I am trying to set up a VxLAN tunnel across 3 physical hosts in a data center.
OK, full disclosure, I am trying to run a single kubernetes cluster across 3 KVM hypervisors, and using private non-routable IPs for the master/worker node VMs. Enter Openvswitch and VxLAN tunneling.
So first I configured the 2 nics in each machine to a bond config using standard linux kernel bonding module. Next I created an ovs bridge on that bond device, then I created 2 vxlan adapters in openvswitch, one for each remote host.
Somehow, this caused my hosts to become unavailable (not immediately, but after a few mins). Further, it created connect issues on other hosts on the same physical switches as my 3 hosts. The NOC came back saying that my 3 hosts had "stolen" the gateway hsrp IP address. Unfortunately I have no idea why or how that happened.
I *thought* that a software network config like OVS would not affect the "underlay" network at all, instead just ride along with encapsulated packets.
If this is not the case, then it looks like setting up Openvswitch in a live data center is not safe for a non expert to even try. Otherwise, it must be the bond setup causing issues.
As I am not a network engineer, I am making assumptions that may be completely untrue. In trying to learn "just enough SDN" to set up Openvswitch, I could very well be coming up short.
Any opinions on whether OVS can be the culprit?
No comments:
Post a Comment