Redis cluster client library
Redis Cluster is coming out later this year, but I can't wait for it so I made this module.
All it does is connect to the nodes of a Redis Cluster, and before sending any commands it checks in which slot the key is with
HASH_SLOT = CRC16(key) mod 16384 and then sends the command to the node that has that slot.
npm install redis-cluster
This module exports two objects.
clusterClient is to be used with a regular Redis Cluster, you just need to supply a link (like
127.0.0.1:6379) and the other members of the cluster will be found, after that you can use it pretty much like the original
var RedisCluster = require'redis-cluster'clusterClient;var redis = RedisCluster;var redisPubSub = RedisCluster;var assert = require'assert';var firstLink = '127.0.0.1:6379'; // Used to discover the rest of the clusterfirstLinkif err throw err;rset'foo' 'bar'if err throw err;assertequalreply'OK';rget'foo'if err throw err;assertequalreply 'bar';;;;firstLinkrsubscribe'channel';for var link in redisPubSubredisLinksredisPubSubredisLinkslinklinkon'message'// New message in a channel, necessarily 'channel' here because it's the only one we're subscribed to.;;
Don't forget that despite being a thin wrapper above
node_redis, you still can't use all the commands you would use against a normal Redis server. For instance, don't expect the
KEYS command to work (in fact, in the Redis Cluster spec it says that "all the operations where in theory keys are not available in the same node are not implemented").
If you really want to have a cluster of Redis nodes but don't want to run unstable software, you can always use the Poor Man's Cluster Client, also supplied by this module.
This time you can't supply a link of one node of the cluster (maybe because it's not a real cluster), you have to supply the links to all the nodes, like this:
var RedisCluster = require'redis-cluster'poorMansClusterClient;var assert = require'assert';var cluster =name: 'redis01' link: '127.0.0.1:6379' slots: 0 5462 options: max_attempts: 5name: 'redis02' link: '127.0.0.1:7379' slots: 5463 12742 options: max_attempts: 5name: 'redis03' link: '127.0.0.1:8379' slots: 12743 16384 options: max_attempts: 5;var r = poorMansClusterClientcluster;rset'foo' 'bar'if err throw err;assertequalreply'OK';rget'foo'if err throw err;assertequalreply 'bar';;;
As you noticed, you must specify the interval of slots allocated to each node. All 16384 slots must be covered, otherwise you will run in some nasty errors (some keys might have no where to go).
Options are optional and may be added or left out. All valid options for the redis client may be found in the redis client documentation.
If you decide to re-allocate the slots, add or remove a node, you must move all the affected keys yourself. The MIGRATE command might help you with that.
Before every operation, a CRC16 of the key gets computed, so we can know in which node of the cluster this key is. It turns out it's not such an expensive operation to run for every command after all. My laptop can hash 2793296.089 strings of 32 characters per second, that will in no way be a bottleneck to all database operations.
Some quick test showed it achieved a very similar performance to the
node_redis module, so I'll assume it's not so bad and as soon as I have time I'll publish some tests
This is of course not intended for production and has probably stupid (not bad, stupid) code inside, but I just needed something that works as there are no modules to work with Redis Clusters yet.
This module shamelessly borrows some code from the mranney's
node_redis module and alexgorbatchev's
node-crc, although I didn't feel the need to include it as a dependency because only CRC16 is needed.