[
https://issues.apache.org/jira/browse/HDFS-6243?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chris Nauroth updated HDFS-6243:
--------------------------------
Attachment: HDFS-6243.2.patch
Thanks for looking at the patch, Jing. Here is v2. I've added an assertion at
the end of the test that {{FSImage#getMostRecentCheckpointTxId}} returns 0 for
the former active. The only checkpoint issued in the test is being canceled,
so we expect this to remain 0.
I've also added one more change in {{StandbyCheckpointer}}. I noticed that the
background thread might still spend a fair amount of time blocked in {{wait}}
inside the throttler. We can make it break out of the throttle faster by
calling {{Future#cancel}} to interrupt the underlying thread.
> HA NameNode transition to active or shutdown may leave lingering image
> transfer thread.
> ---------------------------------------------------------------------------------------
>
> Key: HDFS-6243
> URL: https://issues.apache.org/jira/browse/HDFS-6243
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: ha, namenode
> Affects Versions: 3.0.0, 2.4.0
> Reporter: Chris Nauroth
> Assignee: Chris Nauroth
> Attachments: HDFS-6243.1.patch, HDFS-6243.2.patch
>
>
> With HDFS-4816, the standby uploads the new checkpoint to the active in a
> background thread. This thread may continue to execute after transitioning
> to active or shutting down. This is marginally wasteful of bandwidth if the
> image transfer is large, and it also breaks subsequent tests on Windows since
> the thread continues to hold a lock on a file descriptor in the storage
> directory.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
