On 4/1/26 6:09 AM, Ihar Hrachyshka wrote:
> The localnet_learn_fdb test expects hv1 to accumulate exactly five
> PACKET_IN2 messages after the 14/24 and 15/25 race scenarios. In
> flaky runs, the earlier 14/24 phase sometimes leaves hv1 with an extra
> src14 miss because the ARP 14 -> 24 request is injected before hv1 has
> installed the lookup_fdb flows for 00:00:00:00:10:14. That shifts the
> cumulative count from 5 to 6 even though the later 15/25 phase behaves
> the same way in passing runs.
>
> Synchronize the 10:14 learn before entering the 14/24 race by waiting
> for the FDB row and both hv1/hv2 lookup_fdb flows. Keep the exact hv1
> count check, and update the nearby comment to explain why hv1 is stable
> here while hv2 remains timing-dependent.
>
> Fixes: 584eae5e12fd ("tests: Fix flaky "pod to pod with localnet_learn_fdb".")
> Signed-off-by: Ihar Hrachyshka <[email protected]>
> ---
Hi Ihar,
Thanks for the patch, looks good to me! Applied to main and all stable
branches down to 24.03.
Regards,
Dumitru
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
