[
https://issues.apache.org/jira/browse/PHOENIX-3262?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15479382#comment-15479382
]
James Taylor edited comment on PHOENIX-3262 at 9/10/16 5:50 PM:
----------------------------------------------------------------
There's a shadow column family for each column family the data table has. Those
shadow column families are used by all local index. The local index row key has
the index number as part of the row key.
We created the entire index in ~17 seconds. Why wouldn't we be able to drop it
using delete markers in the same or less time? I think a further optimization
we can do, since the index number won't repeat (i.e. the data will no longer be
accessed), is ideally just not write the data a compaction time (PHOENIX-1590)
or at a minimum issue the delete markers asynchronously (PHOENIX-3180).
was (Author: jamestaylor):
We created the entire index in ~17 seconds. Why wouldn't we be able to drop it
using delete markers in the same or less time?
There's a shadow column family for each column family the data table has. Those
shadow column families are used by all local index. The local index row key has
the index number as part of the row key.
> When dropping a local index, delete the column family instead of individual
> rows.
> ---------------------------------------------------------------------------------
>
> Key: PHOENIX-3262
> URL: https://issues.apache.org/jira/browse/PHOENIX-3262
> Project: Phoenix
> Issue Type: Bug
> Reporter: Lars Hofhansl
> Fix For: 4.9.0
>
>
> Currently when Phoenix drops a local index it places a delete marker for
> every single row, instead we can just drop the L#<x> column family.
> {{HBaseAdmin.deleteColumn}} in 0.98. And {{HBaseAdmin.deleteColumnFamily}} in
> 1.0+.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
