Michael Gruetzner <mgruetzn at HTWM.De> wrote:
>
> Hi,
> 
> I have to meet the challenge of creating a wide area database cluster.  
> Since
> I have made some good experiences with Slony-1 in the past, I'm  
> wondering
> if it would also work with wide-area networks. Of course, bandwidth  
> and latency
> are main issues but also network failures.
> The Slony-1 documetation says that it might not be suitable if some  
> nodes may
> fail oftenly but it does not explain how slony deals with such  
> failures (maybe someone
> can explain this to me?).
> 
> So what I'm asking is: Does anyone have experience with Slony-1 and  
> wide area
> clusters?

Yes.  My opinion is that it works well for typical cases.

Network failures are a problem for a few reasons:
1) It can take Slony a bit to find it's feet again after a network failure.
   So if you have frequent failures, Slony can get into a situation where
   it can't get caught back up.
2) Slony tends to bomb the slaves with lots of bandwidth when they come
   back online after an outage.  This can (potentially) be a problem if
   your bandwidth is limited and it fills up the pipe, interfering with
   other types of traffic.
3) During an outage, Slony has to track all changes until the slave comes
   back online.  This can use up a lot of disk space pretty quickly, so
   an extended outage can be a real issue if you don't size your hardware
   to account for it.

It's a vague question with a vague answer, because whether or not it will
work for you is a combination of many factors: how much spare bandwidth
do you have?  How frequent are the outages?  How much updating is occurring?
How much update lag is acceptable in the slaves during an outage, and how
much lag to getting caught back up after an outage is acceptable?  Can you
size your disks to be large enough to backlog an outages worth of updates
while the network is down?

If you can get all those issues into a range that's acceptable for you,
then Slony will meet your needs, but there are too many questions and
too many details to be able to provide a pat answer like "yes" or "no".

-- 
Bill Moran
Collaborative Fusion Inc.

wmoran at collaborativefusion.com
Phone: 412-422-3463x4023