You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@phoenix.apache.org by "Chao Wang (Jira)" <ji...@apache.org> on 2020/06/30 02:10:00 UTC

[jira] [Commented] (PHOENIX-5507) Local Index Rebuild Issues

    [ https://issues.apache.org/jira/browse/PHOENIX-5507?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17148251#comment-17148251 ] 

Chao Wang commented on PHOENIX-5507:
------------------------------------

hello, I find this issues in my production environment, which the index data will be hidden behind the deletes. only we can drop index ,and recreate index table. Is there a better way?

> Local Index Rebuild Issues
> --------------------------
>
>                 Key: PHOENIX-5507
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-5507
>             Project: Phoenix
>          Issue Type: Bug
>    Affects Versions: 4.15.0, 5.1.0, 4.14.2, 4.14.3
>            Reporter: Geoffrey Jacoby
>            Priority: Major
>
> An umbrella for dealing with issues around rebuilding local indexes. 
> While investigating PHOENIX-5502, a problem with synchronously rebuilding global indexes, I realized that the rebuild functionality for local indexes seems to have several problems.
>  # Same as PHOENIX-5502, synchronous local index rebuilds will run a delete against the old index data, then rebuild. But the deletes will have server time of the rebuild, and the rebuilt index rows should have the same timestamp as their original data, so the index data will be hidden behind the deletes.
>  # The IndexTool theoretically allows for doing rebuilds on local indexes, but it has a worrisome check that prints a warning message if any split has taken place during the rebuild, saying the rebuild has to be redone. We should probably disable splitting during the rebuild, but then how to guarantee it gets re-enabled if the rebuilding client goes away?
> So at the very least, a solution to the above two issues would have to be found to make local index rebuild work. But it also suggests a larger question – given that local indexes have stronger consistency guarantees than global, is "local index rebuild" actually a necessary feature (vs a drop and rebuild if you somehow messed up a local index out-of-band.)?



--
This message was sent by Atlassian Jira
(v8.3.4#803005)