Hi!
Am 04.11.2014 12:46, schrieb Erdősi Péter:
Hmm.. you mean isolated guest network right?
If yes, the IP subnet can be changed, if the user define guest network
(not auto, when first vm started)
There is an option to: Guest Gateway, and Guest Netmask, and if it
filled, the sysvm will use that subnet...
No, I do not mean the guest network (which works fine). I mean the
network between the XenServers, where the guest traffic is carried.
My XenServers have the following interfaces:
- MANAGEMENT
- GUEST (which should be used for GRE-encapsulated private traffic)
- PUBLIC
- ISCSI1
- ISCSI2
All except PUBLIC have IP addresses: MANAGEMENT obviously for XenCenter
and XAPI, ISCSI1/2 for the multipathed shared storage, and GUEST
(intended for GRE traffic, although it is nowhere mentioned that the
interface must have IP addresses).
The intended behaviour is that Cloudstack instructs the OpenVSwitches
within the XenServer nodes to build the GRE tunnels between VMs within
the same Cloudstack isolated Network across the GUEST interface.
But CloudStack chooses one of the other interfaces (MANAGEMENT or ISCSI)
to carry the GRE traffic.
You can see this with "ovs-vsctl show" on one of the XenServers with
currently running instances:
# ovs-vsctl show
...
Bridge "xapi4"
fail_mode: standalone
Port "vif1.0"
Interface "vif1.0"
Port "t1074-8-9"
Interface "t1074-8-9"
type: gre
options: {key="1074", remote_ip="10.33.1.5"}
Port "xapi4"
Interface "xapi4"
type: internal
...
Here you see the remote IP of the other XenServer for the GRE tunnel.
But the IP is wrong, it is on the first ISCSI network.
This works, but of course we want to isolate storage traffic from guest
traffic, so Cloudstack should do as expected and use the "GUEST"
interface ip of the peer XenServer as remote_ip.
Ciao
Martin
