Document Redis::Distributed

[mperham]

I can't find any documentation on how to spread my Redis load across many different machines / redis instances. A wiki page or readme entry would go a long way to help.

[djanowski]

+1. So far I haven't heard of anyone using Redis::Distributed for anything other than caching. Hearing from someone who did use it for spreading out application logic too would be nice.

[mperham]

99% of our Redis usage is for caching but right now we use a single master with an idle slave. We'd like to switch to Distributed.

[yankov]

You can use it like that

https://gist.github.com/2485964

It automatically routes request to a redis node according to a consistent hashing algorithm.

[johnkchow]

What happens if a node goes down? Does the library handle this?

[yankov]

It doesn't. You'll have to handle that manually.

[johnkchow]

Thanks for the reply Artem!

I'm really new to Redis and, while I understand the concepts of sharding, I
don't understand how I'd handle this scenario, especially in Redis. Can you
point me to anything (blogs, reading material, etc) that would help me
learn?

On Thu, May 3, 2012 at 4:12 PM, Artem Yankov <
reply@reply.github.com

wrote:

It doesn't. You'll have to handle that manually.

---

Reply to this email directly or view it on GitHub:
#219 (comment)

[vyper]

Do you have any predictions, and has arrived to study the idea of ​​compatibility with the redistribution cluster coming soon?

[yankov]

@jkchow My attempt on solving this: a simple tool to redistribute keys in cluster when you added or removed nodes. Can't say it's tested enough to use it on sensitive data, but idea is very simple and it can give you a start https://github.com/yankov/redis-migrator

[gerhard]

Are there any updates on this? Redis::Distributed works nicely, but I'm still not 100% sure on the best approach of adding new nodes to the cluster. This is obviously not the right way:

$r.ring.nodes.size
=> 10
1000.times { |i| $r.set i, 1 }
=> 1000
$r.keys.size
=> 1000
$r.ring.nodes.map { |n| n.keys.size }
=> [124, 57, 85, 85, 178, 101, 71, 105, 109, 85]
$r.ring.remove_node($r.ring.nodes.first)
=> 160
$r.ring.nodes.map { |n| n.keys.size }
=> [57, 85, 85, 178, 101, 71, 105, 109, 85]
$r.keys.size
=> 876
# using redis_gun for new_redis_server
$r.add_node(new_redis_server.socket)
$r.ring.nodes.map { |n| n.keys.size }
=> [57, 85, 85, 178, 101, 71, 105, 109, 85, 0]
1000.times { |i| $r.set i, 1 }
=> 1000
$r.ring.nodes.map { |n| n.keys.size }
=> [79, 88, 93, 205, 103, 84, 125, 116, 86, 81]
$r.keys.size
=> 1060

[clupprich]

I wonder if this would be a viable solution (of course adding multiple slaves). Any opinions on that one?

[drandahl]

Just bumping this thread. Anyone know whether Redis::Distributed will at least be supported moving forward?

[yaauie]

@drandahl: if you'd like to see a solution, and to see it supported, you're more than welcome to contribute improvements. This project is run entirely by volunteers, and the to-do list is entirely driven by those who have the time & energy to make this project suit their needs.

[djanowski]

@drandahl The plan is to move Redis::Distributed to a separate gem for 4.0. I'll probably do the split but I'll look for people who want to take up the maintainer role.