[
https://issues.apache.org/jira/browse/HDFS-14792?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16937220#comment-16937220
]
Konstantin Shvachko commented on HDFS-14792:
--------------------------------------------
The main problem here is that Active NN cannot failover to an SBN, if it is in
SafeMode. Which can last for a long time as described in the description. The
solution here should be to stop counting blocks once the startup portion of
edits is processed by SBN. We should take a closer look at
{{isSafeModeTrackingBlocks()}}, which seems to be claiming more in JavaDoc than
it actually implements.
> [SBN read] StanbyNode does not come out of safemode while adding new blocks.
> ----------------------------------------------------------------------------
>
> Key: HDFS-14792
> URL: https://issues.apache.org/jira/browse/HDFS-14792
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: namenode
> Affects Versions: 2.10.0
> Reporter: Konstantin Shvachko
> Priority: Major
>
> During startup StandbyNode reports that it needs additional X blocks to reach
> the threshold 1.0000. Where X is changing up and down.
> This is because with fast tailing SBN adds new blocks from edits while DNs
> have not reported replicas yet. Being in SafeMode SBN counts new blocks
> towards the threshold and can stay in SafeMode for a long time.
> By design, the purpose of startup SafeMode is to disallow modifications of
> the namespace and blocks map until all DN replicas are reported.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
