On Thu, Apr 20, 2006 at 06:45:53PM +0100, Stuart Henderson wrote:
> On 2006/04/20 17:47, Ashley Moran wrote:
> > pf/CARP might worth a try then. The only issue I have is that it's doing
> > whole-server load balancing which is no use if just Apache/lighttpd dies.
> > (I'm more concerned with high-availability than load-balancing.)
>
> You could use a package like monit, or something custom and simpler,
> to restart the httpd if it crashes.
>
> And/or, you could have a simple program checking the backends
> and if one stops answering, remove it from the rdr pool.
>
> > The best alternative I can see is software load balancing on the web
> > servers,
> > and the most promising I've found so far (in terms of lightweightness and
> > simplicity) is Pound.
>
> You then get to start monitoring Pound rather than httpd :)
>
> If you like this route, there's also pen, a general TCP proxy rather
> than HTTP-specific like Pound; it's probably simpler, doesn't need to
> link to a threaded OpenSSL library, and is already in packages/ports.
>
> Advantages and disadvantages either way...
>
> Note that some web browsers will connect to multiple A records
> before returning an error to the user, and that 'machine up but
> httpd down' is detected quickly.
>
> > From what I read, failover is best provided by
> > Heartbeat although so far I have only skimmed a few FAQs.
>
> There's a program called Heartbeat on Linux that appears to
> do a similar job to CARP, so I don't think you'd need that here.
Heartbeat is quite a different beast; it's more akin to a combination of
mon and ifstated.
Some monitoring script sounds like the way to go, though.
Joachim
