Alan, just catching up on the list....
Actually, you CAN do this with a bit of DNS magic... simply set the TTL for
the DNS authority very low, and you can do this. The TTL being low forces
the clients NOT to cache the answer for very long, and so need the
authoritative
answer, which you can change with a heartbeat monitor. Of course, this
does pose other problems, and doesn't really get past the "single source of
failure" problem he's having....
1) Almost every hit on his site is going to need to have a DIRECT DNS lookup
via his DNS server... not a good thing, it seems to me he'll overload that
DNS server quickly.
2) SPoF on the DNS server, of course... although he could try having
secondary
servers normally, if he has his TTL set very low, the secondary servers
may/may not (depending on the DNS version) work properly as backups... they
aren't authorized to cache the results, for example, and therefore may not
do a proper job backing you up.
3) Updates would (possibly) need to be done at the secondary DNS servers
directly... he may control his primary DNS servers, but secondaries are
normally reciprocal servers, he wouldn't have control over them in all
likelihood.
Just my $0.02, and I may be wrong on some of the points... but I know he can
do this with DNS, just not necessarily practically.
Bill Ward
-----Original Message-----
From: Alan Mead [mailto:[EMAIL PROTECTED]]
Sent: Thursday, November 09, 2000 6:11 PM
To: [EMAIL PROTECTED]
Subject: Re: Primary and Alternate Web Site
This sort of behavior is called names like "high availability" or "failure
roll-over". As you note, when the primary host is not operating, it cannot
refer surfers to the second. So this is not a solution you can implement
on the primary host. It has to be implemented "earlier" in the process or
on a second machine.
What you need is the second server to assume the primary server's IP
address. There are packages like 'heartbeat' and ... blocking on the name,
something like 'fake'... that are designed to help build this solution. If
you search for 'high available linux' you should find some literature on
this.
Or you need some DNS magic. With the way DNS works (it caches data) you
cannot change DNS info on a dime and I'm not sure how you would have a
machine far away assume the IP unless both machines are in the same IP
space (e.g., connected by a WAN). If they are served by different DNS
servers, I don't think this would work. Anyway, you would have to control
your DNS server in order for this to work. And don't go to great lengths
to make your web server redundant and then have a single DNS host become
the SPOF, you'll need a roll-over plan for it as well.
And if your site records data, you may need a way to make this data
available on redundant servers. Coda is the name of a file system which is
supposed to be able to do this (maintain synchronized data across several
file servers) although I've never met anyone running it to ask them how
well it actually works.
You can also buy clustering solutions from Red Hat (or does this come with
one of their products?), Turbo Linux, etc. These would solve much the same
problem you are describing and add a performance boost as well.
I think these are are fairly deep waters.
-Alan
At 10:07 PM 11/9/00 , Ted Hilts wrote:
>This is all fine except when the primary hosted web site dies. Now no
>browser can get to the unadvertised alternative domain name because it
>is not known. So this is the dilemna. How does the alternative hosted
>web site make itself look like the primarly hosted web site when the
>primary web site is down?
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
_______________________________________________
Redhat-list mailing list
[EMAIL PROTECTED]
https://listman.redhat.com/mailman/listinfo/redhat-list
