On 10/07/2010 12:59 AM, Andy wrote:
> Ah thanks for the explanation. I was hoping for an automated setup without the need to get paged 24/7.
>
> So HA is still as hard as I thought it would be. I was hoping that with 9.0 things would be easier.
>
>
My 0.02.
Whether you need 3 servers (or 2 or 5 or even just 1) is a business
decision informed by technological constraints. HA/redundancy is
basically just like insurance - how much you should spend depends on the
frequency of failure (MTBF), how long it takes to repair a failure
(MTTR), and the cost to the business of downtime and/or data-loss. For
many businesses, a single server is fine but for others 3 isn't close to
enough. Based on many years of experience running PostgreSQL, I would
say that using a single-server option would have given us well over
>99.9% availability. (I'm referring to failure-related downtime which
we just never see. Scheduled downtime for updates - especially with the
older releases - are a different story and a second server is very handy
for that.)
99.9% gives you over 8-hours/year downtime - generally pretty easy.
Trimming that downtime to a guaranteed 5-minute maximum (5-nines)
becomes exponentially more costly. (And if your primary worry is lost
sleep due to the 24x7 pager, consider that PostgreSQL properly set up on
quality hardware is pretty friggin reliable. I've lost sleep many times
- it just hasn't been PostgreSQL that caused it.)
Replication is one piece of HA and 9.0 does make that much easier (and
lets you to run read queries on the backup machine so it can earn its
keep). But how to fail from one machine to another is still up to you.
Cheers,
Steve