On Mar 29, 10:54 am, Brian Aker <[email protected]> wrote: > "Virtual Buckets" is a feature I called "wheel" that I removed a couple of > years ago because support was dropped in Memcached (and it looks like it will > go back in, in 1.6).
I've only heard you refer to consistent hashing as "wheel." The managed bucket code was removed a while back because we found it was wrong and couldn't see how to make it right at the time. > One thing that I am noticing in the memcached 1.6 tree is that there are > some protocol additions that have not been documented, and there are zero > emails to any of the mailing list about them. Personally I think it is bad > business to see that stuff going into 1.6. If "memcached" is supposed to be > the reference implementation, then there needs to be code in the server > showing how these features work. Which features are you referring to that memcached is lacking? There is code in the server using just about every feature we've thought of that makes sense in a cache, documentation in at least the header files in the branch that has the vbucket features (added last May), and it's consistent with my blog post from last June. There's not been too much discussion on the list on anything related to development (other than some frustration expressed for us not releasing any of the new code we've been working on). So we've had trouble putting a date on a release again, and more conversation of development details from the future has been in other channels. Here, we ask people to check out new things and rarely get responses. There's not a lot of evidence that there's a huge memcached server developer community on this list, or even people all that willing to experiment on new things. I like to think there is, but feedback is minimal here, so we typically get it elsewhere. My hope is that after the recent work we've done on polishing stuff up, we can announce a new dev release, features you can get from it, and hopefully get more users involved. I've been using it in various production environments for over a year (on the same code we've kept published for everyone else). I'd love to have done many releases over the last year or so and have had lots of people excited about them and solving their problems without dirtying it up for other people. It's cool stuff and I know people doing cool stuff with it. Development has been proceeding pragmatically and openly. I don't think it's particularly nice or even beneficial to continually lob stuff into the list with every change we make when we don't have any plans to let users have them.
