[ 
https://issues.apache.org/jira/browse/HBASE-2286?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12852882#action_12852882
 ] 

George P. Stathis commented on HBASE-2286:
------------------------------------------

Clint,

A few additional notes: I downloaded the latest trunk and the patch is outdated 
against it. I was able to manually apply the differences but I am getting test 
failures:

Running org.apache.hadoop.hbase.client.tableindexed.TestIndexedTable
Tests run: 7, Failures: 0, Errors: 7, Skipped: 0, Time elapsed: 1.511 sec <<< 
FAILURE!
Running org.apache.hadoop.hbase.regionserver.transactional.TestTHLog
Tests run: 5, Failures: 0, Errors: 5, Skipped: 0, Time elapsed: 1.307 sec <<< 
FAILURE!
Running org.apache.hadoop.hbase.regionserver.transactional.TestTHLogRecovery
Tests run: 2, Failures: 0, Errors: 1, Skipped: 0, Time elapsed: 450.93 sec <<< 
FAILURE!

But, a --dry-run against the 0.20.3 distro source seems to apply with no 
problems, so I'll try the patch there instead of trunk.

> [Transactional Contrib] Correctly handle or avoid cases where writes occur in 
> same millisecond
> ----------------------------------------------------------------------------------------------
>
>                 Key: HBASE-2286
>                 URL: https://issues.apache.org/jira/browse/HBASE-2286
>             Project: Hadoop HBase
>          Issue Type: Improvement
>          Components: contrib
>            Reporter: Clint Morgan
>             Fix For: 0.20.4, 0.21.0
>
>         Attachments: hbase-2286.patch
>
>
> This patch fixes a few issues where puts/deletes occur with the same 
> timestamp.
> In the indexing layer, we avoid a Delete followed by a Put to the same row 
> for the index update. When the row is the same, we can just do the put.
> In the transactional layer, we correctly handled put, put, scan. This way the 
> scan will see the last put, even if they have the same timestamp.
> Remove the sleep to fix the putPutScan transactional test, and run it many 
> times to be sure we hit the case where they are in the same millisecond.
> Also some small cleanup, null handling, and fail-fast changes.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to