-
Bug
-
Resolution: Unresolved
-
Normal
-
None
-
4.18, 4.19, 4.20
Description of problem:
Submariner[1] connects two clusters and ensures the pod and services connectivity between the clusters.
When submarine traffic enters the clusters with OVN-Kubernetes it gets SNATed at ovn-k8s-mp0 and since OVN moved to nftables there is no way to override this behaviour. Earlier with iptables, we avoided this SNAT with a higher priority rule in the POSTROUTING chain. With nftables , even if we add a higher priority rule, all the rules will be executed before taking a final decision, so there is no way to skip the SNAT.
If OVN-Kubernetes could expose a generic mark value that Submariner could tag the traffic with and avoid an SNAT, it would help us solve the issue.
Why is this needed?
Submariner wants to preserve the source IP when the traffic reaches the destination pod as some use cases rely on the source IP.
ovnk upstream PR:
https://github.com/ovn-kubernetes/ovn-kubernetes/pull/5113
Openshift ovnk PRs:
4.20 : https://github.com/openshift/ovn-kubernetes/pull/2627
4.19: https://github.com/openshift/ovn-kubernetes/pull/2642
- impacts account
-
ACM-17819 Submariner Connectivity E2E failing on OCP 4.18
-
- Review
-