[
https://issues.apache.org/jira/browse/HBASE-18145?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16034120#comment-16034120
]
Anoop Sam John commented on HBASE-18145:
----------------------------------------
Very imp finding..
In shipped() new call to clearAndClose(scannerForDelayedClose);, We should do
this after prevCell clone. That is better as there is no chance at all that
prevCell belong to this scanner which is getting closed now.
clearAndClose(scannerForDelayedClose); call in close() -> Looks ok to do
this whatever be the boolean param. Still it might be better to double check.
May be best would be follow the other list close way of whether boolean true,
then only close here. After shipped() call any way close is there. WDYT? We
have to change the param name in that case
bq.close(boolean withHeapClose)
> The flush may cause the corrupt data for reading
> ------------------------------------------------
>
> Key: HBASE-18145
> URL: https://issues.apache.org/jira/browse/HBASE-18145
> Project: HBase
> Issue Type: Bug
> Affects Versions: 2.0.0
> Reporter: Chia-Ping Tsai
> Assignee: Chia-Ping Tsai
> Priority: Blocker
> Fix For: 2.0.0, 1.4.0, 1.3.2
>
> Attachments: HBASE-18145.v0.patch
>
>
> After HBASE-17887, the store scanner closes the memstore scanner in updating
> the inner scanners. The chunk which stores the current data may be reclaimed.
> So if the chunk is rewrited before we send the data to client, the client
> will receive the corrupt data.
> This issue also breaks the TestAcid*.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
