|
thank you Horms - please bear with me
So let me see here..
Client1 connects from AOL and is routed to RIP1
Client2 connects from AOL before Client1's session times out, so Client2 is
routed to RIP1.
The cookies can then be sent to ensure the two users' sessions aren't
crossed?
I'm sorry Joseph's 1999 lvs-howto gave me the impression that ppc used
source port in routing decisions, or is the current version of ipvs not
using ppc.
Regardless, won't some ISPs swap the masquarade on their outgoing
connections as part of their own load-balancing? and couldn't this allow a
client to connect to the wrong Real Server mid-session?
thanks
-TRevor
Horms wrote:
> On Wed, Nov 01, 2000 at 03:21:56PM +0000, Trevor Marshall wrote:
> > Thanks,
> > I had already read Ted's mail but it didn't mention how the source port
> > comes into play.
> > Am I to assume then that even if the client's source port changes then
> > the cookie will ensure that the director forwards to the same RIP ?
> > regards,
>
> Cookies? There are no cookies in LVS. Persistence is handled by
> templates. A template matches the source ip address and protocol,
> the source port is ignored. The template determines the affinity of
> a matching source to a real server. When a connection is to be scheduled
> for a persistent service if a template can be found then the real
> server information it holds is used, else a new template is created
> that will be used for this and subsequent connections. Templates
> are ip_msaq structures and timeouts are handled by the ip_masq code.
> That is the ip_masq code removes templates once the timeout expires.
>
> --
> Horms
>
|
