Riak 0.7.1 list_buckets times out on Mac OS X but not on Ubuntu
zubairq at gmail.com
Fri Feb 12 00:55:26 EST 2010
Ok, thanks, that was quite a complete answer! :) I guess the more I think
about it the more I see you are right. I was thinking there would only be a
few easily cachable bucket names (<100) on most systems but after reading
more in the Riak user archives I realise that some people may want to store
individual records in their own bucket, thereby creating potentials billions
of buckets, and not so easy to cache. Anyway, I think I can probably do
something like just cache the bucket names in a specially named bucket
together with a reference count for the number of items in each bucket for
now, and just do a list_keys on that bucket, as a temporary solution for
now. Anyway, thanks for the in depth explanation.
> How long would it take to implement list_buckets still showing the buckets
> > when a node goes down?
> It depends on what our answer is, ultimately. There are many options
> for how it should be done, and each has its own effect on the
> guarantees that Riak would provide about that data. It's something
> we're looking into.
> > I disagree with you where you say list_buckets is not
> > used that often, as I think that list buckets is a common operation for
> > DBAs, and if it stops working when Riak goes down then it should be
> > altogether, as a dynamically scalable database shouldn't stop working
> when a
> > machine goes down (I thought this was the whole point of being fault
> > tolerant).
> Rusty's phrasing was unclear. You may want to list_buckets, but you
> shouldn't do it often in a production environment, with Riak's current
> implementation. We disagree about removing it altogether right now,
> because it is a very useful operation, especially when you're in
> development (non-production) mode, often dealing with half-complete
> sets of test data, etc. We recognize that there is some level of
> demand for a production-ready version of this operation, and we are
> evaluating our options for implementing one.
> > Anyway, since I am trying to make a tool to browse a Riak cluster
> > how else can I get a list of the buckets, which is a common operation in
> > use case of a Riak database browser.
> We've implemented quite a few in-house Riak browsers over the last
> couple of years. Many of them weren't used on production datasets, so
> listing buckets was fine. The ones that were used in production
> either knew of a pre-defined list of buckets to check, or asked the
> user to type in a bucket to examine. Since buckets, in a sense,
> "always exist" (the list_buckets operation just lists buckets that
> currently have data in them), asking the user for a bucket to examine
> isn't really that awkward of an idea.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the riak-users