Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[FLINK-19364][python] Add Batch Physical Pandas Group Window Aggregate Rule and RelNode #13460

Closed
wants to merge 1 commit into from
Closed

[FLINK-19364][python] Add Batch Physical Pandas Group Window Aggregate Rule and RelNode #13460

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
93 changes: 93 additions & 0 deletions flink-python/pyflink/table/tests/test_pandas_udaf.py
View file
Open in desktop
Expand Up @@ -17,6 +17,7 @@
################################################################################
import unittest

from pyflink.table import expressions as expr
from pyflink.table.types import DataTypes
from pyflink.table.udf import udaf, udf, AggregateFunction
from pyflink.testing import source_sink_utils
Expand Down Expand Up @@ -94,6 +95,98 @@ def test_group_aggregate_with_aux_group(self):
actual = source_sink_utils.results()
self.assert_equals(actual, ["1,2,2.0,6", "2,3,3.0,8", "3,4,4.0,10"])

def test_tumble_group_window_aggregate_function(self):
import datetime
from pyflink.table.window import Tumble
t = self.t_env.from_elements(
[
(1, 2, 3, datetime.datetime(2018, 3, 11, 3, 10, 0, 0)),
(3, 2, 4, datetime.datetime(2018, 3, 11, 3, 10, 0, 0)),
(2, 1, 2, datetime.datetime(2018, 3, 11, 3, 10, 0, 0)),
(1, 3, 1, datetime.datetime(2018, 3, 11, 3, 40, 0, 0)),
(1, 8, 5, datetime.datetime(2018, 3, 11, 4, 20, 0, 0)),
(2, 3, 6, datetime.datetime(2018, 3, 11, 3, 30, 0, 0))
],
DataTypes.ROW(
[DataTypes.FIELD("a", DataTypes.TINYINT()),
DataTypes.FIELD("b", DataTypes.SMALLINT()),
DataTypes.FIELD("c", DataTypes.INT()),
DataTypes.FIELD("rowtime", DataTypes.TIMESTAMP(3))]))

table_sink = source_sink_utils.TestAppendSink(
['a', 'b', 'c'],
[
DataTypes.TIMESTAMP(3),
DataTypes.TIMESTAMP(3),
DataTypes.FLOAT()])
self.t_env.register_table_sink("Results", table_sink)
self.t_env.create_temporary_system_function("mean_udaf", mean_udaf)
tumble_window = Tumble.over(expr.lit(1).hours) \
.on(expr.col("rowtime")) \
.alias("w")
t.window(tumble_window) \
.group_by("w") \
.select("w.start, w.end, mean_udaf(b)") \
.execute_insert("Results") \
.wait()

actual = source_sink_utils.results()
self.assert_equals(actual,
["2018-03-11 03:00:00.0,2018-03-11 04:00:00.0,2.2",
"2018-03-11 04:00:00.0,2018-03-11 05:00:00.0,8.0"])

def test_slide_group_window_aggregate_function(self):
import datetime
from pyflink.table.window import Slide
t = self.t_env.from_elements(
[
(1, 2, 3, datetime.datetime(2018, 3, 11, 3, 10, 0, 0)),
(3, 2, 4, datetime.datetime(2018, 3, 11, 3, 10, 0, 0)),
(2, 1, 2, datetime.datetime(2018, 3, 11, 3, 10, 0, 0)),
(1, 3, 1, datetime.datetime(2018, 3, 11, 3, 40, 0, 0)),
(1, 8, 5, datetime.datetime(2018, 3, 11, 4, 20, 0, 0)),
(2, 3, 6, datetime.datetime(2018, 3, 11, 3, 30, 0, 0))
],
DataTypes.ROW(
[DataTypes.FIELD("a", DataTypes.TINYINT()),
DataTypes.FIELD("b", DataTypes.SMALLINT()),
DataTypes.FIELD("c", DataTypes.INT()),
DataTypes.FIELD("rowtime", DataTypes.TIMESTAMP(3))]))

table_sink = source_sink_utils.TestAppendSink(
['a', 'b', 'c', 'd', 'e'],
[
DataTypes.TINYINT(),
DataTypes.TIMESTAMP(3),
DataTypes.TIMESTAMP(3),
DataTypes.FLOAT(),
DataTypes.INT()])
self.t_env.register_table_sink("Results", table_sink)
self.t_env.register_function("max_add", udaf(MaxAdd(),
result_type=DataTypes.INT(),
func_type="pandas"))
self.t_env.create_temporary_system_function("mean_udaf", mean_udaf)
slide_window = Slide.over(expr.lit(1).hours) \
.every(expr.lit(30).minutes) \
.on(expr.col("rowtime")) \
.alias("w")
t.window(slide_window) \
.group_by("a, w") \
.select("a, w.start, w.end, mean_udaf(b), max_add(b, c, 1)") \
.execute_insert("Results") \
.wait()
actual = source_sink_utils.results()
self.assert_equals(actual,
["1,2018-03-11 02:30:00.0,2018-03-11 03:30:00.0,2.0,6",
"1,2018-03-11 03:00:00.0,2018-03-11 04:00:00.0,2.5,7",
"1,2018-03-11 03:30:00.0,2018-03-11 04:30:00.0,5.5,14",
"1,2018-03-11 04:00:00.0,2018-03-11 05:00:00.0,8.0,14",
"2,2018-03-11 02:30:00.0,2018-03-11 03:30:00.0,1.0,4",
"2,2018-03-11 03:00:00.0,2018-03-11 04:00:00.0,2.0,10",
"2,2018-03-11 03:30:00.0,2018-03-11 04:30:00.0,3.0,10",
"3,2018-03-11 03:00:00.0,2018-03-11 04:00:00.0,2.0,7",
"3,2018-03-11 02:30:00.0,2018-03-11 03:30:00.0,2.0,7"])


@udaf(result_type=DataTypes.FLOAT(), func_type="pandas")
def mean_udaf(v):
Expand Down
9 changes: 8 additions & 1 deletion ...rs/python/aggregate/arrow/batch/BatchArrowPythonGroupWindowAggregateFunctionOperator.java
View file
Open in desktop
Expand Up @@ -31,6 +31,7 @@
import org.apache.flink.table.runtime.operators.window.TimeWindow;
import org.apache.flink.table.runtime.operators.window.grouping.HeapWindowsGrouping;
import org.apache.flink.table.runtime.operators.window.grouping.WindowsGrouping;
import org.apache.flink.table.runtime.typeutils.RowDataSerializer;
import org.apache.flink.table.runtime.util.RowIterator;
import org.apache.flink.table.types.logical.RowType;

Expand Down Expand Up @@ -92,6 +93,11 @@ public class BatchArrowPythonGroupWindowAggregateFunctionOperator
*/
private transient LinkedList<Tuple2<RowData, TimeWindow>> inputKeyAndWindow;

/**
* The type serializer for the forwarded fields.
*/
private transient RowDataSerializer forwardedInputSerializer;

public BatchArrowPythonGroupWindowAggregateFunctionOperator(
Configuration config,
PythonFunctionInfo[] pandasAggFunctions,
Expand Down Expand Up @@ -122,6 +128,7 @@ public void open() throws Exception {
windowAggResult = new JoinedRowData();
windowsGrouping = new HeapWindowsGrouping(
maxLimitSize, windowSize, slideSize, inputTimeFieldIndex, false);
forwardedInputSerializer = new RowDataSerializer(inputType);
super.open();
}

Expand Down Expand Up @@ -152,7 +159,7 @@ protected void invokeCurrentBatch() throws Exception {

@Override
public void processElementInternal(RowData value) throws Exception {
windowsGrouping.addInputToBuffer((BinaryRowData) value);
windowsGrouping.addInputToBuffer(forwardedInputSerializer.toBinaryRow(value).copy());
triggerWindowProcess();
}

Expand Down
167 changes: 167 additions & 0 deletions ...che/flink/table/planner/plan/rules/physical/batch/BatchExecPythonWindowAggregateRule.java
View file
Open in desktop
@@ -0,0 +1,167 @@
/*
* 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.
*/

package org.apache.flink.table.planner.plan.rules.physical.batch;

import org.apache.flink.table.api.TableException;
import org.apache.flink.table.functions.UserDefinedFunction;
import org.apache.flink.table.functions.python.PythonFunctionKind;
import org.apache.flink.table.planner.calcite.FlinkRelFactories;
import org.apache.flink.table.planner.plan.logical.LogicalWindow;
import org.apache.flink.table.planner.plan.logical.SessionGroupWindow;
import org.apache.flink.table.planner.plan.logical.SlidingGroupWindow;
import org.apache.flink.table.planner.plan.logical.TumblingGroupWindow;
import org.apache.flink.table.planner.plan.nodes.FlinkConventions;
import org.apache.flink.table.planner.plan.nodes.logical.FlinkLogicalWindowAggregate;
import org.apache.flink.table.planner.plan.nodes.physical.batch.BatchExecPythonGroupWindowAggregate;
import org.apache.flink.table.planner.plan.trait.FlinkRelDistribution;
import org.apache.flink.table.planner.plan.utils.AggregateUtil;
import org.apache.flink.table.planner.plan.utils.FlinkRelOptUtil;
import org.apache.flink.table.planner.plan.utils.PythonUtil;
import org.apache.flink.table.types.DataType;

import org.apache.calcite.plan.RelOptRule;
import org.apache.calcite.plan.RelOptRuleCall;
import org.apache.calcite.plan.RelTraitSet;
import org.apache.calcite.rel.RelCollation;
import org.apache.calcite.rel.RelCollations;
import org.apache.calcite.rel.RelFieldCollation;
import org.apache.calcite.rel.RelNode;
import org.apache.calcite.rel.core.AggregateCall;
import org.apache.calcite.rel.type.RelDataType;
import org.apache.calcite.sql.type.SqlTypeName;

import java.util.LinkedList;
import java.util.List;

import scala.Tuple2;
import scala.Tuple3;
import scala.collection.Seq;

/**
* The physical rule is responsible for convert {@link FlinkLogicalWindowAggregate} to
* {@link BatchExecPythonGroupWindowAggregate}.
*/
public class BatchExecPythonWindowAggregateRule extends RelOptRule {

public static final RelOptRule INSTANCE = new BatchExecPythonWindowAggregateRule();

private BatchExecPythonWindowAggregateRule() {
super(operand(FlinkLogicalWindowAggregate.class,
operand(RelNode.class, any())),
FlinkRelFactories.LOGICAL_BUILDER_WITHOUT_AGG_INPUT_PRUNE(),
"BatchExecPythonWindowAggregateRule");
}

@Override
public boolean matches(RelOptRuleCall call) {
FlinkLogicalWindowAggregate agg = call.rel(0);
List<AggregateCall> aggCalls = agg.getAggCallList();

boolean existGeneralPythonFunction =
aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.GENERAL));
boolean existPandasFunction =
aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.PANDAS));
boolean existJavaFunction =
aggCalls.stream().anyMatch(x -> !PythonUtil.isPythonAggregate(x, null));
if (existPandasFunction || existGeneralPythonFunction) {
if (existJavaFunction) {
throw new TableException("Python UDAF and Java/Scala UDAF cannot be used together.");
}
if (existPandasFunction && existGeneralPythonFunction) {
throw new TableException("Pandas UDAF and non-Pandas UDAF cannot be used together.");
}
return true;
} else {
return false;
}

}

@Override
public void onMatch(RelOptRuleCall call) {
FlinkLogicalWindowAggregate agg = call.rel(0);
RelNode input = agg.getInput();
LogicalWindow window = agg.getWindow();

if (!(window instanceof TumblingGroupWindow &&
AggregateUtil.hasTimeIntervalType(((TumblingGroupWindow) window).size())
|| window instanceof SlidingGroupWindow &&
AggregateUtil.hasTimeIntervalType(((SlidingGroupWindow) window).size())
|| window instanceof SessionGroupWindow)) {
// sliding & tumbling count window and session window not supported
throw new TableException("Window " + window + " is not supported right now.");
}

int[] groupSet = agg.getGroupSet().toArray();
RelTraitSet traitSet = agg.getTraitSet().replace(FlinkConventions.BATCH_PHYSICAL());

Tuple2<int[], Seq<AggregateCall>> auxGroupSetAndCallsTuple = AggregateUtil.checkAndSplitAggCalls(agg);
int[] auxGroupSet = auxGroupSetAndCallsTuple._1;
Seq<AggregateCall> aggCallsWithoutAuxGroupCalls = auxGroupSetAndCallsTuple._2;

Tuple3<int[][], DataType[][], UserDefinedFunction[]> aggBufferTypesAndFunctions =
AggregateUtil.transformToBatchAggregateFunctions(
aggCallsWithoutAuxGroupCalls, input.getRowType(), null);
UserDefinedFunction[] aggFunctions = aggBufferTypesAndFunctions._3();

int inputTimeFieldIndex = AggregateUtil.timeFieldIndex(
input.getRowType(), call.builder(), window.timeAttribute());
RelDataType inputTimeFieldType = input.getRowType().getFieldList().get(inputTimeFieldIndex).getType();
boolean inputTimeIsDate = inputTimeFieldType.getSqlTypeName() == SqlTypeName.DATE;

RelTraitSet requiredTraitSet = agg.getTraitSet().replace(FlinkConventions.BATCH_PHYSICAL());
if (groupSet.length != 0) {
FlinkRelDistribution requiredDistribution =
FlinkRelDistribution.hash(groupSet, false);
requiredTraitSet = requiredTraitSet.replace(requiredDistribution);
} else {
requiredTraitSet = requiredTraitSet.replace(FlinkRelDistribution.SINGLETON());
}

RelCollation sortCollation = createRelCollation(groupSet, inputTimeFieldIndex);
requiredTraitSet = requiredTraitSet.replace(sortCollation);

RelNode newInput = RelOptRule.convert(input, requiredTraitSet);
BatchExecPythonGroupWindowAggregate windowAgg = new BatchExecPythonGroupWindowAggregate(
agg.getCluster(),
call.builder(),
traitSet,
newInput,
agg.getRowType(),
newInput.getRowType(),
groupSet,
auxGroupSet,
aggCallsWithoutAuxGroupCalls,
aggFunctions,
window,
inputTimeFieldIndex,
inputTimeIsDate,
agg.getNamedProperties());
call.transformTo(windowAgg);
}

private RelCollation createRelCollation(int[] groupSet, int timeIndex) {
List<RelFieldCollation> fields = new LinkedList<>();
for (int value : groupSet) {
fields.add(FlinkRelOptUtil.ofRelFieldCollation(value));
}
fields.add(FlinkRelOptUtil.ofRelFieldCollation(timeIndex));
return RelCollations.of(fields);
}
}