You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by gatorsmile <gi...@git.apache.org> on 2017/02/05 01:23:15 UTC
[GitHub] spark pull request #14702: [SPARK-15694] Implement ScriptTransformation in s...
Github user gatorsmile commented on a diff in the pull request:
https://github.com/apache/spark/pull/14702#discussion_r99479053
--- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/script/ScriptTransformationExec.scala ---
@@ -0,0 +1,337 @@
+/*
+ * 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.script
+
+import java.io._
+import java.nio.charset.StandardCharsets
+
+import scala.collection.JavaConverters._
+import scala.util.control.NonFatal
+
+import org.apache.hadoop.conf.Configuration
+
+import org.apache.spark.{SparkException, TaskContext}
+import org.apache.spark.internal.Logging
+import org.apache.spark.rdd.RDD
+import org.apache.spark.sql.SQLContext
+import org.apache.spark.sql.catalyst.{CatalystTypeConverters, InternalRow}
+import org.apache.spark.sql.catalyst.expressions._
+import org.apache.spark.sql.catalyst.plans.logical.ScriptInputOutputSchema
+import org.apache.spark.sql.catalyst.plans.physical.Partitioning
+import org.apache.spark.sql.catalyst.util.DateTimeUtils
+import org.apache.spark.sql.catalyst.util.DateTimeUtils.{SQLDate, SQLTimestamp}
+import org.apache.spark.sql.execution.{SparkPlan, UnaryExecNode}
+import org.apache.spark.sql.types.{DataType, DateType, StructType, TimestampType}
+import org.apache.spark.util.{CircularBuffer, RedirectThread, SerializableConfiguration, Utils}
+
+/**
+ * Transforms the input by forking and running the specified script.
+ *
+ * @param input the set of expression that should be passed to the script.
+ * @param script the command that should be executed.
+ * @param output the attributes that are produced by the script.
+ */
+private[sql]
+case class ScriptTransformationExec(
+ input: Seq[Expression],
+ script: String,
+ output: Seq[Attribute],
+ child: SparkPlan,
+ ioschema: ScriptTransformIOSchema)
+ extends UnaryExecNode with ScriptTransformBase {
+
+ override def producedAttributes: AttributeSet = outputSet -- inputSet
+
+ override def outputPartitioning: Partitioning = child.outputPartitioning
+
+ protected override def doExecute(): RDD[InternalRow] =
+ execute(sqlContext, child, schema)
+
+ override def processIterator(
+ inputIterator: Iterator[InternalRow],
+ hadoopConf: Configuration) : Iterator[InternalRow] = {
+
+ val (proc, inputStream, outputStream, stderrBuffer, outputProjection) =
+ init(input, script, child)
+
+ // This new thread will consume the ScriptTransformation's input rows and write them to the
+ // external process. That process's output will be read by this current thread.
+ val writerThread = new ScriptTransformationWriterThread(
+ inputIterator,
+ input.map(_.dataType),
+ outputProjection,
+ ioschema,
+ outputStream,
+ proc,
+ stderrBuffer,
+ TaskContext.get(),
+ hadoopConf
+ )
+
+ val reader = createReader(inputStream)
+
+ val outputIterator: Iterator[InternalRow] = new Iterator[InternalRow] {
+ var curLine: String = null
+ val mutableRow = new SpecificInternalRow(output.map(_.dataType))
+ val fieldDelimiter = ioschema.outputRowFormatMap("TOK_TABLEROWFORMATFIELD")
+
+ override def hasNext: Boolean = {
+ try {
+ if (curLine == null) {
+ curLine = reader.readLine()
+ if (curLine == null) {
+ checkFailureAndPropagate(writerThread.exception, null, proc, stderrBuffer)
+ return false
+ }
+ }
+ true
+ } catch {
+ case NonFatal(e) =>
+ // If this exception is due to abrupt / unclean termination of `proc`,
+ // then detect it and propagate a better exception message for end users
+ checkFailureAndPropagate(writerThread.exception, e, proc, stderrBuffer)
+
+ throw e
+ }
+ }
+
+ override def next(): InternalRow = {
+ if (!hasNext) {
+ throw new NoSuchElementException
+ }
+ val prevLine = curLine
+ curLine = reader.readLine()
+ if (!ioschema.isSchemaLess) {
+ new GenericInternalRow(
+ prevLine.split(fieldDelimiter).map(CatalystTypeConverters.convertToCatalyst))
+ } else {
+ new GenericInternalRow(
+ prevLine.split(fieldDelimiter, 2).map(CatalystTypeConverters.convertToCatalyst))
+ }
+ }
+ }
+
+ writerThread.start()
+ outputIterator
+ }
+}
+
+private[sql] trait ScriptTransformBase extends Serializable with Logging {
+
+ def init(
+ input: Seq[Expression],
+ script: String,
+ child: SparkPlan
+ ): (Process, InputStream, OutputStream, CircularBuffer, InterpretedProjection) = {
+
+ val cmd = List("/bin/bash", "-c", script)
+ val builder = new ProcessBuilder(cmd.asJava)
+
+ val proc = builder.start()
+ val inputStream = proc.getInputStream
+ val outputStream = proc.getOutputStream
+ val errorStream = proc.getErrorStream
+
+ // In order to avoid deadlocks, we need to consume the error output of the child process.
+ // To avoid issues caused by large error output, we use a circular buffer to limit the amount
+ // of error output that we retain. See SPARK-7862 for more discussion of the deadlock / hang
+ // that motivates this.
+ val stderrBuffer = new CircularBuffer(2048)
+ new RedirectThread(
+ errorStream,
+ stderrBuffer,
+ "Thread-ScriptTransformation-STDERR-Consumer").start()
+
+ val outputProjection = new InterpretedProjection(input, child.output)
+ (proc, inputStream, outputStream, stderrBuffer, outputProjection)
+ }
+
+ def execute(sqlContext: SQLContext,
+ child: SparkPlan,
+ schema: StructType): RDD[InternalRow] = {
+ val broadcastedHadoopConf =
+ new SerializableConfiguration(sqlContext.sessionState.newHadoopConf())
+
+ child.execute().mapPartitions { iter =>
+ if (iter.hasNext) {
+ val proj = UnsafeProjection.create(schema)
+ processIterator(iter, broadcastedHadoopConf.value).map(proj)
+ } else {
+ // If the input iterator has no rows then do not launch the external script.
+ Iterator.empty
+ }
+ }
+ }
+
+ def checkFailureAndPropagate(
+ writerException: Option[Throwable],
+ cause: Throwable = null,
+ proc: Process,
+ stderrBuffer: CircularBuffer): Unit = {
+ if (writerException.isDefined) {
+ throw writerException.get
+ }
+
+ // Checks if the proc is still alive (incase the command ran was bad)
+ // The ideal way to do this is to use Java 8's Process#isAlive()
+ // but it cannot be used because Spark still supports Java 7.
+ // Following is a workaround used to check if a process is alive in Java 7
+ // TODO: Once builds are switched to Java 8, this can be changed
+ try {
+ val exitCode = proc.exitValue()
+ if (exitCode != 0) {
+ logError(stderrBuffer.toString) // log the stderr circular buffer
+ throw new SparkException(s"Subprocess exited with status $exitCode. " +
+ s"Error: ${stderrBuffer.toString}", cause)
+ }
+ } catch {
+ case _: IllegalThreadStateException =>
+ // This means that the process is still alive. Move ahead
+ }
+ }
+
+ def createReader(inputStream: InputStream): BufferedReader =
+ new BufferedReader(new InputStreamReader(inputStream, StandardCharsets.UTF_8))
+
+ def processIterator(
+ inputIterator: Iterator[InternalRow],
+ hadoopConf: Configuration) : Iterator[InternalRow]
+}
+
+private[sql] class ScriptTransformationWriterThread(
+ iter: Iterator[InternalRow],
+ inputSchema: Seq[DataType],
+ outputProjection: Projection,
+ ioschema: ScriptTransformIOSchema,
+ outputStream: OutputStream,
+ proc: Process,
+ stderrBuffer: CircularBuffer,
+ taskContext: TaskContext,
+ conf: Configuration
+ ) extends Thread("Thread-ScriptTransformation-Feed") with Logging with Serializable {
+
+ setDaemon(true)
+
+ @volatile protected var _exception: Throwable = null
+
+ protected val lineDelimiter = ioschema.inputRowFormatMap("TOK_TABLEROWFORMATLINES")
+ protected val fieldDelimiter = ioschema.inputRowFormatMap("TOK_TABLEROWFORMATFIELD")
+
+ /** Contains the exception thrown while writing the parent iterator to the external process. */
+ def exception: Option[Throwable] = Option(_exception)
+
+ protected def init(): Unit = {
+ TaskContext.setTaskContext(taskContext)
+ }
+
+ protected def processRow(row: InternalRow, numColumns: Int): Unit = {
+ val data = if (numColumns == 0) {
+ lineDelimiter
+ } else {
+ val sb = new StringBuilder
+ sb.append(row.get(0, inputSchema.head))
+ var i = 1
+ while (i < numColumns) {
+ sb.append(fieldDelimiter)
+ val columnType = inputSchema(i)
+ val fieldValue = row.get(i, columnType)
+ val fieldStringValue = columnType match {
+ case _: DateType =>
+ DateTimeUtils.dateToString(fieldValue.asInstanceOf[SQLDate])
+ case _: TimestampType =>
+ DateTimeUtils.timestampToString(fieldValue.asInstanceOf[SQLTimestamp])
+ case _ =>
+ fieldValue.toString
+ }
+ sb.append(fieldStringValue)
+ i += 1
+ }
+ sb.append(lineDelimiter)
+ sb.toString()
+ }
+ outputStream.write(data.getBytes(StandardCharsets.UTF_8))
+ }
+
+ override def run(): Unit = Utils.logUncaughtExceptions {
+ init()
+
+ // We can't use Utils.tryWithSafeFinally here because we also need a `catch` block, so
+ // let's use a variable to record whether the `finally` block was hit due to an exception
+ var threwException: Boolean = true
+ val numColumns = inputSchema.length
+ try {
+ iter.map(outputProjection).foreach(row => processRow(row, numColumns))
+ threwException = false
+ } catch {
+ case t: Throwable =>
+ // An error occurred while writing input, so kill the child process. According to the
+ // Javadoc this call will not throw an exception:
+ _exception = t
+ proc.destroy()
+ throw t
+ } finally {
+ try {
+ Utils.tryLogNonFatalError(outputStream.close())
+ if (proc.waitFor() != 0) {
+ logError(stderrBuffer.toString) // log the stderr circular buffer
+ }
+ } catch {
+ case NonFatal(exceptionFromFinallyBlock) =>
+ if (!threwException) {
+ throw exceptionFromFinallyBlock
+ } else {
+ log.error("Exception in finally block", exceptionFromFinallyBlock)
+ }
+ }
+ }
+ }
+}
+
+private[sql]
+object ScriptTransformIOSchema {
+ def apply(input: ScriptInputOutputSchema): ScriptTransformIOSchema = {
+ new ScriptTransformIOSchema(
--- End diff --
```Scala
case class ScriptInputOutputSchema(
inputRowFormat: Seq[(String, String)],
outputRowFormat: Seq[(String, String)],
inputSerdeClass: Option[String],
outputSerdeClass: Option[String],
inputSerdeProps: Seq[(String, String)],
outputSerdeProps: Seq[(String, String)],
recordReaderClass: Option[String],
recordWriterClass: Option[String],
schemaLess: Boolean)
```
Except `inputRowFormat `, `outputRowFormat ` and `schemaLess `, we ignore all the other fields. I think we should not silently ignore them. For example, we do not respect any user-specified conf values of `hive.script.recordreader` and `hive.script.recordwriter`. Thus, could we issue an exception when users set them?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org