Uploaded image for project: 'Fast Datapath Product'
  1. Fast Datapath Product
  2. FDP-1032

selection fileds for OVN LBs and ECMP routes do not support IPv6 addresses

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Normal Normal
    • None
    • None
    • ovn24.09
    • 3
    • False
    • Hide

      None

      Show
      None
    • False
    • Hide

      Given a system administrator configures ECMP routes to manage traffic in an environment with mixed IPv4 and IPv6 addresses, 

      When they configure selection fields to include ipv6_src and ipv6_dst addresses for load balancing or ECMP routing, 

      Then, OVN system should correctly translate these fields to the corresponding flow fields and handle IPv6 traffic in the same way it currently handles IPv4 traffic.

      Show
      Given a system administrator configures ECMP routes to manage traffic in an environment with mixed IPv4 and IPv6 addresses,  When they configure selection fields to include ipv6_src and ipv6_dst addresses for load balancing or ECMP routing,  Then, OVN system should correctly translate these fields to the corresponding flow fields and handle IPv6 traffic in the same way it currently handles IPv4 traffic.
    • rhel-9
    • rhel-sst-network-fastdatapath-ovn
    • ssg_networking

      Allowed values today are: "eth_src", "eth_dst", "ip_proto" (ECMP routes only), "ip_src", "ip_dst", "tp_src" and "tp_dst".

      Unit and system tests suggest that ip_src and ip_dst supposed to work for both IPv4 and IPv6. But that is not the case, because they are just translated to flow fields, where ip_src/dst is an alias for nw_src/dst which is IPv4. So, if someone wants to load balance based on IPv6 addresses, they can not.

      The system tests that cover hash fields for load balancing use a single network namespace as a source, so the source IP and MAC are always the same, as well as destination IP, which is a virtual IP of the load balancer. So, this case is not actually tested. The tests are using ip_* fields in the IPv6 tests as if they represented ipv6_* fields.

      I think, this breaks support for IPv6 services with affinity timeout in ovn-kubernetes, as they are load balancing exclusively on ip_src and ip_dst, i.e. there will be no load balancing.

              lorenzobianconi lorenzo bianconi
              imaximet@redhat.com Ilya Maximets
              Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

                Created:
                Updated: