hadoop-common-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Erik Krogen (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HADOOP-15008) Metrics sinks may emit too frequently if multiple sink periods are configured
Date Wed, 01 Nov 2017 22:42:00 GMT

    [ https://issues.apache.org/jira/browse/HADOOP-15008?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16234887#comment-16234887
] 

Erik Krogen edited comment on HADOOP-15008 at 11/1/17 10:41 PM:
----------------------------------------------------------------

Attaching v000 patch. The bug comes from the fact that the {{period}} being passed into the
{{MetricsSinkAdapter}} is in seconds, but the {{logicalTime}} passed in is milliseconds. The
sink adapter gets offered metrics on every period of the {{MetricsSystemImpl}} (where the
period is the GCD of all of the sinks' periods), and filters based on {{logicalTime % period
== 0}}. The mismatch in units would cause this filter to pass more often than it should.

I fixed the issue (converted {{period}} to milliseconds), added a unit test, and changed the
field names in {{MetricsSinkAdapter}} to reflect that they are all in milliseconds to hopefully
prevent such future mistakes.


was (Author: xkrogen):
Attaching v000 patch. The bug comes from the fact that the {{period}} being passed into the
{{MetricsSinkAdapter}} is in seconds, but the {{logicalTime}} passed in is milliseconds. The
sink adapter gets offered metrics on every period of the {{MetricsSystemImpl}} (where the
period is the GCD of all of the sinks' periods), and filters based on {{logicalTime % period
== 0}}. The mismatch in units would cause this filter to pass more often than it should.

> Metrics sinks may emit too frequently if multiple sink periods are configured
> -----------------------------------------------------------------------------
>
>                 Key: HADOOP-15008
>                 URL: https://issues.apache.org/jira/browse/HADOOP-15008
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: metrics
>            Reporter: Erik Krogen
>            Assignee: Erik Krogen
>            Priority: Minor
>         Attachments: HADOOP-15008.000.patch
>
>
> If there are multiple metrics sink periods configured, depending on what those periods
are, some sinks may emit too frequently. For example with the following:
> {code:title=hadoop-metrics2.properties}
> namenode.sink.file10.class=org.apache.hadoop.metrics2.sink.FileSink
> namenode.sink.file5.class=org.apache.hadoop.metrics2.sink.FileSink
> namenode.sink.file10.filename=namenode-metrics_per10.out
> namenode.sink.file5.filename=namenode-metrics_per5.out
> namenode.sink.file10.period=10
> namenode.sink.file5.period=5
> {code}
> I get the following:
> {code}
> ± for f in namenode-metrics_per*.out; do echo "$f" && grep "metricssystem.MetricsSystem"
$f | awk '{last=curr; curr=$1} END { print curr-last }'; done
> namenode-metrics_per10.out
> 5000
> namenode-metrics_per5.out
> 5000
> {code}
> i.e., for both metrics files, each record is 5000 ms apart, even though one of the sinks
has been configured to emit at 10s intervals



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-issues-help@hadoop.apache.org


Mime
View raw message