[
https://issues.apache.org/jira/browse/HDFS-2866?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13197515#comment-13197515
]
Aaron T. Myers commented on HDFS-2866:
--------------------------------------
bq. don't we need a shared copy of fsimage?
Nope, and configuring it as such would be an error.
bq. If it is kept in local dirs, there would be a problem with two NNs
formatting unless something like metadata copy is done from local dirs of NN1
to NN2.
The intention is that when bootstrapping an HA cluster, one must format one NN
and then copy the initially-empty metadata to the other NN. This is similar to
bootstrapping other HA systems with persistent data, e.g. MySQL. See this JIRA
which aims to improve this bootstrapping process: HDFS-2731
bq. Also, working with local image of fsimage might make it difficult to catch
issues if there is divergence.
How so? The standby NN performs checkpoints and thus will periodically download
fsimage files from the active NN. Also note that the edits and image files
include the transaction IDs they contain.
> Standby does not start up due to a gap in transaction id
> --------------------------------------------------------
>
> Key: HDFS-2866
> URL: https://issues.apache.org/jira/browse/HDFS-2866
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: ha
> Affects Versions: HA branch (HDFS-1623)
> Reporter: Hari Mankude
> Priority: Critical
>
> Standby notices a gap in the transaction id in the shared.edits directory.
> The transactions in dfs.edits.dir does not seem to have the gap. The gap
> happens during a failover.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
