[GitHub] spark pull request #22612: [SPARK-24958] Add executors' process tree total m...

[squito <gi...@git.apache.org>]

Github user squito commented on a diff in the pull request:

    https://github.com/apache/spark/pull/22612#discussion_r238366634
  
    --- Diff: core/src/main/scala/org/apache/spark/executor/ProcfsMetricsGetter.scala ---
    @@ -0,0 +1,223 @@
    +/*
    + * 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.executor
    +
    +import java.io._
    +import java.nio.charset.Charset
    +import java.nio.file.{Files, Paths}
    +import java.util.Locale
    +
    +import scala.collection.mutable
    +import scala.collection.mutable.ArrayBuffer
    +import scala.util.Try
    +
    +import org.apache.spark.{SparkEnv, SparkException}
    +import org.apache.spark.internal.{config, Logging}
    +import org.apache.spark.util.Utils
    +
    +
    +private[spark] case class ProcfsMetrics(
    +    jvmVmemTotal: Long,
    +    jvmRSSTotal: Long,
    +    pythonVmemTotal: Long,
    +    pythonRSSTotal: Long,
    +    otherVmemTotal: Long,
    +    otherRSSTotal: Long)
    +
    +// Some of the ideas here are taken from the ProcfsBasedProcessTree class in hadoop
    +// project.
    +private[spark] class ProcfsMetricsGetter(val procfsDir: String = "/proc/") extends Logging {
    +  val procfsStatFile = "stat"
    +  val testing = sys.env.contains("SPARK_TESTING") || sys.props.contains("spark.testing")
    +  val pageSize = computePageSize()
    +  var isAvailable: Boolean = isProcfsAvailable
    +  private val pid = computePid()
    +
    +  private lazy val isProcfsAvailable: Boolean = {
    +    if (testing) {
    +       true
    +    }
    +    else {
    +      val procDirExists = Try(Files.exists(Paths.get(procfsDir))).recover {
    +        case ioe: IOException =>
    +          logWarning("Exception checking for procfs dir", ioe)
    +          false
    +      }
    +      val shouldLogStageExecutorMetrics =
    +        SparkEnv.get.conf.get(config.EVENT_LOG_STAGE_EXECUTOR_METRICS)
    +      val shouldLogStageExecutorProcessTreeMetrics =
    +        SparkEnv.get.conf.get(config.EVENT_LOG_PROCESS_TREE_METRICS)
    +      procDirExists.get && shouldLogStageExecutorProcessTreeMetrics && shouldLogStageExecutorMetrics
    +    }
    +  }
    +
    +  private def computePid(): Int = {
    +    if (!isAvailable || testing) {
    +      return -1;
    +    }
    +    try {
    +      // This can be simplified in java9:
    +      // https://docs.oracle.com/javase/9/docs/api/java/lang/ProcessHandle.html
    +      val cmd = Array("bash", "-c", "echo $PPID")
    +      val out = Utils.executeAndGetOutput(cmd)
    +      val pid = Integer.parseInt(out.split("\n")(0))
    +      return pid;
    +    }
    +    catch {
    +      case e: SparkException =>
    +        logWarning("Exception when trying to compute process tree." +
    +          " As a result reporting of ProcessTree metrics is stopped", e)
    +        isAvailable = false
    +        -1
    +    }
    +  }
    +
    +  private def computePageSize(): Long = {
    +    if (testing) {
    +      return 4096;
    +    }
    +    try {
    +      val cmd = Array("getconf", "PAGESIZE")
    +      val out = Utils.executeAndGetOutput(cmd)
    +      Integer.parseInt(out.split("\n")(0))
    +    } catch {
    +      case e: Exception =>
    +        logWarning("Exception when trying to compute pagesize, as a" +
    +          " result reporting of ProcessTree metrics is stopped")
    +        isAvailable = false
    +        0
    +    }
    +  }
    +
    +  private def computeProcessTree(): Set[Int] = {
    +    if (!isAvailable || testing) {
    +      return Set()
    +    }
    +    var ptree: Set[Int] = Set()
    +    ptree += pid
    +    val queue = mutable.Queue.empty[Int]
    +    queue += pid
    +    while ( !queue.isEmpty ) {
    +      val p = queue.dequeue()
    +      val c = getChildPids(p)
    +      if (!c.isEmpty) {
    +        queue ++= c
    +        ptree ++= c.toSet
    +      }
    +    }
    +    ptree
    +  }
    +
    +  private def getChildPids(pid: Int): ArrayBuffer[Int] = {
    +    try {
    +      val builder = new ProcessBuilder("pgrep", "-P", pid.toString)
    +      val process = builder.start()
    +      val childPidsInInt = mutable.ArrayBuffer.empty[Int]
    +      def appendChildPid(s: String): Unit = {
    +        if (s != "") {
    +          logTrace("Found a child pid:" + s)
    +          childPidsInInt += Integer.parseInt(s)
    +        }
    +      }
    +      val stdoutThread = Utils.processStreamByLine("read stdout for pgrep",
    +        process.getInputStream, appendChildPid)
    +      val errorStringBuilder = new StringBuilder()
    +      val stdErrThread = Utils.processStreamByLine(
    +        "stderr for pgrep",
    +        process.getErrorStream,
    +        line => errorStringBuilder.append(line))
    +      val exitCode = process.waitFor()
    +      stdoutThread.join()
    +      stdErrThread.join()
    +      val errorString = errorStringBuilder.toString()
    +      // pgrep will have exit code of 1 if there are more than one child process
    +      // and it will have a exit code of 2 if there is no child process
    +      if (exitCode != 0 && exitCode > 2) {
    +        val cmd = builder.command().toArray.mkString(" ")
    +        logWarning(s"Process $cmd exited with code $exitCode and stderr: $errorString")
    +        throw new SparkException(s"Process $cmd exited with code $exitCode")
    +      }
    +      childPidsInInt
    +    } catch {
    +      case e: Exception =>
    +        logWarning("Exception when trying to compute process tree." +
    +          " As a result reporting of ProcessTree metrics is stopped.", e)
    +        isAvailable = false
    +        mutable.ArrayBuffer.empty[Int]
    +    }
    +  }
    +
    +  def addProcfsMetricsFromOneProcess(
    +      allMetrics: ProcfsMetrics,
    +      pid: Int): ProcfsMetrics = {
    +
    +    // The computation of RSS and Vmem are based on proc(5):
    +    // http://man7.org/linux/man-pages/man5/proc.5.html
    +    try {
    +      val pidDir = new File(procfsDir, pid.toString)
    +      Utils.tryWithResource(new InputStreamReader(
    +        new FileInputStream(
    +          new File(pidDir, procfsStatFile)), Charset.forName("UTF-8"))) { fReader =>
    +          val in = new BufferedReader(fReader)
    +          val procInfo = in.readLine
    +          val procInfoSplit = procInfo.split(" ")
    +          val vmem = procInfoSplit(22).toLong
    +          val rssMem = procInfoSplit(23).toLong * pageSize
    +          if (procInfoSplit(1).toLowerCase(Locale.US).contains("java")) {
    +            allMetrics.copy(
    +              jvmVmemTotal = allMetrics.jvmVmemTotal + vmem,
    +              jvmRSSTotal = allMetrics.jvmRSSTotal + (rssMem)
    +            )
    +          }
    +          else if (procInfoSplit(1).toLowerCase(Locale.US).contains("python")) {
    +            allMetrics.copy(
    +              pythonVmemTotal = allMetrics.pythonVmemTotal + vmem,
    +              pythonRSSTotal = allMetrics.pythonRSSTotal + (rssMem)
    +            )
    +          }
    +          else {
    +            allMetrics.copy(
    +              otherVmemTotal = allMetrics.otherVmemTotal + vmem,
    +              otherRSSTotal = allMetrics.otherRSSTotal + (rssMem)
    +            )
    +          }
    +      }
    +    } catch {
    +      case f: IOException =>
    +        logWarning("There was a problem with reading" +
    +          " the stat file of the process. ", f)
    +        ProcfsMetrics(0, 0, 0, 0, 0, 0)
    +    }
    +  }
    +
    +  private[spark] def computeAllMetrics(): ProcfsMetrics = {
    +    if (!isAvailable) {
    +      return ProcfsMetrics(0, 0, 0, 0, 0, 0)
    +    }
    +    val pids = computeProcessTree
    +    var allMetrics = ProcfsMetrics(0, 0, 0, 0, 0, 0)
    +    for (p <- pids) {
    +      allMetrics = addProcfsMetricsFromOneProcess(allMetrics, p)
    +    }
    +    allMetrics
    --- End diff --
    
    one weird thing about the error handling here -- if you get any error from `computeProcessTree` through all calls to `addProcsFsMetricsFromOneProcess`, then you'll set `isAvailable = false`, but you'll still return `allMetrics`.  Depending on when the error occurs, you might have some partially accumulated metrics in there.  I think if there is any error, you probably want to return empty metrics.
    
    Also you'll leave `isAvailable=false` forever after that -- is that OK?  I guess I don't really have strong feelings about this one, maybe its OK to give up forever for this executor.


---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org