[HADOOP-14166] Reset the DecayRpcScheduler AvgResponseTime metric to zero when queue is not used

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.9.0, 3.0.0-alpha4, 2.8.2
• Component/s: common
• Hadoop Flags: Reviewed
• Assignee: Surendra Singh Lilhore
• Reporter: Surendra Singh Lilhore
• Created: 09/Mar/17 05:59
• Updated: 24/Apr/18 20:50
• Resolved: 24/May/17 17:31

Description

 "name" : "Hadoop:service=NameNode,name=DecayRpcSchedulerMetrics2.ipc.8020",
    "modelerType" : "DecayRpcSchedulerMetrics2.ipc.8020",
    "tag.Context" : "ipc.8020",
    "tag.Hostname" : "host1",
    "DecayedCallVolume" : 3,
    "UniqueCallers" : 1,
    "Caller(root).Volume" : 266,
    "Caller(root).Priority" : 3,
    "Priority.0.AvgResponseTime" : 6.151201023385511E-5,
    "Priority.1.AvgResponseTime" : 0.0,
    "Priority.2.AvgResponseTime" : 0.0,
    "Priority.3.AvgResponseTime" : 1.184686336544601,
    "Priority.0.CompletedCallVolume" : 0,
    "Priority.1.CompletedCallVolume" : 0,
    "Priority.2.CompletedCallVolume" : 0,
    "Priority.3.CompletedCallVolume" : 2,
    "CallVolume" : 266

"Priority.0.AvgResponseTime" is always "6.151201023385511E-5" even the queue is not used for long time.

      if (lastAvg > PRECISION || averageResponseTime > PRECISION) {
        if (enableDecay) {
          final double decayed = decayFactor * lastAvg + averageResponseTime;
          LOG.info("Decayed "  + i + " time " +   decayed);
          responseTimeAvgInLastWindow.set(i, decayed);
        } else {
          responseTimeAvgInLastWindow.set(i, averageResponseTime);
        }
      }

we should reset it to zero when above condition is false.

Issue Links

• is related to: HADOOP-12916 Allow RPC scheduler/callqueue backoff using response times