|
On Thu, 10 Jan 2008, Raphael Vallazza wrote:
I'm getting arithmetically increasing numbers of replies, so
I'm only replying to lvs-devel this time.
Yes, it should also work without a VIP.
Transparent proxying is very important for us because we
use IPVS to cluster application level proxies on
firewalls. I.E. we have 2 nodes and one of the two acts as
master/director, balancing traffic between node 1
(himself, localnode) and node 2 (slave, realnode).
are you using transparent proxy for localnode?
We would like support transparent proxying for squid and
let the master/director node act as the gateway, outgoing
web traffic to port 80 balanced over the node and
tranparently redirected to port 8080 (squid). Probably
it's an unusual setup for IPVS ;)
sorry if I'm being thick and miss the point of your :-)
here, but you must know that squids were the first major use
of lvs. I haven't kept track of how people are load
balancing squids now. I guess I'd assumed they were putting
a fwmark on anything to 0/80 (or equivalently NAT'ing it to
3128 and then fwmarking it).
what's the localnode + realserver problem with transparent proxy?
It's the NAT problem i've described in the previous mail...
I didn't get it then and I'm still having trouble with it.
if LVS resides in LOCAL_IN, traffic has to be DNATed
before LVS comes into play and the realservers return
packets with the wrong/NATed destination port to the
client.
you're talking about the original LOCAL_IN ipvs. Packets to
localnode is just forced to be accept locally. You have
LVS-NAT with packets being accepted by transparent proxy.
Can't remember what I thought happened with LVS-NAT and
transparent proxy. Just looked at the HOWTO and see the
httpds receiving packets by transparent proxy have to be
setup to reply on the VIP, but I don't know your problem
yet. Can you show the path of the packets and the IPs
involved?
Hmmm, i think i've just understood the solution with
FORWARD :) It should also work with transparent proxies,
LVS has to nat the traffic for the localnode, instead of
letting netfilter DNAT/REDIRECT it.
sounds like part of the problem above that I don't
understand.
I'll try to put LVS inside the FORWARD chain and see what
happens :)
OK. good luck.
The user expects that incoming packets will be messed with
by iptables/conntrack before LVS gets them (and the reverse
on sending packets out to the realservers). If ipvs is in
PREROUTING, you're going to have to make sure that nothing
ever gets put in the PREROUTING chain after LVS. If ipvs is
in FOWARD then the user will already know that they have to
know the order in which packets are hooked (I don't expect
much chain hooking goes on in FORWARD)
Joe
--
Joseph Mack NA3T EME(B,D), FM05lw North Carolina
jmack (at) wm7d (dot) net - azimuthal equidistant map
generator at http://www.wm7d.net/azproj.shtml
Homepage http://www.austintek.com/ It's GNU/Linux!
-
To unsubscribe from this list: send the line "unsubscribe lvs-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
|
