[jira] [Commented] (PHOENIX-1208) Check for existence of views doesn't take into account the fact that SYSTEM.CATALOG could be split across regions

["Jeffrey Zhong (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/PHOENIX-1208?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14112805#comment-14112805 ] 

Jeffrey Zhong commented on PHOENIX-1208:
----------------------------------------

[~jfernando_sfdc] In which case did you see SYSTEM.CATALOG is split? 

[~giacomotaylor] If SYSTEM.CATALOG is split, we have other issues because we relies on multiple rows update in transaction fashion against SYSTEM.CATALOG and MetaDataRegionObserver won't be singleton anymore. Should we enforce system.catalog not splittable or any other suggestion? 

> Check for existence of views doesn't take into account the fact that SYSTEM.CATALOG could be split across regions
> -----------------------------------------------------------------------------------------------------------------
>
>                 Key: PHOENIX-1208
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-1208
>             Project: Phoenix
>          Issue Type: Bug
>    Affects Versions: 5.0.0, 3.1, 4.1
>            Reporter: Jan Fernando
>            Priority: Minor
>         Attachments: PHOENIX-1208.patch
>
>
> It is possible that when SYSTEM.CATALOG gets very large that it will be split across multiple regions. The parent table metadata is guaranteed via MetaDataSplitPolicy to be in the same region. However, child tenant specific views could end split across multiple regions. The hasViews() method, when checking for the existence of any views, scans only the region the parent table metadata is located in. We should detect whether the views span multiple regions and scan across them all.



--
This message was sent by Atlassian JIRA
(v6.2#6252)