directory-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Emmanuel Lecharny <elecha...@apache.org>
Subject Re: [ApacheDS] [Bigbang] All tests are now passing
Date Tue, 27 May 2008 16:50:02 GMT
Alex Karasulu wrote:
> Hi all,
>
> Thanks to Kiran and Emmanuel we were able to get the big bang branch (bbb)
> compiling and passing all unit tests and integration tests over the course
> of the last week.  A job well done thanks to some great collaboration.
>   
Man, you did almost everything with Kiran !
> The bbb now holds a version of ApacheDS which uses Cursors instead of
> NamingEnumerations to carry the results of list() and search() operations.
> Also the JDBM store and partition implementation was revamped with various
> new optimizations along with an additional system index. 
I gonna play my own little performance test on bbb to give you a first 
feedback.
>  Search has been
> considerably altered and simplified as a result of these changes.  Note that
> we decided to use our own Cursor interface instead of the JNDI
> NamingEnumerations because of a few reasons:
>
>    (1) Cursors can be positioned after creation.
>             - This prevents many problems and opens the door for some new
>               features in the server.  NamingEnumeration creation requests
> to
>               the search engine must specify direction.
>             - Extraction of search from Partitions will now be possible
> which is
>               needed to be able to implement virtualization.
>    (2) Cursors decouple advances (i.e. next()) from object lookups (get()).
>             - Allows for lazy candidate evaluation which enables us to add
>               filters for entries after creation without missing entries due
> to
>               prefetch operations used NamingEnumerations.  NEs need unlike
>               Cursors need to prefetch entries since they must return
> entries
>               on next() operations.
>    (3) Cursors support element positioning and bi-directional traversal.
>    (4) Remove the JNDI dependency for NamingEnumeration and it's
> limitations.
>   
Also note that the current implementation is way easier to debug and 
understand !
> Now we have some options.  We can continue removing JNDI constructs and
> workarounds introduced to handle it's shortcommings.  This can finish the
> rest of Phase III in the big bang effort.  Hopefully at this point we can
> (with some other changes on our roadmap) release (1.5.4) the big bang after
> promoting it to the trunk.  Until then we can release 1.5.3 which is siting
> in the trunk now.
>
> We could merge what's been altered in the trunk into the bbb and just
> release 1.5.3 after Phase III but I personally think we should just release
> 1.5.3 ASAP.
>
> Thoughts?
>   
Let's release a 1.5.3 quickly. It brings some new features, and also 
some bug fixes. We can release a 1.5.4 with bigbang later. Consider that 
if we remove more JNDI from the server, this can be the last step before 
1.5.4, and we will be able to ditch bigbang.

We are working on bigband since last november, AFAIR, and it's time to 
get back to trunk. Bigbang, as a branch, is not visible, and makes it 
difficult for those who are not involved into this refactoring to help 
us. We knew it when we started bigbang, but it can last forever !

Thanks a lot guys !
> Thanks,
> Alex
>
>   


-- 
--
cordialement, regards,
Emmanuel L├ęcharny
www.iktek.com
directory.apache.org



Mime
View raw message