd.DataFrame.from_records(self.collect(), columns=self.columns) column_counter = Counter(self.columns) corrected_dtypes: List[Optional[Type]] = [None] * len(self.schema) for index, field in enumerate(self.schema): # We use `iloc` to access columns with duplicate column names. if column_counter[field.name] > 1: pandas_col = pdf.iloc[:, index] else: pandas_col = pdf[field.name] pandas_type = PandasConversionMixin._to_corrected_pandas_type(field.dataType) # SPARK-21766: if an integer field is nullable and has null values, it can be # inferred by pandas as a float column. If we convert the column with NaN back # to integer type e.g., np.int16, we will hit an exception. So we use the # pandas-inferred float type, rather than the corrected type from the schema # in this case. if pandas_type is not None and not ( isinstance(field.dataType, IntegralType) and field.nullable and pandas_col.isnull().any() ): corrected_dtypes[index] = pandas_type # Ensure we fall back to nullable numpy types. if isinstance(field.dataType, IntegralType) and pandas_col.isnull().any(): corrected_dtypes[index] = np.float64 if isinstance(field.dataType, BooleanType) and pandas_col.isnull().any(): corrected_dtypes[index] = np.object # type: ignore[attr-defined] df = pd.DataFrame() for index, t in enumerate(corrected_dtypes): column_name = self.schema[index].name # We use `iloc` to access columns with duplicate column names. if column_counter[column_name] > 1: series = pdf.iloc[:, index] else: series = pdf[column_name] # No need to cast for non-empty series for timedelta. The type is already correct. should_check_timedelta = is_timedelta64_dtype(t) and len(pdf) == 0 if (t is not None and not is_timedelta64_dtype(t)) or should_check_timedelta: series = series.astype(t, copy=False) with catch_warnings(): from pandas.errors import PerformanceWarning simplefilter(action="ignore", category=PerformanceWarning) # `insert` API makes copy of data, # we only do it for Series of duplicate column names. # `pdf.iloc[:, index] = pdf.iloc[:, index]...` doesn't always work # because `iloc` could return a view or a copy depending by context. if column_counter[column_name] > 1: df.insert(index, column_name, series, allow_duplicates=True) else: df[column_name] = series if timezone is None: return df else: from pyspark.sql.pandas.types import _check_series_convert_timestamps_local_tz for field in self.schema: # TODO: handle nested timestamps, such as ArrayType(TimestampType())? if isinstance(field.dataType, TimestampType): df[field.name] = _check_series_convert_timestamps_local_tz( df[field.name], timezone ) return df @staticmethod def _to_corrected_pandas_type(dt: DataType) -> Optional[Type]: """ When converting Spark SQL records to Pandas `pandas.DataFrame`, the inferred data type may be wrong. This method gets the corrected data type for Pandas if that type may be inferred incorrectly. """ import numpy as np if type(dt) == ByteType: return np.int8 elif type(dt) == ShortType: return np.int16 elif type(dt) == IntegerType: return np.int32 elif type(dt) == LongType: return np.int64 elif type(dt) == FloatType: return np.float32 elif type(dt) == DoubleType: return np.float64 elif type(dt) == BooleanType: return np.bool # type: ignore[attr-defined] elif type(dt) == TimestampType: return np.datetime64 elif type(dt) == TimestampNTZType: return np.datetime64 elif type(dt) == DayTimeIntervalType: return np.timedelta64 else: return None def _collect_as_arrow(self, split_batches: bool = False) -> List["pa.RecordBatch"]: """ Returns all records as a list of ArrowRecordBatches, pyarrow must be installed and available on driver and worker Python environments. This is an experimental feature. :param split_batches: split batches such that each column is in its own allocation, so that the selfDestruct optimization is effective; default False. .. note:: Experimental. """ from pyspark.sql.dataframe import DataFrame assert isinstance(self, DataFrame) with SCCallSiteSync(self._sc): ( port, auth_secret, jsocket_auth_server, ) = self._jdf.collectAsArrowToPython() # Collect list of un-ordered batches where last element is a list of correct order indices try: batch_stream = _load_from_socket((port, auth_secret), ArrowCollectSerializer()) if split_batches: # When spark.sql.execution.arrow.pyspark.selfDestruct.enabled, ensure # each column in each record batch is contained in its own allocation. # Otherwise, selfDestruct does nothing; it frees each column as its # converted, but each column will actually be a list of slices of record # batches, and so no memory is actually freed until all columns are # converted. import pyarrow as pa results = [] for batch_or_indices in batch_stream: if isinstance(batch_or_indices, pa.RecordBatch): batch_or_indices = pa.RecordBatch.from_arrays( [ # This call actually reallocates the array pa.concat_arrays([array]) for array in batch_or_indices ], schema=batch_or_indices.schema, ) results.append(batch_or_indices) else: results = list(batch_stream) finally: # Join serving thread and raise any exceptions from collectAsArrowToPython jsocket_auth_server.getResult() # Separate RecordBatches from batch order indices in results batches = results[:-1] batch_order = results[-1] # Re-order the batch list using the correct order return [batches[i] for i in batch_order] class SparkConversionMixin: """ Min-in for the conversion from pandas to Spark. Currently, only :class:`SparkSession` can use this class. """ _jsparkSession: "JavaObject" @overload def createDataFrame( self, data: "PandasDataFrameLike", samplingRatio: Optional[float] = ... ) -> "DataFrame": ... @overload def createDataFrame( self, data: "PandasDataFrameLike", schema: Union[StructType, str], verifySchema: bool = ..., ) -> "DataFrame": ... def createDataFrame( # type: ignore[misc] self, data: "PandasDataFrameLike", schema: Optional[Union[StructType, List[str]]] = None, samplingRatio: Optional[float] = None, verifySchema: bool = True, ) -> "DataFrame": from pyspark.sql import SparkSession assert isinstance(self, SparkSession) from pyspark.sql.pandas.utils import require_minimum_pandas_version require_minimum_pandas_version() timezone = self._jconf.sessionLocalTimeZone() # If no schema supplied by user then get the names of columns only if schema is None: schema = [str(x) if not isinstance(x, str) else x for x in data.columns] if self._jconf.arrowPySparkEnabled() and len(data) > 0: try: return self._create_from_pandas_with_arrow(data, schema, timezone) except Exception as e: if self._jconf.arrowPySparkFallbackEnabled(): msg = ( "createDataFrame attempted Arrow optimization because " "'spark.sql.execution.arrow.pyspark.enabled' is set to true; however, " "failed by the reason below:\n %s\n" "Attempting non-optimization as " "'spark.sql.execution.arrow.pyspark.fallback.enabled' is set to " "true." % str(e) ) warn(msg) else: msg = ( "createDataFrame attempted Arrow optimization because " "'spark.sql.execution.arrow.pyspark.enabled' is set to true, but has " "reached the error below and will not continue because automatic " "fallback with 'spark.sql.execution.arrow.pyspark.fallback.enabled' " "has been set to false.\n %s" % str(e) ) warn(msg) raise converted_data = self._convert_from_pandas(data, schema, timezone) return self._create_dataframe(converted_data, schema, samplingRatio, verifySchema) def _convert_from_pandas( self, pdf: "PandasDataFrameLike", schema: Union[StructType, str, List[str]], timezone: str ) -> List: """ Convert a pandas.DataFrame to list of records that can be used to make a DataFrame Returns ------- list list of records """ import pandas as pd from pyspark.sql import SparkSession assert isinstance(self, SparkSession) if timezone is not None: from pyspark.sql.pandas.types import _check_series_convert_timestamps_tz_local from pandas.core.dtypes.common import is_datetime64tz_dtype, is_timedelta64_dtype copied = False if isinstance(schema, StructType): for field in schema: # TODO: handle nested timestamps, such as ArrayType(TimestampType())? if isinstance(field.dataType, TimestampType): s = _check_series_convert_timestamps_tz_local(pdf[field.name], timezone) if s is not pdf[field.name]: if not copied: # Copy once if the series is modified to prevent the original # Pandas DataFrame from being updated pdf = pdf.copy() copied = True pdf[field.name] = s else: should_localize = not is_timestamp_ntz_preferred() for column, series in pdf.iteritems(): s = series if should_localize and is_datetime64tz_dtype(s.dtype) and s.dt.tz is not None: s = _check_series_convert_timestamps_tz_local(series, timezone) if s is not series: if not copied: # Copy once if the series is modified to prevent the original # Pandas DataFrame from being updated pdf = pdf.copy() copied = True pdf[column] = s for column, series in pdf.iteritems(): if is_timedelta64_dtype(series): if not copied: pdf = pdf.copy() copied = True # Explicitly set the timedelta as object so the output of numpy records can # hold the timedelta instances as are. Otherwise, it converts to the internal # numeric values. ser = pdf[column] pdf[column] = pd.Series( ser.dt.to_pytimedelta(), index=ser.index, dtype="object", name=ser.name ) # Convert pandas.DataFrame to list of numpy records np_records = pdf.to_records(index=False) # Check if any columns need to be fixed for Spark to infer properly if len(np_records) > 0: record_dtype = self._get_numpy_record_dtype(np_records[0]) if record_dtype is not None: return [r.astype(record_dtype).tolist() for r in np_records] # Convert list of numpy records to python lists return [r.tolist() for r in np_records] def _get_numpy_record_dtype(self, rec: "np.recarray") -> Optional["np.dtype"]: """ Used when converting a pandas.DataFrame to Spark using to_records(), this will correct the dtypes of fields in a record so they can be properly loaded into Spark. Parameters ---------- rec : numpy.record a numpy record to check field dtypes Returns ------- numpy.dtype corrected dtype for a numpy.record or None if no correction needed """ import numpy as np cur_dtypes = rec.dtype col_names = cur_dtypes.names record_type_list = [] has_rec_fix = False for i in range(len(cur_dtypes)): curr_type = cur_dtypes[i] # If type is a datetime64 timestamp, convert to microseconds # NOTE: if dtype is datetime[ns] then np.record.tolist() will output values as longs, # conversion from [us] or lower will lead to py datetime objects, see SPARK-22417 if curr_type == np.dtype("datetime64[ns]"): curr_type = "datetime64[us]" has_rec_fix = True record_type_list.append((str(col_names[i]), curr_type)) return np.dtype(record_type_list) if has_rec_fix else None def _create_from_pandas_with_arrow( self, pdf: "PandasDataFrameLike", schema: Union[StructType, List[str]], timezone: str ) -> "DataFrame": """ Create a DataFrame from a given pandas.DataFrame by slicing it into partitions, converting to Arrow data, then sending to the JVM to parallelize. If a schema is passed in, the data types will be used to coerce the data in Pandas to Arrow conversion. """ from pyspark.sql import SparkSession from pyspark.sql.dataframe import DataFrame assert isinstance(self, SparkSession) from pyspark.sql.pandas.serializers import ArrowStreamPandasSerializer from pyspark.sql.types import TimestampType from pyspark.sql.pandas.types import from_arrow_type, to_arrow_type from pyspark.sql.pandas.utils import ( require_minimum_pandas_version, require_minimum_pyarrow_version, ) require_minimum_pandas_version() require_minimum_pyarrow_version() from pandas.api.types import ( # type: ignore[attr-defined] is_datetime64_dtype, is_datetime64tz_dtype, ) import pyarrow as pa # Create the Spark schema from list of names passed in with Arrow types if isinstance(schema, (list, tuple)): arrow_schema = pa.Schema.from_pandas(pdf, preserve_index=False) struct = StructType() prefer_timestamp_ntz = is_timestamp_ntz_preferred() for name, field in zip(schema, arrow_schema): struct.add( name, from_arrow_type(field.type, prefer_timestamp_ntz), nullable=field.nullable ) schema = struct # Determine arrow types to coerce data when creating batches if isinstance(schema, StructType): arrow_types = [to_arrow_type(f.dataType) for f in schema.fields] elif isinstance(schema, DataType): raise ValueError("Single data type %s is not supported with Arrow" % str(schema)) else: # Any timestamps must be coerced to be compatible with Spark arrow_types = [ to_arrow_type(TimestampType()) if is_datetime64_dtype(t) or is_datetime64tz_dtype(t) else None for t in pdf.dtypes ] # Slice the DataFrame to be batched step = -(-len(pdf) // self.sparkContext.defaultParallelism) # round int up pdf_slices = (pdf.iloc[start : start + step] for start in range(0, len(pdf), step)) # Create list of Arrow (columns, type) for serializer dump_stream arrow_data = [ [(c, t) for (_, c), t in zip(pdf_slice.iteritems(), arrow_types)] for pdf_slice in pdf_slices ] jsparkSession = self._jsparkSession safecheck = self._jconf.arrowSafeTypeConversion() col_by_name = True # col by name only applies to StructType columns, can't happen here ser = ArrowStreamPandasSerializer(timezone, safecheck, col_by_name) @no_type_check def reader_func(temp_filename): return self._jvm.PythonSQLUtils.readArrowStreamFromFile(jsparkSession, temp_filename) @no_type_check def create_RDD_server(): return self._jvm.ArrowRDDServer(jsparkSession) # Create Spark DataFrame from Arrow stream file, using one batch per partition jrdd = self._sc._serialize_to_jvm(arrow_data, ser, reader_func, create_RDD_server) assert self._jvm is not None jdf = self._jvm.PythonSQLUtils.toDataFrame(jrdd, schema.json(), jsparkSession) df = DataFrame(jdf, self) df._schema = schema return df def _test() -> None: import doctest from pyspark.sql import SparkSession import pyspark.sql.pandas.conversion globs = pyspark.sql.pandas.conversion.__dict__.copy() spark = ( SparkSession.builder.master("local[4]").appName("sql.pandas.conversion tests").getOrCreate() ) globs["spark"] = spark (failure_count, test_count) = doctest.testmod( pyspark.sql.pandas.conversion, globs=globs, optionflags=doctest.ELLIPSIS | doctest.NORMALIZE_WHITESPACE | doctest.REPORT_NDIFF, ) spark.stop() if failure_count: sys.exit(-1) if __name__ == "__main__": _test()