[
https://issues.apache.org/jira/browse/CASSANDRA-1415?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12902485#action_12902485
]
Jonathan Ellis commented on CASSANDRA-1415:
-------------------------------------------
01 changes the index creation approach used when building the index from the
data file after streaming. the old approach scans through the new data file
and applies the data in those row fragments to the index, but this is buggy: if
the data file contains old data, or newer data arrives via a write after the
file is streamed, we will corrupt our index. instead, 01 creates a
Table.rebuildIndex method that scans the index columns from the _entire_ row
(that is, merged w/ other sstables), with locking so that when we scan that
row, we're guaranteed that it's the most recent version. The KeyIterator class
is what we use to tell this scanner what keys to index. (We don't use
getRangeSlice because the streamed rows aren't guaranteed to be contiguous.)
The rebuild is done in the CompactionManager executor to avoid excessive i/o
conflicts.
02 is a simple refactor, pulling addIndex out of CFS constructor.
03 extends addIndex to adding indexes that weren't previously defined. we add
INDEX_CF to record when an index is created; if addIndex is called for one that
doesn't exist, we initiate the build using a KeyIterator that merges all keys
from all sstables. This makes sure that if we add an index, and the index
build doesn't complete before the server is restarted, the build will restart
too.
> Allow creating indexes on existing data
> ---------------------------------------
>
> Key: CASSANDRA-1415
> URL: https://issues.apache.org/jira/browse/CASSANDRA-1415
> Project: Cassandra
> Issue Type: New Feature
> Components: Core
> Affects Versions: 0.7 beta 1
> Reporter: Jonathan Ellis
> Assignee: Jonathan Ellis
> Fix For: 0.7 beta 2
>
> Attachments: 0001-Table.rebuildIndex.txt,
> 0002-extract-addIndex-method.txt,
> 0003-allow-addIndex-to-create-indexes-that-did-not-previous.txt
>
>
> (This ticket is only for the index-CF creation part, not for the thrift API
> changes, which may come for free from CASSANDRA-1285)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
