Removal of Jiak
alanmckean at me.com
Mon Feb 22 14:45:59 EST 2010
Regarding the POV, I had forgotten that the Jiak implementation of read/write masks was done at the bucket level. What was needed was a switch at the individual operation level ... one that would not cause a global effect for all users.
I'm glad to hear about the pre and post-commit hooks. Will they be configurable at the level of the individual operation or will they be installed once for the bucket? Being able to switch the hook implementations for each operation would enable write masks for each user. Also, will those pre and post-commit hooks be there for reads? If there were hooks for pre-reads, you could mask the reads as well.
How cool would it be to have a 10K object in the store and be able to limit access to specified fields on a user-by-user basis or request only a small portion of it. I would think that it would lead to a trememdous performance improvement to not have to distribute the entire object for many use case
On Feb 22, 2010, at 11:22 AM, Bryan Fink wrote:
> On Mon, Feb 22, 2010 at 2:06 PM, Alan McKean <alanmckean at me.com> wrote:
>> I'm going to miss the read/write mask capabilities of Jiak. As Paul Rogers pointed out with his implementation POV (see the Buoy POV example at http://www.riakrest.com) it enables really nice declarations of data structures that are subsets of the data stored in Riak. Is there any notion of building that into Riak at the bucket level?
> There is no current plan to reimplement read/write mask capabilities
> for the HTTP interface. We found that they added more confusion than
> they added useful features. Indeed, POV was interesting, but it was
> not realistic for concurrent access by multiple actors.
> We've found that one good way to get similar functionality (partial
> reads & updates) is to break objects into smaller logical pieces, and
> then use link-traversal to grab the pieces you want.
>> The other significant feature we are losing is Jiak's ability to hook in to the lifecycle of Riak operations with bucket modules. That's not in the raw interface either, is it?
> It is not yet, but it is coming. In the very near future, either for
> this release or right after, we intend to add "pre- and post-commit
> hooks" to allow applications to tie in at that level.
More information about the riak-users