/ Zope / Apsis / Pound Mailing List / Archive / 2006 / 2006-01 / pound fail-over and proxy behavior

[ << ] [ >> ]

[ Issue with HTTPS URI turning into HTTP / Jeffrey ... ] [ Re: [Pound Mailing List] pound fail-over and ... ]

pound fail-over and proxy behavior
Josef Schmitz <realmde(at)yahoo.de>
2006-01-31 12:29:44 [ FULL ]
Got a hint on the tinydns list to take a look at
pound. 

Need cheap solution for servers in different
geographic regions and subnets for HA. Will
redirecting work over different subnets? Wanted to
place pound and apache on the same hardware seperated
IP and Port binding.

What happens if pound in location A fails and the
requesting client will get his IP? Most modern
Browsers will skip to next IP within a short time, but
Proxies sametimes behave broken. Have seen this with
squid and round-robin DNS entries. You also have no
control about dnscaches that will not work correctly
regarding TTL ageing.


	

	
		
___________________________________________________________ 
Telefonate ohne weitere Kosten vom PC zum PC: http://messenger.yahoo.de

Re: [Pound Mailing List] pound fail-over and proxy behavior
Adam Andrzej Jaworski <emaylx(at)gmail.com>
2006-01-31 13:02:36 [ FULL ]
On Tue, 31 Jan 2006 12:29:44 +0100 (CET), Josef Schmitz wrote:
[...]

redirecting is different than using backend server in pound,
this is just plain redirect to absolute or relative URL,
another story is with load balancer feature - but this works
in backend context, not in redirect special Service

can you explain a little more how it should work in
your setup ?

and a little off topic:

considering pound as a proxy is excellent choice
have you considered as a backend lighttpd instead of Apache?
to get a picture, see:
http://weblog.textdrive.com/article/44/taking-a-full-frontal-slashdot-lighttpdly

Cheers,
Adam

Re: [Pound Mailing List] pound fail-over and proxy behavior
Josef Schmitz <realmde(at)yahoo.de>
2006-01-31 18:39:46 [ FULL ]
> redirecting is different than using backend server[...]

Sorry did not mean the traditional http redirekt
feature. Question should be, is it possible to use
backend-servers in different subnets or locations
 [...]

First idea was 2 Server in different Datacenters. DNS
sends the two IP's if client connects. Most modern
Webbrowser will handle this and connect to both
servers. If one server is down, client drops
connection within a few seconds and uses the remaing
server. Problem between client and server could be
several problems that are not in your reach. Broken
DNS-Caches, Webproxies etc. Squid has a 5 min.
connection time-out and caches DNS entries for 6 hours
in default config. So i think dns based solution will
not work for several reasons. 

Second idea was keepalived, but servers need to be in
the same network segment, so if datacenter fails it
also does not work. Think is the same for pound.

Best practice would be pound with backends at one
(well connected)location with high sla. Is it
possibleand recommended to use pound and webserver at
the same host. What could be done with 2 hosts?

Will all traffic from webserver flow back through
pound to the clients or will the backends deliver
direct?


[...]

pound really looks nice.



	

	
		
___________________________________________________________ 
Telefonate ohne weitere Kosten vom PC zum PC: http://messenger.yahoo.de

MailBoxer