[
https://issues.apache.org/jira/browse/LUCENE-2720?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12924174#action_12924174
]
Michael McCandless commented on LUCENE-2720:
--------------------------------------------
bq. And if we can arrange for addIndexes(Directory...) to throw that too, that
would complete the cycle I think.
I agree.
bq. BTW Mike, I think we should track the version per-segment because only when
all segments of version X are gone, can the minimum version Y be recorded in
segments_N
Yes, each SegmentInfo inside the segments_N file should track the util.Version
when it was created.
bq. This is fine for 3.1 and onwards indexes, however what do we do w/ 3.0
ones? We will need to distinguish >=3.1 and <3.1.
Actually we need only to distinguish 2.x vs 3.x right? I think we can just use
the stored fields file, since it changed as of 3.0?
> IndexWriter should throw IndexFormatTooOldExc on open, not later during
> optimize/getReader/close
> ------------------------------------------------------------------------------------------------
>
> Key: LUCENE-2720
> URL: https://issues.apache.org/jira/browse/LUCENE-2720
> Project: Lucene - Java
> Issue Type: Bug
> Components: Index
> Reporter: Michael McCandless
> Fix For: 3.1, 4.0
>
>
> Spinoff of LUCENE-2618 and also related to the original issue LUCENE-2523...
> If you open IW on a too-old index, you don't find out until much later that
> the index is too old.
> This is because IW does not go and open segment readers on all segments. It
> only does so when it's time to apply deletes, do merges, open an NRT reader,
> etc.
> This is a serious bug because you can in fact succeed in committing with the
> new major version of Lucene against your too-old index, which is catastrophic
> because suddenly the old Lucene version will no longer open the index, and so
> your index becomes unusable.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
