Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

SHS-NG M4.4: Port JobsTab and StageTab to the new backend. #10

Closed
wants to merge 1 commit into from
Closed

SHS-NG M4.4: Port JobsTab and StageTab to the new backend. #10

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion core/src/main/scala/org/apache/spark/SparkContext.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -455,7 +455,7 @@ class SparkContext(config: SparkConf) extends Logging {

_ui =
if (conf.getBoolean("spark.ui.enabled", true)) {
Some(SparkUI.create(Some(this), _stateStore, _conf, listenerBus, _env.securityManager,
Some(SparkUI.create(Some(this), _stateStore, _conf, _env.securityManager,
appName, "", startTime))
} else {
// For tests, do not enable the UI
Expand Down
3 changes: 1 addition & 2 deletions core/src/main/scala/org/apache/spark/deploy/history/FsHistoryProvider.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -326,8 +326,7 @@ private[history] class FsHistoryProvider(conf: SparkConf, clock: Clock)
}

val loadedUI = {
val ui = SparkUI.create(None, new AppStateStore(kvstore), conf, replayBus, secManager,
app.info.name,
val ui = SparkUI.create(None, new AppStateStore(kvstore), conf, secManager, app.info.name,
HistoryServer.getAttemptURI(appId, attempt.info.attemptId),
attempt.info.startTime.getTime())
LoadedAppUI(ui)
Expand Down
84 changes: 78 additions & 6 deletions core/src/main/scala/org/apache/spark/status/AppStateListener.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,7 @@ private[spark] class AppStateListener(
private val liveExecutors = new HashMap[String, LiveExecutor]()
private val liveTasks = new HashMap[Long, LiveTask]()
private val liveRDDs = new HashMap[Int, LiveRDD]()
private val pools = new HashMap[String, SchedulerPool]()

override def onApplicationStart(event: SparkListenerApplicationStart): Unit = {
assert(event.appId.isDefined, "Application without IDs are not supported.")
Expand Down Expand Up @@ -200,9 +201,8 @@ private[spark] class AppStateListener(
missingStages.map(_.numTasks).sum
}

val lastStageInfo = event.stageInfos.lastOption
val lastStageInfo = event.stageInfos.sortBy(_.stageId).lastOption
val lastStageName = lastStageInfo.map(_.name).getOrElse("(Unknown Stage Name)")

val jobGroup = Option(event.properties)
.flatMap { p => Option(p.getProperty(SparkContext.SPARK_JOB_GROUP_ID)) }

Expand All @@ -225,10 +225,41 @@ private[spark] class AppStateListener(
liveUpdate(stage)
}

// Create the graph data for all the job's stages.
event.stageInfos.foreach { stage =>
val graph = RDDOperationGraph.makeOperationGraph(stage, Int.MaxValue)
val uigraph = new RDDOperationGraphWrapper(
stage.stageId,
graph.edges,
graph.outgoingEdges,
graph.incomingEdges,
newRDDOperationCluster(graph.rootCluster))
kvstore.write(uigraph)
}
}

private def newRDDOperationCluster(cluster: RDDOperationCluster): RDDOperationClusterWrapper = {
new RDDOperationClusterWrapper(
cluster.id,
cluster.name,
cluster.childNodes,
cluster.childClusters.map(newRDDOperationCluster))
}

override def onJobEnd(event: SparkListenerJobEnd): Unit = {
liveJobs.remove(event.jobId).foreach { job =>
// Check if there are any pending stages that match this job; mark those as skipped.
job.stageIds.foreach { sid =>
val pending = liveStages.filter { case ((id, _), _) => id == sid }
pending.foreach { case (key, stage) =>
stage.status = v1.StageStatus.SKIPPED
job.skippedStages += stage.info.stageId
job.skippedTasks += stage.info.numTasks
liveStages.remove(key)
update(stage)
}
}

job.status = event.jobResult match {
case JobSucceeded => JobExecutionStatus.SUCCEEDED
case JobFailed(_) => JobExecutionStatus.FAILED
Expand All @@ -253,12 +284,20 @@ private[spark] class AppStateListener(
Option(p.getProperty("spark.scheduler.pool"))
}.getOrElse(SparkUI.DEFAULT_POOL_NAME)

stage.description = Option(event.properties).flatMap { p =>
Option(p.getProperty(SparkContext.SPARK_JOB_DESCRIPTION))
}

stage.jobs.foreach { job =>
job.completedStages = job.completedStages - event.stageInfo.stageId
job.activeStages += 1
liveUpdate(job)
}

val pool = pools.getOrElseUpdate(stage.schedulingPool, new SchedulerPool(stage.schedulingPool))
pool.stageIds = pool.stageIds + event.stageInfo.stageId
update(pool)

event.stageInfo.rddInfos.foreach { info =>
if (info.storageLevel.isValid) {
liveUpdate(liveRDDs.getOrElseUpdate(info.id, new LiveRDD(info)))
Expand Down Expand Up @@ -306,6 +345,8 @@ private[spark] class AppStateListener(
}

val metricsDelta = liveTasks.remove(event.taskInfo.taskId).map { task =>
task.info = event.taskInfo

val errorMessage = event.reason match {
case Success =>
None
Expand All @@ -322,11 +363,13 @@ private[spark] class AppStateListener(
delta
}.orNull

val (completedDelta, failedDelta) = event.reason match {
val (completedDelta, failedDelta, killedDelta) = event.reason match {
case Success =>
(1, 0)
(1, 0, 0)
case _: TaskKilled =>
(0, 0, 1)
case _ =>
(0, 1)
(0, 1, 0)
}

liveStages.get((event.stageId, event.stageAttemptId)).foreach { stage =>
Expand All @@ -336,19 +379,28 @@ private[spark] class AppStateListener(
stage.activeTasks -= 1
stage.completedTasks += completedDelta
stage.failedTasks += failedDelta
stage.killedTasks += killedDelta
if (killedDelta > 0) {
stage.killedSummary = killedTasksSummary(event.reason, stage.killedSummary)
}
liveUpdate(stage)

stage.jobs.foreach { job =>
job.activeTasks -= 1
job.completedTasks += completedDelta
job.failedTasks += failedDelta
job.killedTasks += killedDelta
if (killedDelta > 0) {
job.killedSummary = killedTasksSummary(event.reason, job.killedSummary)
}
liveUpdate(job)
}

val esummary = stage.executorSummary(event.taskInfo.executorId)
esummary.taskTime += event.taskInfo.duration
esummary.succeededTasks += completedDelta
esummary.failedTasks += failedDelta
esummary.killedTasks += killedDelta
if (metricsDelta != null) {
esummary.metrics.update(metricsDelta)
}
Expand Down Expand Up @@ -399,6 +451,11 @@ private[spark] class AppStateListener(
liveUpdate(job)
}

pools.get(stage.schedulingPool).foreach { pool =>
pool.stageIds = pool.stageIds - event.stageInfo.stageId
update(pool)
}

stage.executorSummaries.values.foreach(update)
update(stage)
}
Expand Down Expand Up @@ -456,11 +513,15 @@ private[spark] class AppStateListener(

/** Flush all live entities' data to the underlying store. */
def flush(): Unit = {
liveStages.values.foreach(update)
liveStages.values.foreach { stage =>
update(stage)
stage.executorSummaries.values.foreach(update)
}
liveJobs.values.foreach(update)
liveExecutors.values.foreach(update)
liveTasks.values.foreach(update)
liveRDDs.values.foreach(update)
pools.values.foreach(update)
}

private def updateRDDBlock(event: SparkListenerBlockUpdated, block: RDDBlockId): Unit = {
Expand Down Expand Up @@ -595,6 +656,17 @@ private[spark] class AppStateListener(
stage
}

private def killedTasksSummary(
reason: TaskEndReason,
oldSummary: Map[String, Int]): Map[String, Int] = {
reason match {
case k: TaskKilled =>
oldSummary.updated(k.reason, oldSummary.getOrElse(k.reason, 0) + 1)
case _ =>
oldSummary
}
}

private def update(entity: LiveEntity): Unit = {
entity.write(kvstore)
}
Expand Down
117 changes: 109 additions & 8 deletions core/src/main/scala/org/apache/spark/status/AppStateStore.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,21 +26,32 @@ import org.apache.spark.{JobExecutionStatus, SparkConf}
import org.apache.spark.kvstore.{InMemoryStore, KVStore}
import org.apache.spark.scheduler.{SparkListenerBus, SparkListenerEvent}
import org.apache.spark.status.api.v1
import org.apache.spark.ui.scope._
import org.apache.spark.util.{Distribution, Utils}

/**
* A wrapper around a KVStore that provides methods for accessing the API data stored within.
*/
private[spark] class AppStateStore(store: KVStore) {

def appInfo(): v1.ApplicationInfo = {
// There should be a single application info for a UIStore instance, so do no checks here.
val it = store.view(classOf[ApplicationInfoWrapper]).closeableIterator()
try {
it.next().info
} finally {
it.close()
}
}

def environmentInfo(): v1.ApplicationEnvironmentInfo = {
val klass = classOf[ApplicationEnvironmentInfoWrapper]
store.read(klass, klass.getName()).info
}

def jobsList(statuses: JList[JobExecutionStatus]): Seq[v1.JobData] = {
val it = store.view(classOf[JobDataWrapper]).sorted().asScala.map(_.info)
if (!statuses.isEmpty()) {
val it = store.view(classOf[JobDataWrapper]).sorted().reverse().asScala.map(_.info)
if (statuses != null && !statuses.isEmpty()) {
it.filter { job => statuses.contains(job.status) }.toSeq
} else {
it.toSeq
Expand All @@ -66,21 +77,35 @@ private[spark] class AppStateStore(store: KVStore) {
}

def stageList(statuses: JList[v1.StageStatus]): Seq[v1.StageData] = {
val it = store.view(classOf[StageDataWrapper]).sorted().asScala.map(_.info)
if (!statuses.isEmpty) {
val it = store.view(classOf[StageDataWrapper]).sorted().reverse().asScala.map(_.info)
if (statuses != null && !statuses.isEmpty()) {
it.filter { s => statuses.contains(s.status) }.toSeq
} else {
it.toSeq
}
}

def stageData(stageId: Int): Seq[v1.StageData] = {
def stageData(stageId: Int, details: Boolean = false): Seq[v1.StageData] = {
store.view(classOf[StageDataWrapper]).index("stageId").first(stageId).last(stageId)
.asScala.map(_.info).toSeq
.asScala.map { s =>
if (details) stageWithDetails(s.info) else s.info
}.toSeq
}

def stageAttempt(stageId: Int, stageAttemptId: Int): v1.StageData = {
store.read(classOf[StageDataWrapper], Array(stageId, stageAttemptId)).info
def lastStageAttempt(stageId: Int): v1.StageData = {
val it = store.view(classOf[StageDataWrapper]).index("stageId").reverse().first(stageId)
.closeableIterator()
try {
it.next().info
} finally {
it.close()
}
}

def stageAttempt(stageId: Int, stageAttemptId: Int, details: Boolean = false): v1.StageData = {
val stageKey = Array(stageId, stageAttemptId)
val stage = store.read(classOf[StageDataWrapper], stageKey).info
if (details) stageWithDetails(stage) else stage
}

def taskSummary(
Expand Down Expand Up @@ -179,6 +204,12 @@ private[spark] class AppStateStore(store: KVStore) {
)
}

def taskList(stageId: Int, stageAttemptId: Int, maxTasks: Int): Seq[v1.TaskData] = {
val stageKey = Array(stageId, stageAttemptId)
store.view(classOf[TaskDataWrapper]).index("stage").first(stageKey).last(stageKey).reverse()
.max(maxTasks).asScala.map(_.info).toSeq.reverse
}

def taskList(
stageId: Int,
stageAttemptId: Int,
Expand All @@ -199,6 +230,55 @@ private[spark] class AppStateStore(store: KVStore) {
indexed.skip(offset).max(length).asScala.map(_.info).toSeq
}

private def stageWithDetails(stage: v1.StageData): v1.StageData = {
// TODO: limit tasks returned.
val maxTasks = Int.MaxValue
val tasks = taskList(stage.stageId, stage.attemptId, maxTasks)
.map { t => (t.taskId, t) }
.toMap

val stageKey = Array(stage.stageId, stage.attemptId)
val execs = store.view(classOf[ExecutorStageSummaryWrapper]).index("stage").first(stageKey)
.last(stageKey).closeableIterator().asScala
.map { exec => (exec.executorId -> exec.info) }
.toMap

new v1.StageData(
stage.status,
stage.stageId,
stage.attemptId,
stage.numTasks,
stage.numActiveTasks,
stage.numCompleteTasks,
stage.numFailedTasks,
stage.numKilledTasks,
stage.executorRunTime,
stage.executorCpuTime,
stage.submissionTime,
stage.firstTaskLaunchedTime,
stage.completionTime,
stage.failureReason,
stage.inputBytes,
stage.inputRecords,
stage.outputBytes,
stage.outputRecords,
stage.shuffleReadBytes,
stage.shuffleReadRecords,
stage.shuffleWriteBytes,
stage.shuffleWriteRecords,
stage.memoryBytesSpilled,
stage.diskBytesSpilled,
stage.name,
stage.description,
stage.details,
stage.schedulingPool,
stage.rddIds,
stage.accumulatorUpdates,
Some(tasks),
Some(execs),
stage.killedTasksSummary)
}

def rddList(cachedOnly: Boolean = true): Seq[v1.RDDStorageInfo] = {
store.view(classOf[RDDStorageInfoWrapper]).sorted().asScala.map(_.info).filter { rdd =>
!cachedOnly || rdd.numCachedPartitions > 0
Expand All @@ -217,6 +297,27 @@ private[spark] class AppStateStore(store: KVStore) {
store.view(classOf[StreamBlockData]).asScala.toSeq
}

def operationGraphForStage(stageId: Int): RDDOperationGraph = {
store.read(classOf[RDDOperationGraphWrapper], stageId).toRDDOperationGraph()
}

def operationGraphForJob(jobId: Int): Seq[RDDOperationGraph] = {
val job = store.read(classOf[JobDataWrapper], jobId)
val stages = job.info.stageIds

stages.map { id =>
val g = store.read(classOf[RDDOperationGraphWrapper], id).toRDDOperationGraph()
if (job.skippedStages.contains(id) && !g.rootCluster.name.contains("skipped")) {
g.rootCluster.setName(g.rootCluster.name + " (skipped)")
}
g
}
}

def pool(name: String): PoolData = {
store.read(classOf[PoolData], name)
}

def close(): Unit = {
store.close()
}
Expand Down
Loading