[Mondrian] Flushing SegmentCache

Luc Boudreau lucboudreau at gmail.com
Mon Feb 13 15:59:20 EST 2012


I'm saying we have found some issues with the cache management code in
3.3.X, but it should still work (except a few corner cases). Please
log a Jira case with as much information and reproducible tests as you
can so we can look into your issue. Regardless of the fact that it
works on 3.3 or not, I'd like to make sure we fixed the problem for
the next release.

Also, just a heads up. We have modified the SegmentCache SPI and
simplified it greatly for the next release. For one thing, you won't
need to implement the SPI using Java Futures, nor will you need to
implement SegmentCache.flush() on the SPI side (Mondrian will take
care of that). I strongly suggest that you take a look at the next
version of the SPI; it might make your life that much easier.

http://perforce.eigenbase.org:8080/@rev1=head@//open/mondrian/src/main/mondrian/spi/SegmentCache.java

Please also note that the next version of the SPI includes a
SegmentCacheListener system. Your cache implementation should make use
of it for optimal performance. Mondrian will call
SegmentCache.addListener() at runtime and will register a listener
object with your implementation. This makes Mondrian much more
efficient in maintaining an index of the contents of the external
caches by using deltas to figure out how to sync up the segment
indexes. Listeners and near-real-time updates of the indexes helps
Mondrian perform in-memory rollups of your cell data (this is an order
of magnitude more efficient than SQL access). I know, it sounds pretty
hermetic and complicated. We still haven't had the chance to write a
neat blog post to explain it all yet.

Bottom line: I suggest you invest the little time necessary to develop
your code against 3.4.0. The new & improved SegmentCache SPI is most
likely the one we will support long-term wise. Version 3.4.0 is almost
out of the door, so the transition should be as smooth as can be.

Cheers!

Luc


On Mon, Feb 13, 2012 at 3:07 PM, Pedro Alves <pmgalves at gmail.com> wrote:
>
> This is for a plugin we want to release for 3.3 also, earlier if possible.
>
> Are you saying that it's not possible to do selective cache clearing at all
> :-(
>
>
>
> -pedro
>
>
> Luc Boudreau <lucboudreau at gmail.com> wrote:
>>
>> For our next release (due some time in March) we have modified
>> substantially the code which manages segments and cell data. I remember
>> seeing that issue, so I suggest you grab the latest binary build off our CI
>> server and see if it is gone.
>>
>> If not,  please log a jira case (with a test case on Foodmart preferably).
>>
>> Luc
>>
>> On Feb 13, 2012 2:26 PM, "Tiago Ferreira" <tiago.ferreira at webdetails.pt>
>> wrote:
>>>
>>> Hi,
>>>
>>> I'm using a custom SegmentCache implementation in mondrian 3.3.
>>> Using cacheControl.flush(cacheControl.createMeasuresRegion(cube)) as in
>>> the examples in http://mondrian.pentaho.com/documentation/cache_control.php
>>> for flushing a cube, my SegmentCache's flush method always gets invoked with
>>> an empty ConstrainedColumn[] and segments aren't removed.
>>> Any clues on what I might be doing wrong (or not getting)?
>>>
>>> Thanks
>>>
>>> --
>>> Tiago G.F.
>>>
>>>
>>> _______________________________________________
>>> Mondrian mailing list
>>> Mondrian at pentaho.org
>>> http://lists.pentaho.org/mailman/listinfo/mondrian
>>>
>
> _______________________________________________
> Mondrian mailing list
> Mondrian at pentaho.org
> http://lists.pentaho.org/mailman/listinfo/mondrian
>


More information about the Mondrian mailing list