[ 
https://issues.apache.org/jira/browse/PHOENIX-8067?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Andrew Kyle Purtell updated PHOENIX-8067:
-----------------------------------------
    Description: 
{{UncoveredGlobalIndexRegionScanner}} 's constructor calls 
{{getRegionLocator(dataTable).getEndKeys()}}. In HBase 2.x, 
{{HRegionLocator.getEndKeys()}} scans the meta rows of every region of the data 
table without using the region location cache. This cost is incurred for for 
every index region in every query that uses an uncovered global index, or a 
covered index with a server merge. Each scan does a full meta scan of the data 
table before it reads a row. The cost grows with the number of data table 
regions, and it puts load on the region server that hosts meta.

Scanners should not reload the region boundaries when it opens. For each page 
of data row keys, they should instead find the enclosing region with 
{{RegionLocator.getRegionLocation(row)}}, which uses the location cache.

  was:Placeholder


> Uncovered global index scans read the locations of all data regions from meta 
> at every scanner open
> ---------------------------------------------------------------------------------------------------
>
>                 Key: PHOENIX-8067
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-8067
>             Project: Phoenix
>          Issue Type: Improvement
>    Affects Versions: 5.3.2
>            Reporter: Andrew Kyle Purtell
>            Assignee: Andrew Kyle Purtell
>            Priority: Major
>             Fix For: 5.4.0, 5.3.3
>
>
> {{UncoveredGlobalIndexRegionScanner}} 's constructor calls 
> {{getRegionLocator(dataTable).getEndKeys()}}. In HBase 2.x, 
> {{HRegionLocator.getEndKeys()}} scans the meta rows of every region of the 
> data table without using the region location cache. This cost is incurred for 
> for every index region in every query that uses an uncovered global index, or 
> a covered index with a server merge. Each scan does a full meta scan of the 
> data table before it reads a row. The cost grows with the number of data 
> table regions, and it puts load on the region server that hosts meta.
> Scanners should not reload the region boundaries when it opens. For each page 
> of data row keys, they should instead find the enclosing region with 
> {{RegionLocator.getRegionLocation(row)}}, which uses the location cache.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to