|
| 1 | +/* |
| 2 | + * Licensed to the Apache Software Foundation (ASF) under one or more |
| 3 | + * contributor license agreements. See the NOTICE file distributed with |
| 4 | + * this work for additional information regarding copyright ownership. |
| 5 | + * The ASF licenses this file to You under the Apache License, Version 2.0 |
| 6 | + * (the "License"); you may not use this file except in compliance with |
| 7 | + * the License. You may obtain a copy of the License at |
| 8 | + * |
| 9 | + * http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | + * |
| 11 | + * Unless required by applicable law or agreed to in writing, software |
| 12 | + * distributed under the License is distributed on an "AS IS" BASIS, |
| 13 | + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 14 | + * See the License for the specific language governing permissions and |
| 15 | + * limitations under the License. |
| 16 | + */ |
| 17 | + |
| 18 | +package org.apache.spark.sql.execution.streaming.sources |
| 19 | + |
| 20 | +import java.util |
| 21 | + |
| 22 | +import scala.collection.mutable.ArrayBuffer |
| 23 | + |
| 24 | +import org.apache.spark.{SparkEnv, SparkUnsupportedOperationException} |
| 25 | +import org.apache.spark.rpc.{RpcEndpointRef, RpcEnv, ThreadSafeRpcEndpoint} |
| 26 | +import org.apache.spark.sql.Row |
| 27 | +import org.apache.spark.sql.catalyst.InternalRow |
| 28 | +import org.apache.spark.sql.catalyst.encoders.ExpressionEncoder |
| 29 | +import org.apache.spark.sql.connector.catalog.{SupportsWrite, TableCapability} |
| 30 | +import org.apache.spark.sql.connector.write.{ |
| 31 | + LogicalWriteInfo, |
| 32 | + PhysicalWriteInfo, |
| 33 | + Write, |
| 34 | + WriteBuilder, |
| 35 | + WriterCommitMessage |
| 36 | +} |
| 37 | +import org.apache.spark.sql.connector.write.streaming.{StreamingDataWriterFactory, StreamingWrite} |
| 38 | +import org.apache.spark.sql.internal.connector.SupportsStreamingUpdateAsAppend |
| 39 | +import org.apache.spark.sql.types.StructType |
| 40 | + |
| 41 | +/** |
| 42 | + * A sink that stores the results in memory. This [[org.apache.spark.sql.execution.streaming.Sink]] |
| 43 | + * is primarily intended for use in unit tests and does not provide durability. |
| 44 | + * This is mostly copied from MemorySink, except that the data needs to be available not in |
| 45 | + * commit() but after each write. |
| 46 | + */ |
| 47 | +class ContinuousMemorySink |
| 48 | + extends MemorySink |
| 49 | + with SupportsWrite { |
| 50 | + |
| 51 | + private val batches = new ArrayBuffer[Row]() |
| 52 | + override def name(): String = "ContinuousMemorySink" |
| 53 | + |
| 54 | + override def schema(): StructType = StructType(Nil) |
| 55 | + |
| 56 | + override def capabilities(): util.Set[TableCapability] = { |
| 57 | + util.EnumSet.of(TableCapability.STREAMING_WRITE) |
| 58 | + } |
| 59 | + |
| 60 | + override def newWriteBuilder(info: LogicalWriteInfo): WriteBuilder = { |
| 61 | + new WriteBuilder with SupportsStreamingUpdateAsAppend { |
| 62 | + private val inputSchema: StructType = info.schema() |
| 63 | + |
| 64 | + override def build(): Write = { |
| 65 | + new ContinuousMemoryWrite(batches, inputSchema) |
| 66 | + } |
| 67 | + } |
| 68 | + } |
| 69 | + |
| 70 | + /** Returns all rows that are stored in this [[Sink]]. */ |
| 71 | + override def allData: Seq[Row] = { |
| 72 | + val batches = getBatches() |
| 73 | + batches.synchronized { |
| 74 | + batches.toSeq |
| 75 | + } |
| 76 | + } |
| 77 | + |
| 78 | + override def latestBatchId: Option[Long] = { |
| 79 | + None |
| 80 | + } |
| 81 | + |
| 82 | + override def latestBatchData: Seq[Row] = { |
| 83 | + throw new SparkUnsupportedOperationException( |
| 84 | + errorClass = "UNSUPPORTED_OPERATION_FOR_CONTINUOUS_MEMORY_SINK", |
| 85 | + messageParameters = Map("operation" -> "latestBatchData") |
| 86 | + ) |
| 87 | + } |
| 88 | + |
| 89 | + override def dataSinceBatch(sinceBatchId: Long): Seq[Row] = { |
| 90 | + throw new SparkUnsupportedOperationException( |
| 91 | + errorClass = "UNSUPPORTED_OPERATION_FOR_CONTINUOUS_MEMORY_SINK", |
| 92 | + messageParameters = Map("operation" -> "dataSinceBatch") |
| 93 | + ) |
| 94 | + } |
| 95 | + |
| 96 | + override def toDebugString: String = { |
| 97 | + s"${allData}" |
| 98 | + } |
| 99 | + |
| 100 | + override def write(batchId: Long, needTruncate: Boolean, newRows: Array[Row]): Unit = { |
| 101 | + throw new SparkUnsupportedOperationException( |
| 102 | + errorClass = "UNSUPPORTED_OPERATION_FOR_CONTINUOUS_MEMORY_SINK", |
| 103 | + messageParameters = Map("operation" -> "write") |
| 104 | + ) |
| 105 | + } |
| 106 | + |
| 107 | + override def clear(): Unit = synchronized { |
| 108 | + batches.clear() |
| 109 | + } |
| 110 | + |
| 111 | + private def getBatches(): ArrayBuffer[Row] = { |
| 112 | + batches |
| 113 | + } |
| 114 | + |
| 115 | + override def toString(): String = "ContinuousMemorySink" |
| 116 | +} |
| 117 | + |
| 118 | +class ContinuousMemoryWrite(batches: ArrayBuffer[Row], schema: StructType) extends Write { |
| 119 | + override def toStreaming: StreamingWrite = { |
| 120 | + new ContinuousMemoryStreamingWrite(batches, schema) |
| 121 | + } |
| 122 | +} |
| 123 | + |
| 124 | +/** |
| 125 | + * An RPC endpoint that receives rows and stores them to the ArrayBuffer in real-time. |
| 126 | + */ |
| 127 | +class MemoryRealTimeRpcEndpoint( |
| 128 | + override val rpcEnv: RpcEnv, |
| 129 | + schema: StructType, |
| 130 | + batches: ArrayBuffer[Row] |
| 131 | +) extends ThreadSafeRpcEndpoint { |
| 132 | + private val encoder = ExpressionEncoder(schema).resolveAndBind().createDeserializer() |
| 133 | + |
| 134 | + override def receive: PartialFunction[Any, Unit] = { |
| 135 | + case rows: Array[InternalRow] => |
| 136 | + // synchronized block is optional here since ThreadSafeRpcEndpoint already, just to be safe |
| 137 | + batches.synchronized { |
| 138 | + rows.foreach { row => |
| 139 | + batches += encoder(row) |
| 140 | + } |
| 141 | + } |
| 142 | + } |
| 143 | +} |
| 144 | + |
| 145 | +class ContinuousMemoryStreamingWrite(val batches: ArrayBuffer[Row], schema: StructType) |
| 146 | + extends StreamingWrite { |
| 147 | + |
| 148 | + private val memoryEndpoint = |
| 149 | + new MemoryRealTimeRpcEndpoint( |
| 150 | + SparkEnv.get.rpcEnv, |
| 151 | + schema, |
| 152 | + batches |
| 153 | + ) |
| 154 | + @volatile private var endpointRef: RpcEndpointRef = _ |
| 155 | + |
| 156 | + override def createStreamingWriterFactory(info: PhysicalWriteInfo): StreamingDataWriterFactory = { |
| 157 | + val endpointName = s"MemoryRealTimeRpcEndpoint-${java.util.UUID.randomUUID()}" |
| 158 | + endpointRef = memoryEndpoint.rpcEnv.setupEndpoint(endpointName, memoryEndpoint) |
| 159 | + RealTimeRowWriterFactory(endpointName, endpointRef.address) |
| 160 | + } |
| 161 | + |
| 162 | + override def useCommitCoordinator(): Boolean = false |
| 163 | + |
| 164 | + override def commit(epochId: Long, messages: Array[WriterCommitMessage]): Unit = { |
| 165 | + // We don't need to commit anything in this case, as the rows have already been printed |
| 166 | + if (endpointRef != null) { |
| 167 | + memoryEndpoint.rpcEnv.stop(endpointRef) |
| 168 | + } |
| 169 | + } |
| 170 | + |
| 171 | + override def abort(epochId: Long, messages: Array[WriterCommitMessage]): Unit = { |
| 172 | + if (endpointRef != null) { |
| 173 | + memoryEndpoint.rpcEnv.stop(endpointRef) |
| 174 | + } |
| 175 | + } |
| 176 | +} |
0 commit comments