Node affinity.

38 views
Skip to first unread message

Daniele Aluigi

unread,
May 10, 2015, 8:50:27 AM5/10/15
to redi...@googlegroups.com
I'm evaluating to scale my actual master --> slave configuration to a clustered one.

Is there any affinity / pinning options?
I'll try to explain what i would like to do, maybe my idea could be wrong.

I created a cluster based on 3 VM, each one will be master for its subset of hash and also slave of the others.
Basically each server will have 3 instances of Redis running on differente ports, something like it:

VM A, master of A, slave of B and C
VM B, master of B, slave of A and C
VM C, master of C, slave of A and B

At the moment i'm looking for the way to tell my cluster: both slaves of A should never runs on the same
server (ip - ports ?) where the A master is running, but i think i have to write some code to manage it?

Any suggestions?
Thsnk, Daniele

Daniele Aluigi

unread,
May 21, 2015, 10:20:39 AM5/21/15
to redi...@googlegroups.com
Is this a stupid question or am i missing something important about documentation?

Please, give me a feedback anyway. Thanks!

The Baldguy

unread,
May 22, 2015, 11:32:47 AM5/22/15
to redi...@googlegroups.com


On Sunday, May 10, 2015 at 7:50:27 AM UTC-5, Daniele Aluigi wrote:
I'm evaluating to scale my actual master --> slave configuration to a clustered one.

Is there any affinity / pinning options?
I'll try to explain what i would like to do, maybe my idea could be wrong.

I created a cluster based on 3 VM, each one will be master for its subset of hash and also slave of the others.
Basically each server will have 3 instances of Redis running on differente ports, something like it:

VM A, master of A, slave of B and C
VM B, master of B, slave of A and C
VM C, master of C, slave of A and B



If you need to grow the cluster it will only get worse.

Note: if you lose a VM you lose 1/3rd of your cluster - a master AND a slave. My suggestion is to run each in an independent VM, and if possible one different physical VM hosts. Then you won't lose 1/3rd of your cluster on VM failure and don't have to worry about how to schedule instances. 
Reply all
Reply to author
Forward
0 new messages