[CentOS] "Site down for maintenance" - How is this accomplished?

Warren Young warren at etr-usa.com
Sat Aug 25 00:18:31 UTC 2007


Brian Mathis wrote:
> 
> Messing with DNS is really the wrong way to go on this.  You'd be
> forcing all of the DNS servers involved to start messing with their
> caches, update more frequently, etc.., pushing the problem out onto
> "everyone else", and you have no control over any of it really.  Cache
> time is only a suggestion, and not all DNS servers follow it.

The only moral of this story is that you can't control everything.  That 
doesn't mean you shouldn't do what you can to speed the changeover.

Temporarily lowering the DNS cache time for a server during a time of 
change is well established practice.  The only problem is that it'll 
increase the load on the primary name servers for that domain.  That's 
the best argument for it being temporary.

> The way to go is to assign that same IP address to another box during
> maintenance, and have that box show the page.

The original poster did not explicitly say so, but the impression I got 
is that this "data center" is off-site, and possibly managed by a third 
party.  Letting multiple machines handle a single IP only works within a 
single physical site.  When multiple sites are involved, the only way to 
move an IP is to change global routing rules, which is more fraught with 
problems than the DNS change-over, which you already don't like.

I guess theoretically you can have a single IP at multiple sites, with 
both routes being advertised as viable, but this will cause chaos as 
half the packets go one direction and half go the other.  "Don't cross 
the streams" indeed.



More information about the CentOS mailing list