riak bucket implementations for liza (an kv abstraction)
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
src/liza/store
test/liza/store
.gitignore
.travis.yml
README.md
project.clj

README.md

liza-riak

An implementation of liza's buckets for riak.

Supports merge functions (specified per bucket), serialization/deserialization (ships with a fressian/snappy implementation out of the box), metrics (using yammer metrics), and riak's CRDT counters

Get it from clojars

Design / Goals

liza-riak (unlike liza itself) is meant to be easy - there's only some configuration that this library supports. Specific things that are configurable:

  • serialization (done with a multi method on content-type)
  • allow-siblings (defaults to true)
  • r values
  • last-write-wins
  • host/port
  • storage backends
  • bucket names

Usage

(require '[liza.store :as store])
(require '[liza.store.riak :as riak])
(require '[clojure.set :as set])

(let [riak-client  (riak/connect-client {:host "localhost" :port 8087})
      users-bucket (riak/connect-bucket {:bucket-name "users"
                                         :client riak-client
                                         :merge-fn set/union})]
  (store/modify users-bucket "my-key" (fn [existing]
                                        (conj (or existing #{})
                                              {:email "foo@example.com"})))
  (store/get users-bucket "my-key"))

There are two options here that might not be so obvious: pre-serialize and post-deserialize. These let you convert your data in arbitrary ways before/after passing them to the normal deserialization method. This option is useful if your storage is different from your domain object, but you don't want to write that into custom serialization logic for all buckets.

Important Notes

90% of the time you use this project to mutate data in riak, you'll want the modify function, to prevent vector clock/sibling explosion.

If you're writing anything other than immutable data, you almost definitely want allow-siblings true (which is the default in riak 2.0, and the default in this project). See http://aphyr.com/posts/254-burn-the-library and http://aphyr.com/posts/285-call-me-maybe-riak.

You should (probably) test that your merge function actually is associative/commutative/etc, preferably using generative testing. I like both test.check and clojure.test.generative for that.

Testing

This client includes a connect-test-bucket, which is just a helper for using the in memory riak storage option.

License

Copyright © 2014 Tom Crayford

Distributed under the Eclipse Public License, the same as Clojure.