cross data center leader election?

[Ben Hsu <be...@criticalmedia.com>]

Hello Zookeepers

We would like to use Zookeeper for master election, with the master and the
failover in two different data centers. Concretely, what we are thinking of
is this:

Zookeeper runs in one data center (potentially AWS or other cloud provider)
Process 1 runs in data center 1, in one location
Process 2 runs in data center 2, in a second location

Process 1 and Process 2  both connect to ZK using an election recipe.

Process 1 runs normally, and Process 2 runs as a standby. When Process 1
dies Process 2 will detect the death (using ZK election) and take over.

Our question is, is this feasible with the latencies and levels of
reliability we can expect to see. Has anyone on this list tried this? What
was your experience?

Re: cross data center leader election?

[Jordan Zimmerman <jo...@jordanzimmerman.com>]

This comes up all the time. The short answer: I wouldn’t do it. But, please read Camille Fournier’s post regarding something similar:

http://whilefalse.blogspot.com/2012/12/building-global-highly-available.html


On November 7, 2014 at 5:34:11 PM, Ben Hsu (ben.hsu@criticalmedia.com) wrote:

Hello Zookeepers  

We would like to use Zookeeper for master election, with the master and the  
failover in two different data centers. Concretely, what we are thinking of  
is this:  

Zookeeper runs in one data center (potentially AWS or other cloud provider)  
Process 1 runs in data center 1, in one location  
Process 2 runs in data center 2, in a second location  

Process 1 and Process 2 both connect to ZK using an election recipe.  

Process 1 runs normally, and Process 2 runs as a standby. When Process 1  
dies Process 2 will detect the death (using ZK election) and take over.  

Our question is, is this feasible with the latencies and levels of  
reliability we can expect to see. Has anyone on this list tried this? What  
was your experience?