|
On Tue, May 06, 2003 at 04:38:25PM +0530, Aneesh Kumar K.V wrote:
> On Tue, 2003-05-06 at 16:13, Horms wrote:
> > On Tue, May 06, 2003 at 04:12:21PM +0530, Aneesh Kumar K.V wrote:
> > No.
> >
> > One way to get around this is to use a tool which changes
> > the LVS rules. So the Master would have the LVS rules
> > but the Standby would not. If the Master fails then
> > the Satndby has the rules instantiated.
> >
>
> I will state the full requirement. I handle the LVS integration part of
> www.openssi.org. Now in the case of openssi we added something called
> automatic port binding. What that means is when a bind()/listen()
> happens on any node in the cluster on a port that is specified to be
> load balanced then the service is registered with ipvs master director
> node. In other words everything happens transparent to the user. User as
> normal will run a server and it is automatically registered with ipvs.
> Now this registrations is duplicated across all the potential director
> nodes ( slave director nodes.) . Now when the master director node goes
> down a user space daemon will select the one of the slave director node
> as master send the gratious ARP with the MAC address. There is no way we
> can find what services where registered with master director node at
> this time since all the registration happens transparent to the user(ie
> inside the kernel).
>
> But i can do any setting independent of previous master director node,
> like enable the load balancing on the selected node at this point. Is
> there a /proc variable that i can set which will do the same. Something
> like
>
> echo '1' > /proc/sys/net/ipvs/vs/loadbalance
>
>
> If not is there any other way of achieving similar functionality ?
No, not that I can think of.
However if such a proc value is useful to you.
It should be quite easy to implement by
modifying LVS.
--
Horms
|
