[
https://issues.apache.org/jira/browse/PHOENIX-5261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16835964#comment-16835964
]
Swaroopa Kadam commented on PHOENIX-5261:
-----------------------------------------
Answering the outstanding questions:
I checked against 4.x branch, after a small PoC:
(Phoenix doesn't allow creating a view with a subset of columns)
1. Does splittable system catalog already take care of Item 1? *Yes*
2. What about tenant-owned views? *Yes*
3. Should there be a way to exclude adding the column to a child index to allow
for uncovered usage?
Seems a fair ask; we can have a grammar like below:
{code:java}
ALTER TABLE ADD COLUMN CASCADE <comma spearated list of indexes | ALL > IF NOT
EXISTS <column references>
{code}
> Implement ALTER TABLE ADD COLUMN CASCADE
> ----------------------------------------
>
> Key: PHOENIX-5261
> URL: https://issues.apache.org/jira/browse/PHOENIX-5261
> Project: Phoenix
> Issue Type: Improvement
> Reporter: Geoffrey Jacoby
> Assignee: Swaroopa Kadam
> Priority: Major
>
> Phoenix currently gives no way to alter the columns in an index. This is
> because adding a column to an index would require a full rebuild to cycle
> through the parent table and add the values to the index.
> There is a special case, however, when adding a column to an index would be
> very useful and easy, and that is when adding a new nullable field to the
> index's parent table. In this case it is safe to just go ahead and add the
> column to the index, because at DDL time the field is known to be NULL on
> each row.
> I propose adding an optional parameter CASCADE to ALTER TABLE ADD COLUMN,
> which when used will do the following:
> 1. The new column will be automatically added to any child views.
> 2. The new column will be automatically added as an INCLUDED column in any
> secondary index belonging to the parent table.
> Outstanding questions:
> 1. Does splittable system catalog already take care of Item 1?
> 2. What about tenant-owned views?
> 3. Should there be a way to exclude adding the column to a child index to
> allow for uncovered usage?
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
