Well…
Why have ALL the vhosts defined on a box if you don’t do HTTP load balancing on this box ?
In fact you must things to something different that now, not the actual solution to do this, but more on a new concept.
It could be possible to define a schem load balancing before setting up the load balancing. This schem structure will then setup the sync type and the ipvsadm corresponding
It is what you began to do with the load balancing page in nodeworx
Today the load balanced shem is set for VIP:SERVICES
SO it could be possible do things in two parts :
1- Define the load balancing policy
this part doesn’t setup the load balancing
2- Validate it
this part setup the load balancing and the nodeworx sync policy
For exemple. Tell I have 6 nodes and 10IPs IPs on the CM
the 2 first nodes are for mail only
the 4 others are for Apache only
So I define this in the step 1 :
IPs 1to10:25,110,995,143,993 –> node1 and node2
IPs 1to5:80,443 –> node3 to node4
Ips 6to10:80,443 –> node5 to node6
So why have vhost on node1 and node2 ?
And why have vhost on IP 6to10 on node3 and 4 ?
Having all these vhosts use OS ressources for nothing, as each vhost open at least two descriptors (error_log and transfer_log)
This charges all the boxes even if they don’t treat HTTP requests for example.
Of course this doesn’t tell it is not a good solution to change the descriptors limits in Apache, but I think that if you’ll give this kind of solution, it’ll allow to define an important cluster architecture.
I think it could be a real plus and it is not, for me, only a workaround.
But I agree with you that having a HUGE cluster architecture is maybe not the best solutions !
Pascal