[
https://issues.apache.org/jira/browse/HDFS-2288?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13103299#comment-13103299
]
Todd Lipcon commented on HDFS-2288:
-----------------------------------
Hi Konstantin. Thanks for taking a look at this. I didn't flesh out the
scenario quite enough:
- Writer is writing to a pipeline, alternating writes and hflush()
- A concurrent reader is reading from the replicas (just behind the writer).
Thus maybe it sees 30MB out of the 64MB block in progress.
- The writer and the three DNs crash/restart
- The reader continues trying to read, and is now unable to see the data they
previously saw, until lease recovery proceeds (many minutes)
This seems incorrect. I think the reason it's not mentioned in the design doc
is that the design didn't cover all the cases of concurrent readers behind
writers.
> Replicas awaiting recovery should return a full visible length
> --------------------------------------------------------------
>
> Key: HDFS-2288
> URL: https://issues.apache.org/jira/browse/HDFS-2288
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: data-node
> Affects Versions: 0.23.0
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Priority: Critical
> Fix For: 0.23.0
>
> Attachments: hdfs-2288.txt
>
>
> Currently, if the client calls getReplicaVisibleLength for a RWR, it returns
> a visible length of 0. This causes one of HBase's tests to fail, and I
> believe it's incorrect behavior.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
