On Fri, 25 Oct 2013, hitesh wrote:
The data is going to come from different servers and the connection is going
to be over UDP.
What I am trying out is similar to UDP Multicast scenario where the data is
going to be received only on a single node ( among the multiple nodes) at
the receiver. I dont want the control to be at the sender side to distribute
the data.
I am trying the clone properties of the pacemaker which when set to
"Globally Unique" sends the data to multiple hosts. I dont want the same
data to go to multiple nodes. I am checking on that currently.
Do you have any idea on that.
Ok, using clusterip with UDP traffic is something I've been doing for many years
(again with syslog data).
what happens with clusterip is that all the traffic goes to both systems, but
the OS network stack filters out half the traffic (based on the hash criteria)
so that the application running on each system only processes half the traffic.
If you just use clusterip the way you have it configured, when both nodes are
up, you should see that IP on both boxes. When something sends traffic to that
IP, only one box will have the traffic get to the userspace app, but both boxes
will set the traffic at the interface/OS layer.
Since this is UDP traffic, there aren't connections to deal with as such, but
when an application opens an outbound UDP socket, the OS picks a souce port.
Each time the application opens a new socket, a different port will be used.
If the sending application runs for a long time and keeps it's send of the
socket open, traffic from that server will 'stick' to the node it first hits. If
the sending application keeps closing and opening sockets, it's traffic will
bounce between the systems. Since you haven't said what the application is, I
don't know if the application can handle that or not (syslog can for example,
but some other protocols use UDP, but still have a 'connection' as far as the
application is concerned and won't handle it). If they don't, change the hash
mode to just sourceip instead of sourceip-sourceport.
If you have things setup per your config, check that both nodes are online and
have the resource active. At that point, both systems should get some of the
traffic.
David Lang
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
