[
https://issues.apache.org/jira/browse/CASSANDRA-2982?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13082885#comment-13082885
]
Jonathan Ellis commented on CASSANDRA-2982:
-------------------------------------------
bq. if Memtable flushes successfully and then Lucene is not able to commit
for the existing indexes, we write the indexes first. if we fail, the "main"
memtable will not be written. if we succeed and the "main" memtable fails, the
main (technically, both, which is fine since the index flush is idempotent)
will be replayed from commitlog.
> Refactor secondary index api
> ----------------------------
>
> Key: CASSANDRA-2982
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2982
> Project: Cassandra
> Issue Type: Sub-task
> Components: Core
> Reporter: T Jake Luciani
> Assignee: T Jake Luciani
> Fix For: 1.0
>
> Attachments: 2982-v1.txt, 2982-v2.txt, CASSANDRA-2982-v3.patch
>
>
> Secondary indexes currently make some bad assumptions about the underlying
> indexes.
> 1. That they are always stored in other column families.
> 2. That there is a unique index per column
> In the case of CASSANDRA-2915 neither of these are true. The new api should
> abstract the search concepts and allow any search api to plug in.
> Once the code is refactored and basically pluggable we can remove the
> IndexType enum and use class names similar to how we handle partitioners and
> comparators.
> Basic api is to add a SecondaryIndexManager that handles different index
> types per CF and a SecondaryIndex base class that handles a particular type
> implementation.
> This requires major changes to ColumnFamilyStore and Table.IndexBuilder
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
