On Thu, May 28, 2026 at 2:45 AM Ales Musil <[email protected]> wrote:
>
>
>
> On Wed, May 27, 2026 at 8:53 PM Mark Michelson via dev
> <[email protected]> wrote:
>>
>> In commit 621f85e92437 ("controller: Fix bfd up too early after
>> unexpected reboot."), OVN would manipulate the flow-restore-wait flag in
>> order to attempt to synchronize with OVS. The previous commit reverted
>> this change due to a potential deadlock between OVS and OVN.
>>
>> If users were using a version of OVN that had commit 621f85e92437, then
>> the revert will help to ensure that OVS and OVN won't deadlock any
>> longer. However, the revert on its own does not resolve the deadlock if
>> the previously running version of ovn-controller set flow-restore-wait
>> to true. It requires manual intervention to get flow-restore-wait set to
>> false or deleted from the Open_vSwitch table.
>>
>> This commit seeks to assist by clearing the flow-restore-wait key in the
>> Open_vSwitch table. This way, on upgrade, the deadlock is resolved
>> without manually setting any OVS database values.
>>
>> Reported-at: https://redhat.atlassian.net/browse/FDP-3862
>> Fixes: 621f85e92437 ("controller: Fix bfd up too early after unexpected
>> reboot.")
>> Signed-off-by: Mark Michelson <[email protected]>
>> ---
>
>
> Hi Mark,
>
> thank you for the patch, I see one slight issue with this.
> This will remove any "flow-restore-wait" unconditionally.
> However, someone else might set it. The reverted code
> actually used "ovn-managed-flow-restore-wait" as an
> indication. We could leverage that and remove
> both "ovn-managed-flow-restore-wait" and "flow-restore-wait"
> only if "ovn-managed-flow-restore-wait=true" WDYT?
>
Good suggestion! I posted a v2 that does this.
>> controller/ovn-controller.c | 14 ++++++++++++++
>> 1 file changed, 14 insertions(+)
>>
>> diff --git a/controller/ovn-controller.c b/controller/ovn-controller.c
>> index 7e6c9e69a..51dc394f0 100644
>> --- a/controller/ovn-controller.c
>> +++ b/controller/ovn-controller.c
>> @@ -7895,6 +7895,20 @@ main(int argc, char *argv[])
>> ovsrec_open_vswitch_update_other_config_setkey(
>> cfg, "vlan-limit", "0");
>> }
>> + /* Clear flow-restore-wait. OVN at one point would set
>> + * flow-restore-wait in order to try to synchronize with
>> + * OVS. However, that resulted in a bug, so that behavior
>> + * was reverted. If upgrading from a version where OVN
>> + * manipulted flow-restore-wait, then flow-restore-wait
>> + * needs to be cleared in order for OVS to function
>> + * properly. This is (hopefully) a temporary measure until
>> + * a more reliable method of synchronizing with OVS is
>> + * devised.
>> + */
>> + if (smap_get(&cfg->other_config, "flow-restore-wait")) {
>> + ovsrec_open_vswitch_update_other_config_delkey(
>> + cfg, "flow-restore-wait");
>> + }
>> }
>>
>> static bool chassis_idx_stored = false;
>> --
>> 2.52.0
>>
>> _______________________________________________
>> dev mailing list
>> [email protected]
>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>>
>
> Regards,
> Ales
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
