Questions

Gustavo Niemeyer gustavo.niemeyer at canonical.com
Tue Mar 19 13:25:44 UTC 2013


On Tue, Mar 19, 2013 at 9:56 AM, Dimiter Naydenov
<dimiter.naydenov at canonical.com> wrote:
> On 19.03.2013 03:57, Mike Sam wrote:
>> 1> Why was zookeper removed in juju-core and replaced with mongodb?
> It was a design decision. Mostly related with instability and issues with
> zookeeper. Cannot be more specific, perhaps you should ask Gustavo Niemeyer.

I wouldn't claim instability was an issue. In fact, the problem was on
our side: despite the fact we do coordination work with our data
store, our usage resembles a lot more a traditional database than a
coordination service such as ZooKeeper. Then, there are a few things
MongoDB does well for us. Some notes off the top of my head:

- We benefit form the fact the data isn't all in memory
- We benefit from primary and secondary indexes
- We benefit from the several atomic operations
- Adding/removing nodes in MongoDB is a lot simpler
- Setting up the replica set in MongoDB is a lot simpler
- etc

I wouldn't claim ZooKeeper is unstable, though. I'd still use
ZooKeeper (or Doozer) for the kind of coordination work that these
services are meant to handle.


gustavo @ http://niemeyer.net



More information about the Juju mailing list