| layout | title |
|---|---|
post |
Riak::Light, and other Riak Perl clients |
I am not a Riak pro, especially on the administration side. I have used it a lot, but if you see any error, misconception or missing information, please let me know in the comments.
Riak is an open source key value distributed database. It consists of a so-called Ring of VNodes, which run on one or more servers, storing data in a redundant manner, making it scalable, fast, secure, fault-tolerant.
Riak provides a limited numbers of features, but they are carefully chosen, and properly implemented.
It's the main feature: set/get a value. Nothing fancy here, except that it handles minimal replication of the data, quorum decision between underlying nodes, and provides a way to resolve conflicts (to avoid race conditions).
Riak can be used with Bitcask, or LevelDB. There is also the memory backend
-
Bitcask basically is a simple backend, storing all the keys in memory, and values on disk and memory cache. It has a slightly better performance (actually, a more steady performance) than LevelDB, but requires a minimum amount of memory (to store keys for the given node), and doesn't implement secondary indexes. However, it implements automatic expiration
-
LevelDB maintains a sorted list of keys, and keys plus values are stored on disk, and relies on the OS for caching (I think). It can then implement secondary indexes, but not automatic expiration.
-
The memory backend: I've never used it, so I won't talk about it. It's rarely used in production. Redis is usually a better option, and dramatically faster.
Each key belongs to a bucket, some kind of namespace. Keys must be unique inside a bucket, so you can have the same key name in different buckets
I've not used this feature, but an object can point to an other one, by reading its properties. So you can do link walking, which is quite handy for graphs and related concepts.
To be able to access a key, you have to know its name. You can technically list all the keys, but don't do that on production, it's slow. The key name can be considered as first index to retrieve the value.
You can use the REST API to connect to Riak, but I tend to prefer the Protocol Buffers API connection, as it's faster.
Available with the elevelDB backend, secondary indexes are additional properties that can be added to objects, so that they can be retrieved via other means than just their key names.
As numerous NoSQL databases, Riak supports Map Reduce operations
For some time, Net::Riak was the only Perl client. It's powerful and full featured, but its code is quite complex (some might say overengineered), and uses Moose extensively. Specifically, every value retrieved from Riak is an object. This and other stuff makes it quite slow (from a client CPU usage point of view). And sometimes it maters, when you want to process very quickly a lot of keys, and have a very fast Riak server nearby.