[ 
https://issues.apache.org/jira/browse/PHOENIX-3681?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15985822#comment-15985822
 ] 

James Taylor commented on PHOENIX-3681:
---------------------------------------

I think you're going down a rabbit hole as there are many assumptions in the 
code that:
- local index column families are prefixed with "L#"
- table column families are 1:1 with shadow local index column families

The scan of a local index won't be impacted by putting them all in the same 
local index column families since the row key is prefixed with the local index 
ID. You're right that compaction of local index column families may start to 
drive when compaction occurs, but it feels like a second order problem (just my 
two cents). If we do a first level of perf testing of local indexes, my guess 
is that there will be a number of first order issues that would fall out of 
that. I'd hate for you to go down this rabbit hole and swear off any future 
Phoenix work. :-)



> Store local indexes in a column family per index
> ------------------------------------------------
>
>                 Key: PHOENIX-3681
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-3681
>             Project: Phoenix
>          Issue Type: Bug
>            Reporter: Lars Hofhansl
>
> Currently all local indexes are stored in a single column family. That makes 
> maintenance (such as dropping an index) more expensive than necessary.
> Let's have each local index in its own column family (or be able to declare 
> which column family an index should go into).
> As [~jamestaylor] points out, this won't work for indexes on views as there 
> might be 1000's of them.
> Another issue are covered local indexes, but I'd argue that local indexes 
> would benefit little from being covered. (that also needs to be 
> experimentally verified)
> Local indexes in individual column families would be great to isolate any 
> maintenance and even usage from each other.
> [~rajeshbabu], [~mujtabachohan]



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Reply via email to