[
https://issues.apache.org/jira/browse/HDFS-15562?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17228786#comment-17228786
]
Konstantin Shvachko commented on HDFS-15562:
--------------------------------------------
Hey guys, I think generally the checkpointer should persist until the
checkpoint completes and the image is transferred. With large image we see
transfers fail once in a while, so just ignoring image transfer failures isn't
right.
I understand that with multiple ObserverNodes some of them can be down.
We already have logic for ActiveNN and ObserverNodes to reject an image if they
already have one recent enough. So frequent checkpoints should not overwhelm
the active or the Observers.
We may add a logic for the Checkpointer to not re-create an image if it was
created recently. But this does not seem to be a big concern.
> StandbyCheckpointer will do checkpoint repeatedly while connecting
> observer/active namenode failed
> --------------------------------------------------------------------------------------------------
>
> Key: HDFS-15562
> URL: https://issues.apache.org/jira/browse/HDFS-15562
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: SunHao
> Assignee: Aihua Xu
> Priority: Major
> Labels: pull-request-available
> Attachments: HDFS-15562.patch
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> We find the standby namenode will do checkpoint over and over while
> connecting observer/active namenode failed.
> StandbyCheckpointer won't update “lastCheckpointTime” when upload new fsimage
> to the other namenode failed, so that the standby namenode will keep doing
> checkpoint repeatedly.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
