I'll take a stab... Inline.
@siculars on twitter
http://siculars.posterous.com
Sent from my iPhone
On Dec 31, 2010, at 14:15, David Hollander <[email protected]> wrote:
I am developing a python web application and have several questions:
1. Can Riak use unix sockets in addition to TCP?
I think the closest you can get to raw sockets is via the native
erlang interface.
2. Load-balancing a REST interface with nginx would be
straightforward.
However, if using protocol buffers with the python client should I
also
be implementing load balancing? Should I use a function to cycle the
port\client connection used?
Maybe ha proxy would shuffle pbuff. Not sure. But I've done it in
nodejs (http/pb) by cycling as you indicated.
3. Is there a limit to the number of buckets? What is the big O access
time for a bucket as number of buckets in the database increases? Is
it
a constant time hashmap? I am considering programmatically generating
many bucket names to partition keys into groups I would map-reduce
over.
Rather than put everything in one bucket and map-reduce over
unnecessary
keys.
There is no limit to the number of buckets. Buckets are a virtual
concept primarily used for namespacing and as a way to isolate
deviations from default settings (think nval, hooks, backends).
Currently all keys in the system are considered when doing an m/r over
a bucket. Afaik, there are plans to fix this in future versions. I
would still take the time to engineer a meaningful bucket key naming
scheme because in the next release there will be all sorts of key
selection goodness that will help with current limitations. Recall
that with the default bitcask backend all keys are kept in ram with
values on disk.
4. Eventual consistency: let's imagine a site has a popular item
with a
score data key where the value is incremented by voting. Let's say the
score is 0, and it receives two votes incrementing it by +1 at similar
times. If these update commands are sent to different nodes, is it
possible for the vote to ultimately read 1 instead of the desired 2?
If
so, would always sending the update commands to the same node prevent
this?
I have a problem relying on a system like riak for autoincrement use
cases for a number of reasons. It's one thing if you can accept a
situation of "more or less" accurate. But if you need to be exact I am
of the opinion that riak is not the system to do it. Sending all
requests to one node will not have the desired effect.
Thanks
_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
_______________________________________________
riak-users mailing list
[email protected]
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
