[discuss] Which Lucene index should be used as a default

[Chetan Mehrotra <ch...@gmail.com>]

Hi Team,

The work required for following Lucene features is nearing completion.

* OAK-2268 - Support index time Aggregation of repository nodes
* OAK-2278 - Same Lucene index to support both property restrictions
and fulltext constraints
* OAK-2261 - Enable support for NodeType based indexing rules

Due to the new features introduced and change in index schema existing
index cannot be used (untill reindex). To manage that I had to
introduce IndexFormatVersion (OAK-2276). In brief currently there are
two versions

* V1 - Compatible with existing LuceneIndex and its fulltext support
* V2 - Newer version supporting LucenePropertyIndex and newer fulltext support

Oak would be able to support both indexes on same system depending on
the configuration.

The newer Lucene fulltext support would not support some cases. See
[1] and probably would not be supporting them in near future. To allow
users to try newer index and adapt the indexing configuration to get
such queries work I was thinking to keep V1 as the default for now and
later switch the default to V2 (say Oak 1.0.10?). To make use of new
features they need to set 'compatMode' to 2 in index configuration

Q. Would it be fine to continue to use V1 (old LuceneIndex) for newer
setups as default for next release 1.0.9

For now I have kept V1 as the default and opened OAK-2290 to track later switch

Chetan Mehrotra
[1] https://issues.apache.org/jira/browse/OAK-2268?focusedCommentId=14225916l#comment-14225916

Re: [discuss] Which Lucene index should be used as a default

[Chetan Mehrotra <ch...@gmail.com>]

On Thu, Nov 27, 2014 at 2:11 PM, Davide Giannella <da...@apache.org> wrote:
> Will it be too painful for merging?

Should not be much issue as just need to change the default in
org.apache.jackrabbit.oak.plugins.index.lucene.IndexFormatVersion#getDefault.
Only issue is changing the default would cause some tests to fail
(tracked in OAK-2282). So would need to manage that part.

For now would merge it first with default to V1. And for trunk would
change it to V2 once I get the test case part managed by early next
week


Chetan Mehrotra

Re: [discuss] Which Lucene index should be used as a default

[Davide Giannella <da...@apache.org>]

Good morning Chetan,

On 26/11/2014 09:46, Chetan Mehrotra wrote:
> ...
> The newer Lucene fulltext support would not support some cases. See
> [1] and probably would not be supporting them in near future. To allow
> users to try newer index and adapt the indexing configuration to get
> such queries work I was thinking to keep V1 as the default for now and
> later switch the default to V2 (say Oak 1.0.10?). To make use of new
> features they need to set 'compatMode' to 2 in index configuration
>
> Q. Would it be fine to continue to use V1 (old LuceneIndex) for newer
> setups as default for next release 1.0.9
>
will it be possible to have oak trunk with V2 as default and 1.0 with
V1? In this way customer already running the stable version won't see
any difference while the ones using the unstable for experimental
purposes could leverage the new algorithms and report issues before it
goes in production. Will it be too painful for merging?

Cheers
Davide