[
https://issues.apache.org/jira/browse/HADOOP-10794?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14067976#comment-14067976
]
Zhijie Shen commented on HADOOP-10794:
--------------------------------------
bq. Can you explain which components make this assumption, and what happens if
it's violated? So far the only example presented is statistics in YARN.
Not sure about the HDFS part, but the given example in the description is the
problem that happens to MR (see YARN-2251 for more details). For YARN side, at
least, NodeHealthStatus#lastHealthReportTime is acquired at NM, has been
reported to RM, and will further be sent to AM. If the clock is out of sync, RM
and AM will get a incorrect time about the NM health status. It may be more
other problematic cases or not. To see what really breaks, we'd better to setup
a cluster with asynchronous clocks.
> A hadoop cluster needs clock synchronization
> --------------------------------------------
>
> Key: HADOOP-10794
> URL: https://issues.apache.org/jira/browse/HADOOP-10794
> Project: Hadoop Common
> Issue Type: Bug
> Reporter: Zhijie Shen
>
> As a distributed system, a hadoop cluster wants the clock on all the
> participating hosts synchronized. Otherwise, some problems might happen. For
> example, in YARN-2251, due to the clock on the host for the task container
> falls behind that on the host of the AM container, the computed elapsed time
> (the diff between the timestamps produced on two hosts) becomes negative.
> In YARN-2251, we tried to mask the negative elapsed time. However, we should
> seek for a decent long term solution, such as providing mechanism to do and
> check clock synchronization.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
