[SPARK-24398] SQLAppStatusListener.aggregateMetrics() invocation is too slow

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 2.3.0
• Component/s: Web UI
• Assignee: Unassigned
• Reporter: furongbin
• Created: 26/May/18 08:22
• Updated: 16/Dec/18 22:31
• Resolved: 16/Dec/18 22:31

Description

RROR 18/05/24 14:08:25 AsyncEventQueue: Dropping event from queue appStatus. This likely means one of the listeners is too slow and cannot keep up with the rate at which tasks are being started by the scheduler. [dag-scheduler-event-loop]

WARN 18/05/24 16:39:54 AsyncEventQueue: Dropped 208205 events from appStatus since Thu May 24 16:30:57 CST 2018. [dag-scheduler-event-loop]

WARN 18/05/24 16:40:19 SQLAppStatusListener: agregateMetrics  cost=18.6s, metricIds.size=775, metrics.size=5292940, aggregatedMetrics=615 [pool-22-thread-1]

private def aggregateMetrics(exec: LiveExecutionData): Map[Long, String] = {

val metricIds = exec.metrics.map(_.accumulatorId).sorted

val metricTypes = exec.metrics.map

{ m => (m.accumulatorId, m.metricType) }

.toMap
val metrics = exec.stages.toSeq
    .flatMap { stageId => Option(stageMetrics.get(stageId)) }
    .flatMap(_.taskMetrics.values().asScala)
    .flatMap { metrics => metrics.ids.zip(metrics.values) }

val aggregatedMetrics = (metrics ++ exec.driverAccumUpdates.toSeq)
.filter { case (id, _) => metricIds.contains(id) }
.groupBy(_._1)
.map

{ case (id, values) =>         id -> SQLMetrics.stringValue(metricTypes(id), values.map(_._2).toSeq) }

if (exec.metricsValues != null)

{       exec.metricsValues }

else

{     aggregatedMetrics   }

}

Case:

when aggregatedMetrics.size=5000000, metrics.size=1000, then the execution(val aggregatedMetrics = (metrics ++ exec.driverAccumUpdates.toSeq)
.filter { case (id, _) => metricIds.contains(id) })  is very slow

 

Issue Links

• duplicates: SPARK-26003 Improve performance in SQLAppStatusListener
• links to: [Github] Pull Request #21438 (frb502)
• links to: GitHub Pull Request #21438