[
https://issues.apache.org/jira/browse/HADOOP-12409?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14744990#comment-14744990
]
Xianyin Xin commented on HADOOP-12409:
--------------------------------------
Hi [[email protected]], i have two questions,
# in fact i don't understand why create and link the same jiras in MAPREDUCE
and YARN. Will they finally resolve it as duplicate?
# {quote}
Looking at uses of {{Clock}}, I see a bigger issue, which is that there are
lots of places which use Clock to time sleep intervals, e.g.
{{org.apache.hadoop.yarn.server.nodemanager.containermanager.linux.resources.CGroupsHandlerImpl}}.
This is dangerous as System time is not monotonic and can go backwards if NTP
tells it to.
{quote}
How about adding a new Clock (where we can wrap System.nanoTime() to avoid NTP
problem) which is used uniquely for time sleep intervals? Do you have some
considerations on this?
> Move org.apache.hadoop.yarn.util.Clock and relative implementations to hadoop
> Common
> ------------------------------------------------------------------------------------
>
> Key: HADOOP-12409
> URL: https://issues.apache.org/jira/browse/HADOOP-12409
> Project: Hadoop Common
> Issue Type: Improvement
> Components: util
> Affects Versions: 3.0.0
> Reporter: Xianyin Xin
> Assignee: Xianyin Xin
> Priority: Minor
> Attachments: Hadoop-12409.001.patch, Hadoop-12409.002.patch
>
>
> It is widely used by MR and YARN.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
