hequn8128 commented on a change in pull request #9707: [FLINK-14015][python] Introduces PythonScalarFunctionOperator to execute Python user-defined functions URL: https://github.com/apache/flink/pull/9707#discussion_r326526010
########## File path: flink-python/src/main/java/org/apache/flink/table/runtime/operators/python/PythonScalarFunctionOperator.java ########## @@ -0,0 +1,198 @@ +/* + * 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.runtime.operators.python; + +import org.apache.flink.annotation.Internal; +import org.apache.flink.annotation.VisibleForTesting; +import org.apache.flink.api.common.typeinfo.TypeInformation; +import org.apache.flink.api.common.typeutils.TypeSerializer; +import org.apache.flink.api.java.typeutils.RowTypeInfo; +import org.apache.flink.python.PythonFunctionRunner; +import org.apache.flink.table.functions.ScalarFunction; +import org.apache.flink.table.functions.python.PythonFunctionInfo; +import org.apache.flink.table.functions.python.PythonScalarFunctionRunner; +import org.apache.flink.table.runtime.types.CRow; +import org.apache.flink.table.runtime.types.CRowTypeInfo; +import org.apache.flink.table.types.logical.RowType; +import org.apache.flink.table.types.utils.TypeConversions; +import org.apache.flink.types.Row; + +import org.apache.beam.sdk.fn.data.FnDataReceiver; + +import java.util.concurrent.LinkedBlockingQueue; + +/** + * The {@link PythonScalarFunctionOperator} is responsible for executing Python {@link ScalarFunction}s. + * It executes the Python {@link ScalarFunction}s in separate Python execution environment. + * + * <p>The inputs are assumed as the following format: + * {{{ + * +------------------+--------------+ + * | forwarded fields | extra fields | + * +------------------+--------------+ + * }}}. + * + * <p>The Python UDFs may take input columns directly from the input row or the execution result of Java UDFs: + * 1) The input columns from the input row can be referred from the 'forwarded fields'; + * 2) The Java UDFs will be computed and the execution results can be referred from the 'extra fields'. + * + * <p>The outputs will be as the following format: + * {{{ + * +------------------+-------------------------+ + * | forwarded fields | scalar function results | + * +------------------+-------------------------+ + * }}}. + */ +@Internal +public class PythonScalarFunctionOperator extends AbstractPythonScalarFunctionOperator<CRow, CRow> { + + private static final long serialVersionUID = 1L; + + /** + * The collector used to collect records. + */ + private transient StreamRecordCRowWrappingCollector cRowWrapper; + + /** + * The queue holding the input elements for which the execution results have not been received. + */ + private transient LinkedBlockingQueue<CRow> forwardedInputQueue; + + /** + * The queue holding the user-defined function execution results. The execution results are in + * the same order as the input elements. + */ + private transient LinkedBlockingQueue<Row> udfResultQueue; + + /** + * The type serializer for the forwarded fields. + */ + private transient TypeSerializer<CRow> forwardedInputSerializer; + + public PythonScalarFunctionOperator( + PythonFunctionInfo[] scalarFunctions, + RowType inputType, + RowType outputType, + int[] udfInputOffsets, + int forwardedFieldCnt) { + super(scalarFunctions, inputType, outputType, udfInputOffsets, forwardedFieldCnt); + } + + @Override + public void open() throws Exception { + super.open(); + this.cRowWrapper = new StreamRecordCRowWrappingCollector(output); + this.forwardedInputQueue = new LinkedBlockingQueue<>(); + this.udfResultQueue = new LinkedBlockingQueue<>(); + + CRowTypeInfo forwardedInputTypeInfo = new CRowTypeInfo(new RowTypeInfo( + getInputType().getFields().stream() + .limit(getForwardedFieldCnt()) + .map(RowType.RowField::getType) + .map(TypeConversions::fromLogicalToDataType) + .map(TypeConversions::fromDataTypeToLegacyInfo) + .toArray(TypeInformation[]::new))); + forwardedInputSerializer = forwardedInputTypeInfo.createSerializer(getExecutionConfig()); + } + + @Override + public void bufferInput(CRow input) { + CRow forwardedFields = new CRow(getForwardedRow(input.row()), input.change()); + if (getExecutionConfig().isObjectReuseEnabled()) { + forwardedFields = forwardedInputSerializer.copy(forwardedFields); + } + forwardedInputQueue.add(forwardedFields); + } + + @Override + @SuppressWarnings("ConstantConditions") + public void emitResults() { + Row udfResult; + while ((udfResult = udfResultQueue.poll()) != null) { + CRow input = forwardedInputQueue.poll(); + cRowWrapper.setChange(input.change()); + cRowWrapper.collect(Row.join(input.row(), udfResult)); + } + } + + @Override + public PythonFunctionRunner<CRow> createPythonFunctionRunner() { + final FnDataReceiver<Row> udfResultReceiver = input -> { + // handover to queue, do not block the result receiver thread + udfResultQueue.put(input); + }; + + return new PythonScalarFunctionRunnerWrapper(createPythonFunctionRunner(udfResultReceiver)); + } + + @VisibleForTesting + PythonFunctionRunner<Row> createPythonFunctionRunner( + FnDataReceiver<Row> resultReceiver) { + return new PythonScalarFunctionRunner( + getRuntimeContext().getTaskName(), + resultReceiver, + getScalarFunctions(), + getScalarFunctions()[0].getPythonFunction().getPythonEnv(), + getUdfInputType(), + getUdfOutputType(), + getContainingTask().getEnvironment().getTaskManagerInfo().getTmpDirectories()[0]); + } + + private Row getForwardedRow(Row input) { + int[] inputs = new int[getForwardedFieldCnt()]; Review comment: Maybe we can change this method to ``` private Row getForwardedRow(Row input) { Row returnRow = new Row(getForwardedFieldCnt()); for (int i = 0; i < returnRow.getArity(); ++i) { returnRow.setField(i, input.getField(i)); } return returnRow; } ``` Thus, we can: - Avoid new extract inputs array. Need less memory. - Avoid extract loop(only loop once). What do you think? ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
