Hi,
Seeing something odd and wondered if anyone else had seen this or had any ideas.
In a nutshell, trying to get OSPF neighbour up between UDLR instance and an ESG. It works fine if both on the same edge cluster host. It fails if you separate them between edge cluster hosts. However, the key point is that when on separate hosts you can ping the corresponding interfaces. This last fact tells me that VxLAN transport between hosts is working.
Although I am not sure of the frame sizes involved with OSPF hellos (imagine very small), we know it is not an MTU related issue as the underlay has interfaces set to 9k.
The environment is simple, just started build out, so only the ESG and UDLR instances and the transit universal logical switch.
Replication mode has been set to unicast.
NSX Version is 6.4.1, building a cross-vCentre NSX deployment, will be using Active-Active with Local Egress eventually.
Interestingly the OSPF peering to an external router from the ESG works fine also, but that is obviously connected via more traditional dvSwitch and associated port group.
Have also tested with the external connection to router from ESG disabled and again makes no difference.
Any ideas?
Thanks
Thanks all for your replies.
This issue has since gone, but not because we found the solution, but rather we changed approach to using BGP and have had no issues with this.
I wanted to use BGP from the off to match VVD for SDDC but was initially met with some resistance that has now been resolved. All is working.
I am still intrigued to what the issue was, if I ever get time I may look into it in a lab.
Thanks
Trying to get OSPF neighbour up between UDLR instance and an ESG. It works fine if both on the same edge cluster host
This test proves OSPF config is fine
The key point is that when on separate hosts you can ping the corresponding interfaces. This last fact tells me that VxLAN transport between hosts is working
Interface which you are referring is VTEP VMK IP's ? - End to End MTU is must ( Including Server profiles based on the architecture) . Hope your VXLAN transport VLAN is also correct ?
When Edge and UDLR CVM is residing on different host .
Do we have IP connectivity from Edge to CVM Protocol address IP ? Perform a debug ip OSPF and verify the output followed by no debug ip ospf to disable the same.
Thanks for your reply, just to clarify the ping bit:
"The key point is that when on separate hosts you can ping the corresponding interfaces. This last fact tells me that VxLAN transport between hosts is working
Interface which you are referring is VTEP VMK IP's ? - End to End MTU is must ( Including Server profiles based on the architecture) . Hope your VXLAN transport VLAN is also correct ? "
From the DLR VM (on Edge host x), Can ping the ESG interface that is used as the OSPF neighbour (on Edge host y), thus VxLAN working between host X and Y
MTU and VXLAN Transport VLAN are fine.
We did do debug, It was done in both directions. Both sides are sending OSPF hello but not receiving any
So Hello packets itself is getting dropped . Can i get a sample routing table output with OSPF configs for DLR and Edge ?
I have had this issue before where even though the MTU was correct on each interface, OSPF failed when routers were on different hosts. What I did to fix the issue was Edit the Area to Interface Mapping setting. Under Routing, OSPF, edit the Area to Interface Mapping. Check the box that states "Ignore Interface MTU setting". It then worked with no issues. This was a lab though and do not know if this is appropriate for a production system. Might be worth a try to see if you can get it working.
Thanks all for your replies.
This issue has since gone, but not because we found the solution, but rather we changed approach to using BGP and have had no issues with this.
I wanted to use BGP from the off to match VVD for SDDC but was initially met with some resistance that has now been resolved. All is working.
I am still intrigued to what the issue was, if I ever get time I may look into it in a lab.
Thanks