[
https://issues.apache.org/jira/browse/YARN-3304?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14352549#comment-14352549
]
Karthik Kambatla commented on YARN-3304:
----------------------------------------
Sample output when logging this metric. Note that, by default, these metrics
are published when a container completes. It is a little odd for the CPU usage
to be negative. Also, it is unavailable primarily when the container has not
started yet or enough time hasn't elapsed since start; showing zero for these
cases seems reasonable. What do others think?
{noformat}
kkambatl@pdsl24:~$ tail -f /tmp/nm-container-usage.out
1425872341126
container.ContainerResource_container_1425871888831_0001_01_000013:
ContainerResource=container_1425871888831_0001_01_000013, Context=container,
ContainerPid=8874, Hostname=pdsl24.cs.purdue.edu, PMemUsageMBsNumUsage=299,
PMemUsageMBsAvgMBs=161.25083612040132, PMemUsageMBsStdevMBs=15.650566309518153,
PMemUsageMBsIMinMBs=45.0, PMemUsageMBsIMaxMBs=171.0, PMemUsageMBsMinMBs=45.0,
PMemUsageMBsMaxMBs=171.0, PCpuUsagePercentNumUsage=299,
PCpuUsagePercentAvgPercents=2.0535117056856196,
PCpuUsagePercentStdevPercents=16.322375613869188,
PCpuUsagePercentIMinPercents=-1.0, PCpuUsagePercentIMaxPercents=194.0,
PCpuUsagePercentMinPercents=-1.0, PCpuUsagePercentMaxPercents=194.0,
MilliVcoreUsageNumUsage=299, MilliVcoreUsageAvgMilliVcores=24.40802675585284,
MilliVcoreUsageStdevMilliVcores=163.1979921241116,
MilliVcoreUsageIMinMilliVcores=-10.0, MilliVcoreUsageIMaxMilliVcores=1940.0,
MilliVcoreUsageMinMilliVcores=-10.0, MilliVcoreUsageMaxMilliVcores=1940.0,
pMemLimitMBs=1024, vMemLimitMBs=2150, vCoreLimit=1
1425872341127
container.ContainerResource_container_1425871888831_0001_01_000005:
ContainerResource=container_1425871888831_0001_01_000005, Context=container,
ContainerPid=8878, Hostname=pdsl24.cs.purdue.edu, PMemUsageMBsNumUsage=299,
PMemUsageMBsAvgMBs=153.38127090301018, PMemUsageMBsStdevMBs=14.116311560413372,
PMemUsageMBsIMinMBs=50.0, PMemUsageMBsIMaxMBs=162.0, PMemUsageMBsMinMBs=50.0,
PMemUsageMBsMaxMBs=162.0, PCpuUsagePercentNumUsage=299,
PCpuUsagePercentAvgPercents=2.070234113712376,
PCpuUsagePercentStdevPercents=16.222811543457478,
PCpuUsagePercentIMinPercents=-1.0, PCpuUsagePercentIMaxPercents=178.0,
PCpuUsagePercentMinPercents=-1.0, PCpuUsagePercentMaxPercents=178.0,
MilliVcoreUsageNumUsage=299, MilliVcoreUsageAvgMilliVcores=24.53846153846158,
MilliVcoreUsageStdevMilliVcores=162.46689190542918,
MilliVcoreUsageIMinMilliVcores=-10.0, MilliVcoreUsageIMaxMilliVcores=1785.0,
MilliVcoreUsageMinMilliVcores=-10.0, MilliVcoreUsageMaxMilliVcores=1785.0,
pMemLimitMBs=1024, vMemLimitMBs=2150, vCoreLimit=1
{noformat}
> Default return values for resources in ResourceCalculatorProcessTree are
> inconsistent in case where functionality is not available.
> -----------------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-3304
> URL: https://issues.apache.org/jira/browse/YARN-3304
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Reporter: Junping Du
> Assignee: Karthik Kambatla
> Priority: Blocker
>
> Per discussions in YARN-3296, getCpuUsagePercent() will return -1 for
> unavailable case while other resource metrics are return 0 in the same case
> which sounds inconsistent.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
