[ 
https://issues.apache.org/jira/browse/PHOENIX-1427?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14209269#comment-14209269
 ] 

Lars Hofhansl commented on PHOENIX-1427:
----------------------------------------

Checkout the part that completely avoids doing anything with the family in 
updateStatistics if this is running in a major compaction.
In that case we know the family ahead of time.

In the other case I had an earlier patch that does the getFamilyArray huh hah 
as well. Then a consideration is that we might be holding on to the byte[] of a 
very large KV for a while, so I undid that part again.

So the main new thing is the ability to set the family and familyMap just once 
ahead of time when we're doing a major compaction.

> Reduce work in StatsCollector
> -----------------------------
>
>                 Key: PHOENIX-1427
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-1427
>             Project: Phoenix
>          Issue Type: Bug
>            Reporter: Lars Hofhansl
>            Assignee: James Taylor
>         Attachments: 1427-4.2.txt, PHOENIX-1427.patch
>
>
> I noticed that the StatsCollector does a non-trivial amount of work during 
> HBase compactions.
> In a sort of worst case scenario (single node cluster, all data on SSD), it 
> adds almost 50% to the compaction time - in a real setup the relative time 
> spent there would be much less of course.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to