Release 1.0.0 is a major feature release and is the first where
riak-client, ripple, and riak-sessions will be released
independently (see below). Because there too many individual changes
to recount, this entry will cover the major features and bugfixes
present in the release.
riak-client is fully compatible with Riak 1.0.x and
(yet-to-be-released) 1.1.x, including supporting secondary indexes,
integrated search, and cluster membership commands.
Riak::Client can now connect to multiple Riak nodes at once. This
greatly improves throughput and allows the client to recover from
intermittent connection errors while continuing normal operation. To
enable this, all uses of the Pump/Fiber logic were removed in favor of
connection pools from which any new request can draw an existing or
create a new connection. Which node is selected for any new connection
is based on a quickly-decaying EWMA of its success rate on recent
requests. A huge thanks to Kyle Kingsbury
who did most of the work on this!
The Riak::TestServer class has been generalized such that you can
generate regular nodes and even clusters that store data on disk. This
is especially useful if you want separate nodes or clusters for each
project that uses Riak, and to keep them separate from your base
install. TestServer also now launches the node in a separate process
(not a child process) so you can keep it running between test
suites. Clearing the in-memory data is performed by connecting to the
console via the exposed Unix pipes, rather than over stdio.
An important part of dealing with eventual consistency is the ability
to handle when conflicts (also called siblings) are created. Now you
can resolve them automatically by registering blocks (callbacks) using
Riak::RObject.on_conflict. The block will be called when fetching a
key in conflict and receives a RObject that has siblings. To resolve
the conflict, it simply returns the resolved object, or nil if it
didn't handle the conflict. A huge thanks to
Myron Marston who implemented this!
Before, serialization of Ruby objects into Riak was constrained to
three formats: JSON, YAML and Marshal. Now you can define your own
serializers so that you can store data in BSON, MsgPack, NetStrings,
or whatever format you like. Use Riak::Serializers[content_type] = serializer to assign a serializer for the selected media type. The
serializer must respond to #dump and #load. (More handiwork of Myron
Marston, thanks!)
If you don't like the keys that Riak hands out when you store an
RObject without a key, and you want something naturally ordered, you
can now generate them client-side using Riak::Stamp, which will
generate 64-bit integers in a fashion similar to Twitter's Snowflake,
but uses Riak::Client#client_id as the machine identifier.
In an effort to decouple development of the individual projects and
reduce top-level dependencies, the ripple repository was split into
new repositories containing its corresponding sub-projects.
Additionally, the Riak::CacheStore has become its own project/gem.
The new gem and repository locations are below:
riak-client— basho/riak-ruby-clientripple— seancribbs/rippleriak-sessions— seancribbs/riak-sessionsriak-cache— seancribbs/riak-cache
Attempting to use the Protocol Buffers transport with a 0.14.x cluster may cause the connection to dump because of incompatibilities in certain protocol messages. This will be addressed in a future patch/bugfix release.
The new node generation and test server intermittently fails on JRuby, specifically from deadlocks related to blocking opens for the console FIFOs. The JRuby team has helped on this issue, but there may not be a clear resolution path until JRuby 1.7 or later.
Other known issues may be found on the Github issue tracker.