[
https://issues.apache.org/jira/browse/PHOENIX-4717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16837626#comment-16837626
]
Daniel Wong commented on PHOENIX-4717:
--------------------------------------
The document we were working on is a start of this type of work for sure though
I hope for it to be broader in topic and discuss PK layout in all situations.
[~swaroopa] and I have a beginning of a document for this but its still in
draft. FYI [~tdsilva]
> Document restrictions on adding primary key columns
> ---------------------------------------------------
>
> Key: PHOENIX-4717
> URL: https://issues.apache.org/jira/browse/PHOENIX-4717
> Project: Phoenix
> Issue Type: Task
> Reporter: Thomas D'Silva
> Priority: Major
>
> For both views and tables
> 1. We allow nullable columns in the PK, but only if they're variable length.
> Variable length types may be null, since we use a null-byte terminator (which
> is a disallowed character in variable length types). Fixed width types do not
> have a way of representing null.
> There is the following TODO in PColumnImpl.init
> // TODO: we may be able to allow this for columns at the end of the PK
> 2. We disallow adding a column to the PK
> a) if the last PK column is VARBINARY
> b) if the last PK column is fixed width and nullable //not sure if this is
> possible currently because of #1
> c) if the column is not nullable (in order to handle existing rows)
> For views:
> 1. We disallow adding a column to the PK If the last pk column is variable
> length
> If the last pk column is variable length then we read all the bytes of the
> rowkey without looking for a separator byte so we cannot add a pk column to a
> view if the last pk column of the parent is variable length (see
> https://issues.apache.org/jira/browse/PHOENIX-978?focusedCommentId=14617847&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-14617847)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
