On 6/24/21 5:44 PM, Dumitru Ceara wrote:
> On 6/22/21 7:25 AM, Han Zhou wrote:
>> On Sat, Jun 19, 2021 at 2:51 AM Mark Gray <mark.d.g...@redhat.com> wrote:
>>> If a new table is added to a logical flow pipeline, the mapping between
>>> 'external_ids:stage-name' from the 'Logical_Flow' table in the
>>> 'OVN_Southbound' database and the 'stage' number may change for some
>> tables.
>>> If 'ovn-northd' is started against a populated Southbound database,
>>> 'external_ids:stage-name' will not be updated to reflect the new, correct
>>> name. This will cause the stage name to be incorrectly displayed by some
>>> tools and commands such as `ovn-sbctl dump-flows`.
>>>
>>> This commit, reconciles changes to the stage name as part of
>> build_lflows().
>> This is interesting. It means a flow F existed in stage S (named "foo") of
>> ovn-northd version V1, and in version V2, S becomes S + 1 ("foo"), but in
>> the V2's stage S ("bar") there happens to be an exactly same flow like F
>> existing, but the stage name shown in ovn-sbctl dump-flows will be "foo"
>> instead of "bar". Is this the scenario the patch is trying to fix?
>>
>> If so, I think it is better not only fixing the "stage-name" but also other
>> external_ids, including "source" and "stage-hint", for the same reason.
>>
>
> Hi Mark, Han,
>
> Wouldn't this cause lots of additional string comparisons? What if we
> restrict ourselves to checking one flow per stage (for each ovn-northd
> iteration). Would that be enough to detect that we need to update all
> of them?
>
> This might be an ugly hack though. :)
Maybe we can check if 'northd_internal_version' changed in SB and re-check
all external-ids only in this case?
Best regards, Ilya Maximets.
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
