1. Justin Sheehy
  2. riak


riak / releasenotes / riak-0.5.txt


By popular demand, this release changes the identifier type for
buckets (Riak's namespaces) from atoms to binaries.  Old code using
the native interface must change accordingly; code using the jiak http
interface will continue to work unchanged.  To convert an old
cluster's storage, simply do a backup and restore -- this will convert
all of the data on the way back in.

  - preferred use is now riak:client_connect(Nodename) instead of IP/Port
  - bugfix where timeouts could cause incorrect responses

  - no longer gen_event, now using simple erlang messaging
  - fully capable matchspecs now available for event matching
  - use monitor to watch for death of handlers
  - the example riak_event_logger is now a little friendlier

  - more robust due to explicit ring fetch and node connection

  - more coverage, especially of backend interface, riak_object

  - less messaging via batched groups to each vnode
  - intermediate per-vnode reduce phase

build process:
  - much more useful error message when building with erlang < R13

ring management:
  - simplify metadata structure
  - fix filtered_preflist bug in case of small number of nodes
  - internal naming cleanup
  - prune old state files more aggressively

  - new 'start_apps' param for starting dependency applications
    (useful for some custom storage backends)

native client interface:
  - added list_buckets and filter_keys

client libraries:
  - url-style quoting of buckets and keys in http client content
  - initial version of ruby client
  - java client uses List instead of ArrayList