[
https://issues.apache.org/jira/browse/CASSANDRA-11046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15115363#comment-15115363
]
Sam Tunnicliffe commented on CASSANDRA-11046:
---------------------------------------------
||branch||testall||dtest||
|[11046-3.0|https://github.com/beobal/cassandra/tree/11046-3.0]|[testall|http://cassci.datastax.com/view/Dev/view/beobal/job/beobal-11046-3.0-testall]|[dtest|http://cassci.datastax.com/view/Dev/view/beobal/job/beobal-11046-3.0-dtest]|
Seems like the dtest run I kicked off when I posted the patch didn't run for
whatever reason, so I've started another
> Existing indexes are always rebuilt on upgrade to 3.0
> -----------------------------------------------------
>
> Key: CASSANDRA-11046
> URL: https://issues.apache.org/jira/browse/CASSANDRA-11046
> Project: Cassandra
> Issue Type: Bug
> Components: Distributed Metadata
> Reporter: Sam Tunnicliffe
> Assignee: Sam Tunnicliffe
> Priority: Critical
> Fix For: 3.0.x, 3.x
>
>
> CASSANDRA-10127 made the naming of secondary indexes consistent internally.
> Previously some places used just the index name, whilst others required the
> name in the format {{table.index}}.
> One place including the additional table name was the {{system.IndexInfo}}
> table, which records whether or not the index has been built. On upgrade to
> 3.0, as the node restarts and initialises and index, it checks this table to
> determine whether an initial build task is necessary. Since 10127, this check
> expects the row to include just the index name, but the actual row will still
> be in the old {{table.index}} format, causing the index manager to assume the
> index is not built and submit a build task.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
