diff options
author | Josh Rosen <joshrosen@databricks.com> | 2016-11-07 16:14:19 -0800 |
---|---|---|
committer | Josh Rosen <joshrosen@databricks.com> | 2016-11-07 16:14:19 -0800 |
commit | 3a710b94b0c853a2dd4c40dca446ecde4e7be959 (patch) | |
tree | 2714991c21133436d28bc726fc55078d7aa98f92 /core/src | |
parent | 19cf208063f035d793d2306295a251a9af7e32f6 (diff) | |
download | spark-3a710b94b0c853a2dd4c40dca446ecde4e7be959.tar.gz spark-3a710b94b0c853a2dd4c40dca446ecde4e7be959.tar.bz2 spark-3a710b94b0c853a2dd4c40dca446ecde4e7be959.zip |
[SPARK-18236] Reduce duplicate objects in Spark UI and HistoryServer
## What changes were proposed in this pull request?
When profiling heap dumps from the HistoryServer and live Spark web UIs, I found a large amount of memory being wasted on duplicated objects and strings. This patch's changes remove most of this duplication, resulting in over 40% memory savings for some benchmarks.
- **Task metrics** (6441f0624dfcda9c7193a64bfb416a145b5aabdf): previously, every `TaskUIData` object would have its own instances of `InputMetricsUIData`, `OutputMetricsUIData`, `ShuffleReadMetrics`, and `ShuffleWriteMetrics`, but for many tasks these metrics are irrelevant because they're all zero. This patch changes how we construct these metrics in order to re-use a single immutable "empty" value for the cases where these metrics are empty.
- **TaskInfo.accumulables** (ade86db901127bf13c0e0bdc3f09c933a093bb76): Previously, every `TaskInfo` object had its own empty `ListBuffer` for holding updates from named accumulators. Tasks which didn't use named accumulators still paid for the cost of allocating and storing this empty buffer. To avoid this overhead, I changed the `val` with a mutable buffer into a `var` which holds an immutable Scala list, allowing tasks which do not have named accumulator updates to share the same singleton `Nil` object.
- **String.intern() in JSONProtocol** (7e05630e9a78c455db8c8c499f0590c864624e05): in the HistoryServer, executor hostnames and ids are deserialized from JSON, leading to massive duplication of these string objects. By calling `String.intern()` on the deserialized values we can remove all of this duplication. Since Spark now requires Java 7+ we don't have to worry about string interning exhausting the permgen (see http://java-performance.info/string-intern-in-java-6-7-8/).
## How was this patch tested?
I ran
```
sc.parallelize(1 to 100000, 100000).count()
```
in `spark-shell` with event logging enabled, then loaded that event log in the HistoryServer, performed a full GC, and took a heap dump. According to YourKit, the changes in this patch reduced memory consumption by roughly 28 megabytes (or 770k Java objects):
![image](https://cloud.githubusercontent.com/assets/50748/19953276/4f3a28aa-a129-11e6-93df-d7fa91396f66.png)
Here's a table illustrating the drop in objects due to deduplication (the drop is <100k for some objects because some events were dropped from the listener bus; this is a separate, existing bug that I'll address separately after CPU-profiling):
![image](https://cloud.githubusercontent.com/assets/50748/19953290/6a271290-a129-11e6-93ad-b825f1448886.png)
Author: Josh Rosen <joshrosen@databricks.com>
Closes #15743 from JoshRosen/spark-ui-memory-usage.
Diffstat (limited to 'core/src')
6 files changed, 79 insertions, 36 deletions
diff --git a/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala b/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala index f2517401cb..7fde34d897 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala @@ -1089,7 +1089,8 @@ class DAGScheduler( // To avoid UI cruft, ignore cases where value wasn't updated if (acc.name.isDefined && !updates.isZero) { stage.latestInfo.accumulables(id) = acc.toInfo(None, Some(acc.value)) - event.taskInfo.accumulables += acc.toInfo(Some(updates.value), Some(acc.value)) + event.taskInfo.setAccumulables( + acc.toInfo(Some(updates.value), Some(acc.value)) +: event.taskInfo.accumulables) } } } catch { diff --git a/core/src/main/scala/org/apache/spark/scheduler/TaskInfo.scala b/core/src/main/scala/org/apache/spark/scheduler/TaskInfo.scala index eeb7963c9e..59680139e7 100644 --- a/core/src/main/scala/org/apache/spark/scheduler/TaskInfo.scala +++ b/core/src/main/scala/org/apache/spark/scheduler/TaskInfo.scala @@ -17,8 +17,6 @@ package org.apache.spark.scheduler -import scala.collection.mutable.ListBuffer - import org.apache.spark.TaskState import org.apache.spark.TaskState.TaskState import org.apache.spark.annotation.DeveloperApi @@ -54,7 +52,13 @@ class TaskInfo( * accumulable to be updated multiple times in a single task or for two accumulables with the * same name but different IDs to exist in a task. */ - val accumulables = ListBuffer[AccumulableInfo]() + def accumulables: Seq[AccumulableInfo] = _accumulables + + private[this] var _accumulables: Seq[AccumulableInfo] = Nil + + private[spark] def setAccumulables(newAccumulables: Seq[AccumulableInfo]): Unit = { + _accumulables = newAccumulables + } /** * The time when the task has completed successfully (including the time to remotely fetch diff --git a/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala b/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala index f4a04609c4..9ce8542f02 100644 --- a/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala +++ b/core/src/main/scala/org/apache/spark/ui/jobs/UIData.scala @@ -21,7 +21,7 @@ import scala.collection.mutable import scala.collection.mutable.{HashMap, LinkedHashMap} import org.apache.spark.JobExecutionStatus -import org.apache.spark.executor.{ShuffleReadMetrics, ShuffleWriteMetrics, TaskMetrics} +import org.apache.spark.executor._ import org.apache.spark.scheduler.{AccumulableInfo, TaskInfo} import org.apache.spark.util.AccumulatorContext import org.apache.spark.util.collection.OpenHashSet @@ -147,9 +147,8 @@ private[spark] object UIData { memoryBytesSpilled = m.memoryBytesSpilled, diskBytesSpilled = m.diskBytesSpilled, peakExecutionMemory = m.peakExecutionMemory, - inputMetrics = InputMetricsUIData(m.inputMetrics.bytesRead, m.inputMetrics.recordsRead), - outputMetrics = - OutputMetricsUIData(m.outputMetrics.bytesWritten, m.outputMetrics.recordsWritten), + inputMetrics = InputMetricsUIData(m.inputMetrics), + outputMetrics = OutputMetricsUIData(m.outputMetrics), shuffleReadMetrics = ShuffleReadMetricsUIData(m.shuffleReadMetrics), shuffleWriteMetrics = ShuffleWriteMetricsUIData(m.shuffleWriteMetrics)) } @@ -171,9 +170,9 @@ private[spark] object UIData { speculative = taskInfo.speculative ) newTaskInfo.gettingResultTime = taskInfo.gettingResultTime - newTaskInfo.accumulables ++= taskInfo.accumulables.filter { + newTaskInfo.setAccumulables(taskInfo.accumulables.filter { accum => !accum.internal && accum.metadata != Some(AccumulatorContext.SQL_ACCUM_IDENTIFIER) - } + }) newTaskInfo.finishTime = taskInfo.finishTime newTaskInfo.failed = taskInfo.failed newTaskInfo @@ -197,8 +196,32 @@ private[spark] object UIData { shuffleWriteMetrics: ShuffleWriteMetricsUIData) case class InputMetricsUIData(bytesRead: Long, recordsRead: Long) + object InputMetricsUIData { + def apply(metrics: InputMetrics): InputMetricsUIData = { + if (metrics.bytesRead == 0 && metrics.recordsRead == 0) { + EMPTY + } else { + new InputMetricsUIData( + bytesRead = metrics.bytesRead, + recordsRead = metrics.recordsRead) + } + } + private val EMPTY = InputMetricsUIData(0, 0) + } case class OutputMetricsUIData(bytesWritten: Long, recordsWritten: Long) + object OutputMetricsUIData { + def apply(metrics: OutputMetrics): OutputMetricsUIData = { + if (metrics.bytesWritten == 0 && metrics.recordsWritten == 0) { + EMPTY + } else { + new OutputMetricsUIData( + bytesWritten = metrics.bytesWritten, + recordsWritten = metrics.recordsWritten) + } + } + private val EMPTY = OutputMetricsUIData(0, 0) + } case class ShuffleReadMetricsUIData( remoteBlocksFetched: Long, @@ -212,17 +235,30 @@ private[spark] object UIData { object ShuffleReadMetricsUIData { def apply(metrics: ShuffleReadMetrics): ShuffleReadMetricsUIData = { - new ShuffleReadMetricsUIData( - remoteBlocksFetched = metrics.remoteBlocksFetched, - localBlocksFetched = metrics.localBlocksFetched, - remoteBytesRead = metrics.remoteBytesRead, - localBytesRead = metrics.localBytesRead, - fetchWaitTime = metrics.fetchWaitTime, - recordsRead = metrics.recordsRead, - totalBytesRead = metrics.totalBytesRead, - totalBlocksFetched = metrics.totalBlocksFetched - ) + if ( + metrics.remoteBlocksFetched == 0 && + metrics.localBlocksFetched == 0 && + metrics.remoteBytesRead == 0 && + metrics.localBytesRead == 0 && + metrics.fetchWaitTime == 0 && + metrics.recordsRead == 0 && + metrics.totalBytesRead == 0 && + metrics.totalBlocksFetched == 0) { + EMPTY + } else { + new ShuffleReadMetricsUIData( + remoteBlocksFetched = metrics.remoteBlocksFetched, + localBlocksFetched = metrics.localBlocksFetched, + remoteBytesRead = metrics.remoteBytesRead, + localBytesRead = metrics.localBytesRead, + fetchWaitTime = metrics.fetchWaitTime, + recordsRead = metrics.recordsRead, + totalBytesRead = metrics.totalBytesRead, + totalBlocksFetched = metrics.totalBlocksFetched + ) + } } + private val EMPTY = ShuffleReadMetricsUIData(0, 0, 0, 0, 0, 0, 0, 0) } case class ShuffleWriteMetricsUIData( @@ -232,12 +268,17 @@ private[spark] object UIData { object ShuffleWriteMetricsUIData { def apply(metrics: ShuffleWriteMetrics): ShuffleWriteMetricsUIData = { - new ShuffleWriteMetricsUIData( - bytesWritten = metrics.bytesWritten, - recordsWritten = metrics.recordsWritten, - writeTime = metrics.writeTime - ) + if (metrics.bytesWritten == 0 && metrics.recordsWritten == 0 && metrics.writeTime == 0) { + EMPTY + } else { + new ShuffleWriteMetricsUIData( + bytesWritten = metrics.bytesWritten, + recordsWritten = metrics.recordsWritten, + writeTime = metrics.writeTime + ) + } } + private val EMPTY = ShuffleWriteMetricsUIData(0, 0, 0) } } diff --git a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala index 6593aab33f..4b4d2d10cb 100644 --- a/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala +++ b/core/src/main/scala/org/apache/spark/util/JsonProtocol.scala @@ -702,8 +702,8 @@ private[spark] object JsonProtocol { val index = (json \ "Index").extract[Int] val attempt = Utils.jsonOption(json \ "Attempt").map(_.extract[Int]).getOrElse(1) val launchTime = (json \ "Launch Time").extract[Long] - val executorId = (json \ "Executor ID").extract[String] - val host = (json \ "Host").extract[String] + val executorId = (json \ "Executor ID").extract[String].intern() + val host = (json \ "Host").extract[String].intern() val taskLocality = TaskLocality.withName((json \ "Locality").extract[String]) val speculative = Utils.jsonOption(json \ "Speculative").exists(_.extract[Boolean]) val gettingResultTime = (json \ "Getting Result Time").extract[Long] @@ -721,7 +721,7 @@ private[spark] object JsonProtocol { taskInfo.finishTime = finishTime taskInfo.failed = failed taskInfo.killed = killed - accumulables.foreach { taskInfo.accumulables += _ } + taskInfo.setAccumulables(accumulables) taskInfo } @@ -903,8 +903,8 @@ private[spark] object JsonProtocol { if (json == JNothing) { return null } - val executorId = (json \ "Executor ID").extract[String] - val host = (json \ "Host").extract[String] + val executorId = (json \ "Executor ID").extract[String].intern() + val host = (json \ "Host").extract[String].intern() val port = (json \ "Port").extract[Int] BlockManagerId(executorId, host, port) } diff --git a/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala b/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala index 8418fa74d2..da853f1be8 100644 --- a/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala +++ b/core/src/test/scala/org/apache/spark/ui/jobs/JobProgressListenerSuite.scala @@ -403,7 +403,7 @@ class JobProgressListenerSuite extends SparkFunSuite with LocalSparkContext with internal = false, countFailedValues = false, metadata = None) - taskInfo.accumulables ++= Seq(internalAccum, sqlAccum, userAccum) + taskInfo.setAccumulables(List(internalAccum, sqlAccum, userAccum)) val newTaskInfo = TaskUIData.dropInternalAndSQLAccumulables(taskInfo) assert(newTaskInfo.accumulables === Seq(userAccum)) diff --git a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala index d5146d70eb..85da79180f 100644 --- a/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala +++ b/core/src/test/scala/org/apache/spark/util/JsonProtocolSuite.scala @@ -788,11 +788,8 @@ private[spark] object JsonProtocolSuite extends Assertions { private def makeTaskInfo(a: Long, b: Int, c: Int, d: Long, speculative: Boolean) = { val taskInfo = new TaskInfo(a, b, c, d, "executor", "your kind sir", TaskLocality.NODE_LOCAL, speculative) - val (acc1, acc2, acc3) = - (makeAccumulableInfo(1), makeAccumulableInfo(2), makeAccumulableInfo(3, internal = true)) - taskInfo.accumulables += acc1 - taskInfo.accumulables += acc2 - taskInfo.accumulables += acc3 + taskInfo.setAccumulables( + List(makeAccumulableInfo(1), makeAccumulableInfo(2), makeAccumulableInfo(3, internal = true))) taskInfo } |