[PhillyOnRails] ha and rails

[Flinn Mueller]

Thanks Mark, this is what I was looking for.


On Jun 18, 2007, at 8:23 AM, Mark Chadwick wrote:

> Hey Flinn,
>
> I'm running a production Rails app that replicates with slony.   
> It's been running that way for a couple years (though not always  
> Rails) without problems.  You can keep the fail-over logic out of  
> your code by using Heartbeat.  The primary DB server owns a virtual  
> IP, and accepts all the connections.  If something happens to it,  
> the secondary takes over, and claims the IP.  You might get a bad  
> request or two in there, but it's mostly seamless.
>
> That said, this was all set up before pgcluster was available.   
> I've set up a smaller (not yet in production) environment with  
> pgcluster, and it's a sight nicer.  You, as advertised, talk to  
> whichever host you'd like for reads and writes.  It's very nice.
>
> Even though it's fault-tolerant already by design, I'd probably  
> recommend a pair of machines to run pgcluster's "Load Balancer"  
> with heartbeat between them.  That way you can keep the logic of  
> your environment out of your code.
>
> Otherwise, I'd just suggest catching connection refused exceptions  
> down in ActiveRecord's guts, and retrying on another host.  That's  
> going to get messy pretty quickly, though.
>
> -Mark