[
https://issues.apache.org/jira/browse/HBASE-6446?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13421656#comment-13421656
]
Jean-Daniel Cryans commented on HBASE-6446:
-------------------------------------------
@Ted
It wouldn't, the EOF comes out of the RS's HLog.
@terry
You are right that we shouldn't be so verbose in the replication code, but I
don't think your patch is the right way to do it. Checking the length adds
another NN hit (we already do many of them) and it doesn't really tell us
anything useful. Best case the log is still empty so we don't print a message,
worst case something was added so now we print that we got an EOF after
learning that it's not empty anymore.
In fact, there are other reasons to get EOF. For example, if you open the
reader at the last position you read from and nothing new was added then you
get the EOF right away. Meaning that, even with your patch, since the length
isn't 0, the exception would always be thrown plus you'll hit the NN more than
without it.
IMO we could skip printing the stack trace and have a more meaningful message.
> Replication source will throw EOF exception when hlog size is 0
> ---------------------------------------------------------------
>
> Key: HBASE-6446
> URL: https://issues.apache.org/jira/browse/HBASE-6446
> Project: HBase
> Issue Type: Bug
> Components: replication
> Affects Versions: 0.94.0
> Reporter: terry zhang
> Attachments: hbase-6446.patch
>
>
> when master cluster startup new hlog which size is 0 will be created. if we
> start replication, replication source will print many EOF exception when
> openreader. I think we need to ignore this case and do not print so many
> exception warning log .
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
