[
https://issues.apache.org/jira/browse/PHOENIX-3604?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15832642#comment-15832642
]
James Taylor commented on PHOENIX-3604:
---------------------------------------
Instead of tracking the dropped columns, we can query the physical table and
views at compaction time to compute the "gaps" in column qualifier names. As
compaction occurs, we can strip the value out and update the offset.
[~lhofhansl] - is it possible to modify a cell during major compaction?
> Support partial drop column on immutable tables with encoded data
> -----------------------------------------------------------------
>
> Key: PHOENIX-3604
> URL: https://issues.apache.org/jira/browse/PHOENIX-3604
> Project: Phoenix
> Issue Type: Task
> Reporter: Samarth Jain
>
> We need to support dropping a column from an encoded immutable table. For v1,
> we should support dropping the column only from the metadata. This way users
> won't be able to access that column anymore through regular phoenix queries.
> They still would be able to access the data through hbase until we support
> version v2.
> To support v2 (see PHOENIX-3605 ), we would need to add a "DROPPED" metadata
> column in SYSTEM.CATALOG. The compaction hook can then look for this
> attribute in the metadata row for the column and decide whether or not it
> should be replacing the contents with null. Our code in MetadataEndPointImpl
> too would then conditionally add this column to PTable so that any queries
> accessing the dropped column would fail.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
