[ 
https://issues.apache.org/jira/browse/HBASE-7902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13586669#comment-13586669
 ] 

Sergey Shelukhin commented on HBASE-7902:
-----------------------------------------

This depends on your definition of "incorrectly" :) 
Given that there's no control over compaction trigger time in most normal 
scenarios, and no sync between puts and compactions at all strictly speaking, I 
view put-in-the-past to memstore -> major compaction and major compaction -> 
put-in-the-past to memstore as purely timing difference, not semantic 
difference. Unless user wants to manually sync every put and every major 
compaction.
This patch will exacerbate the "timing" problem quite a bit so probably in this 
form it shouldn't be included (although scan type rename and separation from 
major in general is needed, I'll make a patch tomorrow).
Now, for stripes it /seems/ the workaround (again, excluding memstore) will be 
relatively easy (there will have to be special handling in policy, compactor 
and maybe in top level scanner; essentially different file scanners will have 
different scan types; but this is from memory, need to check the code).
But for level, if we implement it, we never have all files for any key.

                
> deletes may be removed during minor compaction,  in non-standard compaction 
> schemes
> -----------------------------------------------------------------------------------
>
>                 Key: HBASE-7902
>                 URL: https://issues.apache.org/jira/browse/HBASE-7902
>             Project: HBase
>          Issue Type: Improvement
>          Components: Compaction
>            Reporter: Sergey Shelukhin
>            Assignee: Sergey Shelukhin
>            Priority: Minor
>         Attachments: HBASE-7902-v0.patch, HBASE-7902-v0-with-7843.patch
>
>
> Deletes are only removed during major compaction now. However, in presence of 
> file ordering, deletes can be removed during minor compaction too, as long as 
> there's no file that is not being compacted that is older than the files that 
> are.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to