Description of problem:
When launching an instance and on a node other than the one that holds that NAPT switch - there is no connectivity to an external IP.
The setup is bare metal setup with VLAN. One controller and 2 compute nodes.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1.Create an external and internal network
2.Create a router and attach the external and internal networks
3.Note the compute node that the NAPT switch is on.
4.Launch an instance on a compute node other than the one in step 3.
5.Check connectivity to an external IP.
Created attachment 1438016 [details]
Moved to POST given the solution was merged upstream
On further analysis:
Currently, SNAT traffic always goes through overlay tunnels by design, even when the tenant networking is VLAN.
This is a design oversight on Netvirt and the solution is not simple and would have to be taken up as an RFE.
To get SNAT to work on VLAN setup we need to:
1. Make sure the tunnels are properly set up, so that SNAT traffic can flow through them.
2. Make sure the VMs get MTU 1450, otherwise the overlay encap would make the packet bigger and likely to drop due to default MTU 1500.
Given this information it seems we should:
a. Close the existing bug as VERIFIED with release notes detailing this situation.
b. Open a bug/RFE to fix this design mess in a future release.
In the setup that @Itzik was using, the issue was [1 - above]. After the tunnels were properly setup, SNAT traffic was going fine.
The following RHBZ is opened to track the RFE.
This is not a bug fix, it's a design which should be implemented in Fluorine or a following release.
Since OSP13 ODL is using Oxygen (prior to Fluorine) we can't get this feature.
Yes , this more a feature than a bug and needs design considerations.