trying to replace calico with OVN/frr for a telco customer.  we send and receive routes successfully, we support the network functions successfully, but when we do resiliency testing, specifically the bond failover/BGP failover, we have traffic impact.. it appears the received routes take an additional 5 seconds to be pushed to the kernel by Zebra.

This test fails with OVN/frr but passes with calico

    4.20.10 with additionalRoutingCapabilities enabled

    100%

    1. set up a secondary bond with active/passive (mode 1)
    2. set BGP up with FRRConfiguration (BGP is active/passive too)
    3. set up any kind of traffic
    4. flip the bond with "ip l set dev xxx down"
    5. failover is fast (depending on your timers)
    6. routes are exchanged
    7. 5 seconds delay
    8. routes are finally pushed to the kernel routing table
    9. End-of-RIB message appears in the logs of FRR
    

    with an active/passive BGP setup, it is important that the routes are pushed to the kernel routing table fast, calico/bird seems to be able to push these routes under 1 second, but FRR/zebra adds a 5 seconds delay

    zebra should not delay pushing the new routes to the kernel table