[
https://issues.apache.org/jira/browse/HDFS-5145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13819209#comment-13819209
]
Kihwal Lee commented on HDFS-5145:
----------------------------------
An immediate question would be whether it is okay to automatically revert a
decision to leave start-up safe mode. On one hand, we want the cluster to
become usable soon even if a small portion of data is temporarily unavailable.
But if the cluster loses a significant portion of data nodes or blocks after
reaching the threshold once, we might prefer it from leaving safe mode.
However, if a decision to leave safe mode is made and an action (calling
leave()) is taken immediately, this problem won't exist, since data
availability won't degrade while trying to leave safe mode.
Then the question becomes why do we have safe mode extension? What is the risk
of getting rid of it or making it 0?
> Revisit the needEnter() check in SafeModeInfo#canLeave()
> ---------------------------------------------------------
>
> Key: HDFS-5145
> URL: https://issues.apache.org/jira/browse/HDFS-5145
> Project: Hadoop HDFS
> Issue Type: Improvement
> Affects Versions: 3.0.0
> Reporter: Jing Zhao
> Assignee: Jing Zhao
> Priority: Minor
>
> The current code in SafeModeInfo#canLeave checks needEnter() again. This may
> be unnecessary as per discussion in HDFS-5140. Let's revisit
> SafeModeInfo#canLeave in this jira.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
