[
https://issues.apache.org/jira/browse/HBASE-10443?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13887698#comment-13887698
]
ramkrishna.s.vasudevan commented on HBASE-10443:
------------------------------------------------
After a long fight with this issue, we have nailed down this.
When the store file is being read, as part of reseek to a key if we find the if
we have jumped a given key we do a move previous. Incase of DBEs we have
already read this key and made an entry in the LRU dictionary for tags.
Now after moving to previous we once again read this kv and again will make a
duplicate entry in the KV in the next immediate index. Now there are two
entries.
Index 17 and 18 may have the same entry (for eg). Now another read tries to
check what value is in index 18 expecting Tag B he may get back Tag A.
Just tried out ruling out duplicate entries am able to overcome this bug.
So better to disable compress_tags for this version?
> Getting Index Out of bound exception when using Compress_tags (to verify if
> it is because of this)
> --------------------------------------------------------------------------------------------------
>
> Key: HBASE-10443
> URL: https://issues.apache.org/jira/browse/HBASE-10443
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.98.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 0.98.1
>
>
> As HBASE-10438 got closed, we still need to fix the Index out of bound
> exception that occurs. If we have a proper fix will fix this, if the bug was
> a false alarm would close this.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
