[
https://issues.apache.org/jira/browse/HDFS-8326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14528775#comment-14528775
]
Andrew Wang commented on HDFS-8326:
-----------------------------------
Hi Misty, a few review comments:
* I believe the config key is "txns" not "txn"
* Would be good to put units for the period, which I believe is seconds.
* While we're in this doc, let's remove the "This key metadata item..."
sentence, since it really dates this doc and is unnecessary.
* The bit about a consistency tradeoff is not quite right. We write to the edit
log because it's very efficient, more efficient than trying to update the
fsimage in place. We need to checkpoint to prevent an unbounded number of
edits, which is bad because it fills up disks and also delays NN startup time
since the edits need to be applied to the base fsimage. You might consider
taking some content from the blog post I wrote:
http://blog.cloudera.com/blog/2014/03/a-guide-to-checkpointing-in-hadoop/
> Documentation about when checkpoints are run is out of date
> -----------------------------------------------------------
>
> Key: HDFS-8326
> URL: https://issues.apache.org/jira/browse/HDFS-8326
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: documentation
> Affects Versions: 2.3.0
> Reporter: Misty Stanley-Jones
> Assignee: Misty Stanley-Jones
> Attachments: HDFS-8326.patch
>
>
> Apparently checkpointing by interval or transaction size are both supported
> in at least HDFS 2.3, but the documentation does not reflect this.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
