[
https://issues.apache.org/jira/browse/HDFS-6029?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13915108#comment-13915108
]
Jing Zhao commented on HDFS-6029:
---------------------------------
The reason is that for non-HA setup, the rollingupgrade prepare step will
create a rollback image in NN, and SNN could not download it while
checkpointing (it can only recognize "fsimage"). I will provide a patch later.
> Secondary NN fails to checkpoint after -rollingUpgrade prepare
> --------------------------------------------------------------
>
> Key: HDFS-6029
> URL: https://issues.apache.org/jira/browse/HDFS-6029
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: datanode, ha, hdfs-client, namenode
> Affects Versions: HDFS-5535 (Rolling upgrades)
> Reporter: Haohui Mai
> Assignee: Jing Zhao
>
> If Secondary NN crashes right after calling {{dfsadmin -rollingUpgrade
> prepare}}, it fails to checkpoint because it fails to download the newest
> fsimage whose name is {{fsimage_rollback_txid}}.
> To reproduce this bug:
> # Run NN and SNN.
> # Run {{dfsadmin -rollingUpgrade prepare}}.
> # Shutdown SNN, clear all fsimage created by SNN.
> # Restart SNN, SNN tries to download the {{fsimage_txid}} instead of
> {{fsimage_rollback_txid}}, resulting the failure.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
