[ 
https://issues.apache.org/jira/browse/LUCENE-6658?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14614638#comment-14614638
 ] 

Uwe Schindler commented on LUCENE-6658:
---------------------------------------

bq. This sounds easy, but there are probably traps. For instance, there might 
be files in the directory which don't really belong to the index. Earlier 
versions of Lucene used to have a FilenameFilter which was usable to 
distinguish one from the other, but that seems to have disappeared, making it 
less obvious how to do this.

Because you want to upgrade the index, you can also open it with a 
DirectoryReader and check maxDoc() for 0. Opening a DirectoryReader should be 
easy if you have the JAR files at hand - you need them anyways for upgrading.

We can add similar behaviour to the IndexUpgrader in 5.x, but not in older 
versions!

> IndexUpgrader doesn't upgrade an index if it has zero segments
> --------------------------------------------------------------
>
>                 Key: LUCENE-6658
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6658
>             Project: Lucene - Core
>          Issue Type: Bug
>          Components: core/index
>    Affects Versions: 4.10.4, 5.2.1
>            Reporter: Trejkaz
>
> IndexUpgrader uses merges to do its job. Therefore, if you use it to upgrade 
> an index with no segments, it will do nothing - it won't even update the 
> version numbers in the segments file, meaning that later versions of Lucene 
> will fail to open the index, despite the fact that you "upgraded" it.
> The suggested workaround when this was raised on the mailing list in January 
> seems to be to use filesystem magic to look at the files, figure out whether 
> there are any segments, and write a new empty index if there are none.
> This sounds easy, but there are probably traps. For instance, there might be 
> files in the directory which don't really belong to the index. Earlier 
> versions of Lucene used to have a FilenameFilter which was usable to 
> distinguish one from the other, but that seems to have disappeared, making it 
> less obvious how to do this.
> This issue is presumed to exist in 3.x as well, I just haven't encountered it 
> yet because the only empty indices I have hit have been later versions.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to