[CentOS] High Availability using 2 sites

Les Mikesell lesmikesell at gmail.com
Fri Jan 6 02:03:38 UTC 2006


On Thu, 2006-01-05 at 18:48, Bryan J. Smith wrote:

> > You can if you always offer distributed locations and
> > let the client choose the address.
> 
> The problem with that is it is too arbitrary.

No, that's why letting the client decide is the best
approach.  Nothing else can know for sure whether a
connection is possible to the given IP addresses.

> > Web browsers already do that.
> 
> I think we disagree there.  And I think you are stretching
> some things to fit web browsers that are simply not true.

Have you tried the test I suggested yet?

> > I have F5 3dns boxes as the primary DNS servers but
> normally
> > let them give out both addresses for all services, all the
> > time.
> 
> Once again, you're looking at it from your perspective very
> close to the authority.  That's completely different than any
> arbitrary user who may be several non-authoritative
> resolutions away.

Yes, I control it from the registered primary dns servers
for the zone but the users are scattered over the world
behind all sorts of intermediate DNS servers.  That doesn't
matter.  You put 2 A records in the servers.  The clients
get 2 IP addresses.  No amount of caching changes that.

> > IE mostly just works.  Our own client software takes care
> of
> > failover using the addresses supplied by DNS. It has its
> own
> > heartbeat on the server connection and will reconnect
> anytime
> > it notices a problem with the connection, trying every
> > address in the list.  When it reconnects it refreshes
> certain
> > things from the new server connection.
> 
> Whoa!  Whoa!  Whoa!!!
> 
> You're talking about heartbeats and other "keep alives" that
> are not common to web servers with many, many clients from
> many, many web clients.  You're almost approaching a stateful
> client/connection when you do such, along with the
> associated, added traffic.

Web just make a new connection whenever they need one.  If
there is a visible problem the user will punch the
reload button to force it.  Other apps tend to be stateful
which is why you need to build in the logic to fix it
when they reconnect.  This will be the case even if you
fudge the failover with expensive hardware tricks instead
of making the app smart enough to do it on the client side.

> So, again, your context is _very_different_ than what I
> understand the need to be here for generic web servers and
> browsers.

I think the original question was about web and other
services. In the 'other' case it might be their own
program where they can make it work.

> > Try it.  If you are resolving names with netbios you might
> > see something different.
> 
> *SMACK*  ;->  Right there, you don't understand a thing about
> how ADS-DNS works.  No offense.  ;->

No I don't, but if you can't put in two A records and
have any client's DNS lookup receive them (as demonstrated
by the 'nslookup www.ibm.com' example) it is broken.

> It is _not_ NetBIOS.  MS IE does some nasty stuff when it has
> ADS.  MS IE does some stupid stuff when it doesn't as well.

It can't be bad enough that other zone's A records disappear
or you wouldn't be able to use the internet.

-- 
  Les Mikesell
   lesmikesell at gmail.com





More information about the CentOS mailing list