alex-balikov commented on code in PR #37893: URL: https://github.com/apache/spark/pull/37893#discussion_r974707164
########## sql/core/src/main/scala/org/apache/spark/sql/execution/python/ApplyInPandasWithStatePythonRunner.scala: ########## @@ -0,0 +1,201 @@ +/* + * 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.spark.sql.execution.python + +import java.io._ + +import scala.collection.JavaConverters._ + +import org.apache.arrow.vector.VectorSchemaRoot +import org.apache.arrow.vector.ipc.ArrowStreamWriter +import org.json4s._ +import org.json4s.jackson.JsonMethods._ + +import org.apache.spark.api.python._ +import org.apache.spark.sql.Row +import org.apache.spark.sql.api.python.PythonSQLUtils +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.encoders.ExpressionEncoder +import org.apache.spark.sql.catalyst.expressions.UnsafeRow +import org.apache.spark.sql.execution.python.ApplyInPandasWithStatePythonRunner.{InType, OutType, OutTypeForState, STATE_METADATA_SCHEMA_FROM_PYTHON_WORKER} +import org.apache.spark.sql.execution.python.ApplyInPandasWithStateWriter.STATE_METADATA_SCHEMA +import org.apache.spark.sql.execution.streaming.GroupStateImpl +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.sql.types._ +import org.apache.spark.sql.vectorized.{ArrowColumnVector, ColumnarBatch} + + +/** + * A variant implementation of [[ArrowPythonRunner]] to serve the operation + * applyInPandasWithState. + */ +class ApplyInPandasWithStatePythonRunner( + funcs: Seq[ChainedPythonFunctions], + evalType: Int, + argOffsets: Array[Array[Int]], + inputSchema: StructType, + override protected val timeZoneId: String, + initialWorkerConf: Map[String, String], + stateEncoder: ExpressionEncoder[Row], + keySchema: StructType, + valueSchema: StructType, + stateValueSchema: StructType, + softLimitBytesPerBatch: Long, + minDataCountForSample: Int, + softTimeoutMillsPurgeBatch: Long) + extends BasePythonRunner[InType, OutType](funcs, evalType, argOffsets) + with PythonArrowInput[InType] + with PythonArrowOutput[OutType] { + + override protected val schema: StructType = inputSchema.add("__state", STATE_METADATA_SCHEMA) + + override val simplifiedTraceback: Boolean = SQLConf.get.pysparkSimplifiedTraceback + + override val bufferSize: Int = SQLConf.get.pandasUDFBufferSize + require( + bufferSize >= 4, + "Pandas execution requires more than 4 bytes. Please set higher buffer. " + + s"Please change '${SQLConf.PANDAS_UDF_BUFFER_SIZE.key}'.") + + // applyInPandasWithState has its own mechanism to construct the Arrow RecordBatch instance. + // Configurations are both applied to executor and Python worker, set them to the worker conf + // to let Python worker read the config properly. + override protected val workerConf: Map[String, String] = initialWorkerConf + + (SQLConf.MAP_PANDAS_UDF_WITH_STATE_SOFT_LIMIT_SIZE_PER_BATCH.key -> + softLimitBytesPerBatch.toString) + + (SQLConf.MAP_PANDAS_UDF_WITH_STATE_MIN_DATA_COUNT_FOR_SAMPLE.key -> + minDataCountForSample.toString) + + (SQLConf.MAP_PANDAS_UDF_WITH_STATE_SOFT_TIMEOUT_PURGE_BATCH.key -> + softTimeoutMillsPurgeBatch.toString) + + private val stateRowDeserializer = stateEncoder.createDeserializer() + + override protected def handleMetadataBeforeExec(stream: DataOutputStream): Unit = { + super.handleMetadataBeforeExec(stream) + // Also write the schema for state value + PythonRDD.writeUTF(stateValueSchema.json, stream) + } + + protected def writeIteratorToArrowStream( + root: VectorSchemaRoot, + writer: ArrowStreamWriter, + dataOut: DataOutputStream, + inputIterator: Iterator[InType]): Unit = { + val w = new ApplyInPandasWithStateWriter(root, writer, softLimitBytesPerBatch, + minDataCountForSample, softTimeoutMillsPurgeBatch) + + while (inputIterator.hasNext) { + val (keyRow, groupState, dataIter) = inputIterator.next() + assert(dataIter.hasNext, "should have at least one data row!") + w.startNewGroup(keyRow, groupState) + + while (dataIter.hasNext) { + val dataRow = dataIter.next() + w.writeRow(dataRow) + } + + w.finalizeGroup() + } + + w.finalizeData() + } + + protected def deserializeColumnarBatch(batch: ColumnarBatch, schema: StructType): OutType = { Review Comment: method level comments ########## sql/core/src/main/scala/org/apache/spark/sql/execution/python/ApplyInPandasWithStateWriter.scala: ########## @@ -0,0 +1,246 @@ +/* + * 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.spark.sql.execution.python + +import scala.collection.JavaConverters._ + +import org.apache.arrow.vector.{FieldVector, VectorSchemaRoot} +import org.apache.arrow.vector.ipc.ArrowStreamWriter + +import org.apache.spark.sql.Row +import org.apache.spark.sql.api.python.PythonSQLUtils +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.expressions.{GenericInternalRow, UnsafeRow} +import org.apache.spark.sql.execution.arrow.ArrowWriter +import org.apache.spark.sql.execution.arrow.ArrowWriter.createFieldWriter +import org.apache.spark.sql.execution.streaming.GroupStateImpl +import org.apache.spark.sql.types.{BinaryType, BooleanType, IntegerType, StringType, StructField, StructType} +import org.apache.spark.unsafe.types.UTF8String + +/** + * This class abstracts the complexity on constructing Arrow RecordBatches for data and state with + * bin-packing and chunking. The caller only need to call the proper public methods of this class + * `startNewGroup`, `writeRow`, `finalizeGroup`, `finalizeData` and this class will write the data + * and state into Arrow RecordBatches with performing bin-pack and chunk internally. + * + * This class requires that the parameter `root` has initialized with the Arrow schema like below: Review Comment: the parameter 'root' has *been* initialized ########## sql/core/src/main/scala/org/apache/spark/sql/execution/python/ApplyInPandasWithStateWriter.scala: ########## @@ -0,0 +1,246 @@ +/* + * 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.spark.sql.execution.python + +import scala.collection.JavaConverters._ + +import org.apache.arrow.vector.{FieldVector, VectorSchemaRoot} +import org.apache.arrow.vector.ipc.ArrowStreamWriter + +import org.apache.spark.sql.Row +import org.apache.spark.sql.api.python.PythonSQLUtils +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.expressions.{GenericInternalRow, UnsafeRow} +import org.apache.spark.sql.execution.arrow.ArrowWriter +import org.apache.spark.sql.execution.arrow.ArrowWriter.createFieldWriter +import org.apache.spark.sql.execution.streaming.GroupStateImpl +import org.apache.spark.sql.types.{BinaryType, BooleanType, IntegerType, StringType, StructField, StructType} +import org.apache.spark.unsafe.types.UTF8String + +/** + * This class abstracts the complexity on constructing Arrow RecordBatches for data and state with + * bin-packing and chunking. The caller only need to call the proper public methods of this class + * `startNewGroup`, `writeRow`, `finalizeGroup`, `finalizeData` and this class will write the data + * and state into Arrow RecordBatches with performing bin-pack and chunk internally. + * + * This class requires that the parameter `root` has initialized with the Arrow schema like below: + * - data fields + * - state field + * - nested schema (Refer ApplyInPandasWithStateWriter.STATE_METADATA_SCHEMA) + * + * Please refer the code comment in the implementation to see how the writes of data and state + * against Arrow RecordBatch work with consideration of bin-packing and chunking. + */ +class ApplyInPandasWithStateWriter( + root: VectorSchemaRoot, + writer: ArrowStreamWriter, + softLimitBytesPerBatch: Long, + minDataCountForSample: Int, + softTimeoutMillsPurgeBatch: Long) { + + import ApplyInPandasWithStateWriter._ + + // We logically group the columns by family (data vs state) and initialize writer separately, + // since it's lot more easier and probably performant to write the row directly rather than + // projecting the row to match up with the overall schema. + // + // The number of data rows and state metadata rows can be different which could be problematic Review Comment: it is confusing what can be problematic. It we are maintaining separate batches for data and state then it is not clear why these batches need to have same number of rows. If we packing everything into the same batch, which we should explain why we are doing this, then it makes sense that some rows may contain only data and no state. -- 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. To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org