Geo redundant OSS database
squitoey at gmail.com
Fri Jan 20 00:12:06 MST 2012
On Jan 19, 2012, at 10:04 PM, Bryan Sant <bryan.sant at gmail.com> wrote:
> Has anyone had any experience working with PostgreSQL or MySQL in a
> geo redundant configuration (replication/clustering)? I see that
> MySQL has MySQL Cluster
> (http://www.mysql.com/products/cluster/)--which is their "Carrier
> Grade" offering. Looks compelling. I know that PostgreSQL (>=
> version 9) has clustering as well, but I'm not sure how feature
> complete it is. I'm interested in any insights you might have.
I wasn't terribly impressed with MySQL clustering. Granted this was a
few years ago that I set it up. What bothered me, was that it all had
to fit in memory, and if the entire cluster died (like what happened
to my servers last week in the C7 data center power outage) you lose
Depending on the application, I would recommend an eventually
consistent approach (assuming your project would allow for it). Throw
in a distributed hash table for who controls what resource, and you
can end up with a pretty robust setup with out some of the
complexities of clustering the database. That said, you get a little
more complexity in the app itself. Can you tell I am leaning toward
the "shared nothing" approach.
If you are interested I can explain further how I have been
approaching this for my projects.
Hope this helps, I've never used the clustering for postgres, so maybe
someone smarter than I can comment there.
More information about the PLUG