On 6/9/26 10:38 AM, Xavier Simonart via dev wrote:
> Hello Aditya
> 
> Thanks for the patch and for v2.
> I only have a minor comment below that the maintainer can address when
> applying the fix.
> 
> Acked-by: Xavier Simonart <[email protected]>
> Thanks
> Xavier
> 
> On Fri, Jun 5, 2026 at 7:41 PM Aditya Mehakare <[email protected]>
> wrote:
> 
>> When a container (or virtual/mirror) child port is deleted,
>> handle_deleted_vif_lport() calls if_status_mgr_remove_ovn_installed()
>> on the associated lbinding's OVS interface.  For child port types,
>> lbinding points to the parent's local_binding, so this incorrectly
>> strips ovn-installed from the parent's OVS interface.
>>
>> This causes a cascade of problems:
>> 1. ovn-installed is removed from the parent's OVS interface.
>> 2. On the next full recompute (triggered by any SB commit failure,
>>    IDL reconnection, etc.), claim_lport() notices the missing
>>    ovn-installed and re-claims the parent port.
>> 3. The parent port re-enters OIF_CLAIMED -> OIF_INSTALL_FLOWS,
>>    during which local_binding_set_down() marks the parent AND all
>>    remaining children as down in the Southbound database.
>> 4. After flows are re-installed, all ports are set back up.
>>
>> This results in a disruptive and unnecessary down/up bounce of the
>> parent port and all sibling container ports whenever a single child
>> port is deleted.
>>
>> Fix this by skipping ovn-installed removal for port types that are
>> bound through a parent's local_binding rather than their own, so
>> that deleting a child port no longer disturbs the parent's interface
>> state.
>>
>> Fixes: ec1db7ae09fa ("ovn-controller: fixed ovn-installed not always
>> properly added or removed.")
>> Signed-off-by: Aditya Mehakare <[email protected]>
>> Acked-by: Naveen Yerramneni <[email protected]>
>> ---

Hi Aditya, Xavier,

Thanks for the patch and review!

>> +
>> +# Make sure that ovn-controller has not crashed.
>> +AT_CHECK([kill -0 $(cat hv1/ovn-controller.pid)])
>>
> I do not think this is necessary; OVN_CLEANUP now performs recompute, which
> catches an ovn-controller crash.
> 

I took care of this and applied it to main and 26.03.

Regards,
Dumitru

>> +
>> +OVN_CLEANUP([hv1])
>> +AT_CLEANUP
>> +])
>> --
>> 2.43.5
>>
>> _______________________________________________
>> dev mailing list
>> [email protected]
>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>>
>>
> _______________________________________________
> dev mailing list
> [email protected]
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> 

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to